vern-llm 2.6.1 → 2.6.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +18 -36
- package/dist/index.cjs +2 -2
- package/dist/index.cjs.map +1 -1
- package/dist/index.d.cts +27 -54
- package/dist/index.d.cts.map +1 -1
- package/dist/index.d.mts +27 -54
- package/dist/index.d.mts.map +1 -1
- package/dist/index.mjs +2 -2
- package/dist/index.mjs.map +1 -1
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -24,11 +24,11 @@
|
|
|
24
24
|
<img src="https://img.shields.io/badge/TypeScript-strict-3178C6?logo=typescript&logoColor=white" alt="TypeScript" />
|
|
25
25
|
</p>
|
|
26
26
|
|
|
27
|
-
<p align="center">The LLM call framework. Resilience, observability, and control for every call.</p>
|
|
27
|
+
<p align="center">The LLM call framework. Resilience, observability, and control for every call, in your own process.</p>
|
|
28
28
|
|
|
29
|
-
|
|
29
|
+
The `vern-llm` npm package: one client across OpenAI-compatible, Anthropic, Gemini, and Bedrock, with retries, circuit breaking, fallback, rate limiting, caching, and middleware built in.
|
|
30
30
|
|
|
31
|
-
**Full documentation: [vernllm.dev](https://vernllm.dev)**
|
|
31
|
+
**Full documentation: [vernllm.dev](https://vernllm.dev)** for installation, structured output, caching, circuit breaker, provider fallback, rate limiting, observability, every adapter, and the complete API reference. This README is a quick reference, not the manual.
|
|
32
32
|
|
|
33
33
|
## Install
|
|
34
34
|
|
|
@@ -43,44 +43,22 @@ import Anthropic from '@anthropic-ai/sdk';
|
|
|
43
43
|
import OpenAI from 'openai';
|
|
44
44
|
import { fromAnthropic, fromOpenAI, VernLLM } from 'vern-llm';
|
|
45
45
|
|
|
46
|
+
const openai = fromOpenAI(new OpenAI({ apiKey: process.env.OPENAI_API_KEY }));
|
|
47
|
+
const anthropic = fromAnthropic(new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY }));
|
|
48
|
+
|
|
46
49
|
const llm = new VernLLM({
|
|
47
|
-
client:
|
|
50
|
+
client: openai,
|
|
48
51
|
model: 'gpt-4o',
|
|
52
|
+
fallback: { client: anthropic, model: 'claude-sonnet-5', circuitBreaker: true },
|
|
53
|
+
rateLimit: { requestsPerMinute: 500, tokensPerMinute: 100_000, maxConcurrent: 20 },
|
|
54
|
+
retryBudget: { windowMs: 60_000, minCalls: 20, retryRatio: 0.2 },
|
|
49
55
|
maxRetries: 3,
|
|
50
56
|
timeoutMs: 10_000,
|
|
51
|
-
|
|
52
|
-
|
|
53
|
-
fallback: {
|
|
54
|
-
client: fromAnthropic(new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY })),
|
|
55
|
-
model: 'claude-sonnet-5',
|
|
56
|
-
circuitBreaker: true,
|
|
57
|
-
},
|
|
58
|
-
onEvent: (event) => {
|
|
59
|
-
if (event.kind === 'fallback') console.warn(`falling over ${event.from} -> ${event.to}`);
|
|
60
|
-
},
|
|
61
|
-
});
|
|
62
|
-
|
|
63
|
-
const getWeather = {
|
|
64
|
-
name: 'get_weather',
|
|
65
|
-
description: 'Gets the current weather for a city',
|
|
66
|
-
parameters: { type: 'object', properties: { city: { type: 'string' } }, required: ['city'] },
|
|
67
|
-
};
|
|
68
|
-
|
|
69
|
-
const { chunks, finalResult } = await llm.cachedCall({
|
|
70
|
-
cacheKey: 'weather-demo-001',
|
|
71
|
-
ttl: 60,
|
|
72
|
-
call: {
|
|
73
|
-
userContent: "What's the weather in New York?",
|
|
74
|
-
tools: [getWeather],
|
|
75
|
-
stream: true,
|
|
76
|
-
},
|
|
57
|
+
defaultMaxTokens: 1000,
|
|
58
|
+
defaultReasoningEffort: 'medium',
|
|
77
59
|
});
|
|
78
60
|
|
|
79
|
-
|
|
80
|
-
if (chunk.type === 'text-delta') process.stdout.write(chunk.delta);
|
|
81
|
-
}
|
|
82
|
-
|
|
83
|
-
const result = await finalResult; // cached, retried, and streamed, tool calls included
|
|
61
|
+
const result = await llm.call({ userContent: "What's the weather in New York?" });
|
|
84
62
|
```
|
|
85
63
|
|
|
86
64
|
## Why vern-llm?
|
|
@@ -93,13 +71,17 @@ const result = await finalResult; // cached, retried, and streamed, tool calls i
|
|
|
93
71
|
- **Streaming**: set `stream: true` on any call and get live chunks alongside the same validated result the call would otherwise resolve to
|
|
94
72
|
- **Provider-native JSON Schema mode**: constrain generation itself, not just validate after the fact
|
|
95
73
|
- **Caching**: wrap any LLM call with `cachedCall`, bring your own cache adapter
|
|
96
|
-
- **Middleware
|
|
74
|
+
- **Middleware pipeline**: `transform` patches the outgoing request per attempt, `wrap` runs once around the whole logical call regardless of retries or fallback, patches from separate middleware merge instead of clobbering each other, order is controlled by `priority`, and each entry can be conditionally `enabled` per call
|
|
97
75
|
- **Circuit breaker**: trips after repeated failures, recovers automatically once the provider's back, independent per fallback target too
|
|
98
76
|
- **Observability**: one `onEvent` stream reports retries, fallovers, circuit transitions, and rate-limit waits
|
|
99
77
|
- **Usage tracking**: `onUsage` and `onUsageFailure` report token spend on success and on failure, so nothing goes unaccounted for when a call fails after the provider already responded
|
|
100
78
|
- **One interface, every provider**: OpenAI, Groq, Mistral, DeepSeek, Cerebras, Together, Fireworks, Ollama, Anthropic, Gemini, Bedrock, or raw HTTP via `fromFetch`
|
|
101
79
|
- **Zero runtime dependencies**: `zod` and provider SDKs are not required dependencies; vern-llm relies on compatible interfaces rather than specific implementations.
|
|
102
80
|
|
|
81
|
+
### Why not a gateway?
|
|
82
|
+
|
|
83
|
+
The customization is something a gateway can't match. vern-llm runs in your process, so you bring your own rate limiter, cache, and middlewares which transform the call instead of picking from a config panel, and hooks like `reserveUsage`, `onEvent` and even more give you superior control over billing and observability.
|
|
84
|
+
|
|
103
85
|
See the [docs](https://vernllm.dev) for adapter setup, caching, the circuit breaker, provider fallback, rate limiting, and structured output in depth.
|
|
104
86
|
|
|
105
87
|
## License
|
package/dist/index.cjs
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
Object.defineProperty(exports,Symbol.toStringTag,{value:`Module`});const e=new Set([`unknown_tool`,`duplicate_tool_call_id`,`tool_choice_none_violated`,`unexpected_tool_calls`]),t=new Set([`rate_limit_queue_full`,`rate_limit_queue_timeout`,`rate_limit_capacity_exceeded`,`retry_budget_exhausted`]),n=new Set([`middleware_timeout`]),r=new Set([`parse`,`validation`,`invalid_params`,`aborted`]);function i(i,a){return!(r.has(i)||a&&e.has(a)||a&&t.has(a)||a&&n.has(a))}const a=new Set([`quota_exceeded`]);function o(e,t){return!(!i(e,t)||a.has(e))}function s(e){if(e!==void 0)try{return JSON.stringify(e),e}catch{return`[Unserializable: issues contained a circular reference]`}}function c(e){if(e!==void 0)try{return JSON.parse(JSON.stringify(e))}catch{return`[Unserializable: request body contained a circular reference]`}}const l=new Set([`authorization`,`x-api-key`,`x-goog-api-key`,`api-key`]);function u(e){if(e===void 0)return;let t={};for(let[n,r]of Object.entries(e))l.has(n.toLowerCase())||(t[n]=r);return t}function d(e,t=0){if(e!==void 0)return t>=20?[]:e.map(e=>({...e,error:{...e.error,issues:s(e.error.issues),attempts:d(e.error.attempts,t+1)},request:e.request&&{...e.request,body:c(e.request.body),headers:u(e.request.headers)}}))}function f(e,t,n,r,i=Date.now()){return{provider:e,model:t,body:c(n),headers:u(r),startedAt:i}}var p=class extends Error{type;status;issues;cause;retryAfterMs;code;attempts;constructor(e,t,n={}){super(e),this.type=t,this.name=`LLMError`,this.status=n.status,this.issues=n.issues,this.cause=n.cause,this.retryAfterMs=n.retryAfterMs,this.code=n.code,this.attempts=n.attempts}get retryable(){return i(this.type,this.code)}get countsTowardBreaker(){return o(this.type,this.code)}toSnapshot(){return{message:this.message,type:this.type,status:this.status,issues:s(this.issues),retryAfterMs:this.retryAfterMs,code:this.code,retryable:this.retryable,attempts:d(this.attempts)}}toJSON(){return{name:this.name,message:this.message,type:this.type,status:this.status,issues:s(this.issues),retryAfterMs:this.retryAfterMs,code:this.code,retryable:this.retryable,attempts:d(this.attempts)}}};function m(e){return e instanceof p}function h(e,t){return e.code===t&&e.issues!==void 0}const g={onAccess(){},onInsert(){},getEvictee:e=>e.keys().next().value};function _(e,t){if(!e.has(t))return;let n=e.get(t);e.delete(t),e.set(t,n)}const v={onAccess:_,onInsert:_,getEvictee:e=>e.keys().next().value};function y(e){return e===`lru`?v:g}var b=class{maxSize;store=new Map;eviction;constructor(e=1e3,t=`fifo`){this.maxSize=e,this.eviction=y(t)}async get(e){let t=this.store.get(e);return t?Date.now()>=t.expiresAt?(this.store.delete(e),{hit:!1,value:null}):(this.eviction.onAccess(this.store,e),{hit:!0,value:t.value}):{hit:!1,value:null}}async set(e,t,n){this.cleanupExpiredEntries(),this.store.set(e,{value:t,expiresAt:Date.now()+n*1e3}),this.eviction.onInsert(this.store,e),this.enforceSizeLimit()}async delete(e){this.store.delete(e)}cleanupExpiredEntries(){let e=Date.now();for(let[t,n]of this.store)e>=n.expiresAt&&this.store.delete(t)}enforceSizeLimit(){for(;this.store.size>this.maxSize;){let e=this.eviction.getEvictee(this.store);if(e===void 0)break;this.store.delete(e)}}},x=class{inner;constructor(e=new b){this.inner=e}normalize(e){return e.toLowerCase().trim().replace(/[^\p{L}\p{N}\s]/gu,` `).replace(/\s+/g,` `).trim()}async resolveKey(e){return this.normalize(e)}async get(e){return this.inner.get(this.normalize(e))}async set(e,t,n){await this.inner.set(this.normalize(e),t,n)}async delete(e){await this.inner.delete?.(this.normalize(e))}},S=class{l1;l2;l1Ttl;constructor(e,t,n){this.l1=e,this.l2=t,this.l1Ttl=n}async resolveKey(e){return this.l1.resolveKey?this.l1.resolveKey(e):this.l2.resolveKey?this.l2.resolveKey(e):e}async get(e){let t=await this.l1.get(e);if(t.hit)return t;let n=await this.l2.get(e);return n.hit&&await this.l1.set(e,n.value,this.l1Ttl??60),n}async set(e,t,n){await Promise.all([this.l1.set(e,t,this.l1Ttl??n),this.l2.set(e,t,n)])}async delete(e){await Promise.all([this.l1.delete?.(e),this.l2.delete?.(e)])}};function C(e){return e}function ee(e){return typeof e==`object`&&!!e&&`type`in e&&e.type===`tool_calls`&&Array.isArray(e.toolCalls)}const te=new Set([`unknown_tool`,`duplicate_tool_call_id`,`tool_choice_none_violated`,`unexpected_tool_calls`]),ne=e=>e.type===`parse`||e.type===`validation`||e.type===`aborted`||e.type===`quota_exceeded`||e.code&&te.has(e.code)?`stop`:`next`;var w=class extends p{attempts;constructor(e){let t=e[e.length-1]?.error;super(`${e.length} provider${e.length===1?``:`s`} attempted and failed: ${e.map(e=>`${e.provider}(${e.error.type})`).join(` then `)}`,`fallback_exhausted`,{status:t?.status,cause:t,retryAfterMs:t?.retryAfterMs,code:`fallback_exhausted`,attempts:e}),this.attempts=e}get retryable(){let e=this.attempts[this.attempts.length-1]?.error;return e?e.retryable:super.retryable}};function re(e){return e instanceof w}function ie(){return{}}function ae(e){return{debugName:e}}function T(){let e=new Map;return{get(t){return e.get(t)},set(t,n){e.set(t,n)}}}function oe(e){let{onError:t,...n}=e;return t?{...n,wrap:async(e,n,r)=>{try{return await n()}catch(e){try{await t(e,r)}catch{}throw e}}}:n}function se(e,t){let n=e.completion_tokens_details?.reasoning_tokens;return{promptTokens:e.prompt_tokens??0,completionTokens:e.completion_tokens??0,totalTokens:e.total_tokens??0,...n===void 0?{}:{reasoningTokens:n},requestId:t.requestId,model:t.model,provider:t.providerName,usedFallback:t.isFallback}}async function ce(e,t,n){if(!e.reserveUsage)return!1;try{return await e.reserveUsage({coalesced:t,signal:n}),!0}catch(e){throw n?.aborted?new p(`LLM request aborted`,`aborted`):new p(e instanceof Error?e.message:`Usage reservation failed`,`quota_exceeded`,{cause:e})}}function le(e,t,n,r){return async i=>{try{await e.refundUsage?.({coalesced:t,signal:n})}catch(e){r(i,e)}}}async function E(e,t,n,r,i){if(r?.aborted)throw new p(`LLM request aborted`,`aborted`);let a=await ce(e,t,r),o=le(e,t,r,i);if(r?.aborted)throw a&&await o(`[VernLLM] refundUsage failed after abort`),new p(`LLM request aborted`,`aborted`);let s;try{s=await n()}catch(e){throw a&&await o(`[VernLLM] refundUsage failed`),e}if(r?.aborted)throw a&&await o(`[VernLLM] refundUsage failed after abort`),new p(`LLM request aborted`,`aborted`);return s}async function ue(e,t,n,r){if(n?.aborted)throw new p(`LLM request aborted`,`aborted`);let i=await ce(e,!1,n),a=le(e,!1,n,r);if(n?.aborted)throw i&&await a(`[VernLLM] refundUsage failed after abort`),new p(`LLM request aborted`,`aborted`);let o;try{o=await t()}catch(e){throw i&&await a(`[VernLLM] refundUsage failed after stream-open failure`),e}let s=o.finalResult.then(e=>e,async e=>{throw i&&await a(`[VernLLM] refundUsage failed after stream error`),e});return s.catch(()=>{}),{chunks:o.chunks,finalResult:s}}function de(){let e=new Map;function t(t){return e.get(t)}function n(t,n){return e.set(t,n),n.catch(()=>{}).finally(()=>{e.get(t)===n&&e.delete(t)}),n}return{get:t,track:n}}function fe(e){return typeof e==`string`?e:JSON.stringify(e)??``}function pe(e,t){let n=[];if(t){let t=e;t.type===`tool_calls`?(t.toolCalls.forEach((e,t)=>{n.push({type:`tool_call_delta`,index:t,id:e.id,name:e.name,argsDelta:JSON.stringify(e.arguments??{}),complete:!0})}),t.content&&n.push({type:`text-delta`,delta:t.content})):n.push({type:`text-delta`,delta:fe(t.content)})}else n.push({type:`text-delta`,delta:fe(e)});return{async*[Symbol.asyncIterator](){for(let e of n)yield e}}}function me(e,t){return{async*[Symbol.asyncIterator](){yield*pe(await e,t)}}}var he=class{cache;logger;inFlight=de();constructor(e,t){this.cache=e,this.logger=t}async resolveCacheKey(e){return this.cache.resolveKey?await this.cache.resolveKey(e):e}async deleteCache(e){if(this.cache.delete)try{await this.cache.delete(await this.resolveCacheKey(e))}catch(e){this.logger.warn(`[VernLLM] cache delete failed: ${e instanceof Error?e.message:`unknown`}`)}}logRefundError(e,t){this.logger.error(e,{message:t instanceof Error?t.message:`unknown`})}async getCached(e){try{return await this.cache.get(e)}catch(e){return this.logger.warn(`[VernLLM] cache read failed: ${e instanceof Error?e.message:`unknown`}`),{hit:!1}}}async resolveKeyAndReadCache(e){let t=await this.resolveCacheKey(e.cacheKey);return{resolvedParams:t===e.cacheKey?e:{...e,cacheKey:t},cached:await this.getCached(t)}}joinInFlight(e,t){return E(e,!0,()=>t,e.signal,(e,t)=>this.logRefundError(e,t))}async runCached(e){let{resolvedParams:t,cached:n}=await this.resolveKeyAndReadCache(e);if(n.hit)return n.value;let r=this.inFlight.get(t.cacheKey);return r?this.joinInFlight(t,r):this.registerTrigger(t)}registerTrigger(e){let t=E(e,!1,()=>this.runAndCache(e),e.signal,(e,t)=>this.logRefundError(e,t));return this.inFlight.track(e.cacheKey,t),t}async runAndCache(e){let t=await e.fn();try{await this.cache.set(e.cacheKey,t,e.ttl)}catch(e){this.logger.warn(`[VernLLM] cache write failed: ${e instanceof Error?e.message:`unknown`}`)}return t}async runCachedStream(e,t){let{resolvedParams:n,cached:r}=await this.resolveKeyAndReadCache(e);if(r.hit){let e=r.value;return{chunks:pe(e,t),finalResult:Promise.resolve(e)}}let i=this.inFlight.get(n.cacheKey);if(i){let e=this.joinInFlight(n,i);return e.catch(()=>{}),{chunks:me(e,t),finalResult:e}}return this.registerStreamTrigger(n)}registerStreamTrigger(e){let t,n,r=new Promise((e,r)=>{t=e,n=r});this.inFlight.track(e.cacheKey,r);let i=ue(e,async()=>{let t=await e.openStream(),n=t.finalResult.then(async t=>{try{await this.cache.set(e.cacheKey,t,e.ttl)}catch(e){this.logger.warn(`[VernLLM] cache write failed: ${e instanceof Error?e.message:`unknown`}`)}return t},e=>{throw e});return{chunks:t.chunks,finalResult:n}},e.signal,(e,t)=>this.logRefundError(e,t));return i.then(e=>{e.finalResult.then(t,n)},e=>{n(e)}),i}};const ge=1e4;function _e(e){return!e||e<=0||e===1/0?void 0:e}function ve(e){return Math.min(e,2147483647)}async function ye(e,t,n){let r=new AbortController,i=_e(t),a=i===void 0?void 0:setTimeout(()=>{r.abort()},ve(i)),o=n?AbortSignal.any([n,r.signal]):r.signal;try{return await e(o)}catch(e){throw r.signal.aborted&&!n?.aborted&&e instanceof DOMException&&e.name===`AbortError`?new p(`Request timed out`,`timeout`,{code:`request_timeout`}):e}finally{clearTimeout(a)}}function be(e,t,n,r){let i=_e(t);if(i===void 0)return e();let a=!1;return new Promise((t,o)=>{let s=setTimeout(()=>{a=!0,n?.(),o(new p(`No stream chunk received for ${i}ms (idle timeout)`,`timeout`,{code:`idle_timeout`}))},ve(i));e().then(e=>{if(clearTimeout(s),a){r?.debug(`[VernLLM] chunk resolved after idle timeout already fired; discarding`);return}a=!0,t(e)},e=>{if(clearTimeout(s),a){r?.debug(`[VernLLM] chunk rejection arrived after idle timeout already fired; discarding`);return}a=!0,o(e)})})}const xe=[{name:`Retry-After-Ms`,kind:`ms`},{name:`X-Retry-After-Ms`,kind:`ms`},{name:`Retry-After`,kind:`seconds`},{name:`Retry-After`,kind:`date`}];function Se(e,t){let n=e;if(typeof n.get==`function`)return n.get(t)??void 0;let r=Object.entries(e).find(([e])=>e.toLowerCase()===t.toLowerCase())?.[1];return typeof r==`string`?r:void 0}function D(e,t){return e<0?void 0:Math.min(e,t)}function Ce(e,t=ge){if(!e||typeof e!=`object`)return;let n=e,r=n.headers??n.response?.headers;if(!(!r||typeof r!=`object`))for(let{name:e,kind:n}of xe){let i=Se(r,e)?.trim();if(i){if(n===`ms`&&/^\d+$/.test(i))return D(Number(i),t);if(n===`seconds`&&/^\d+(\.\d+)?$/.test(i))return D(Number(i)*1e3,t);if(n===`date`){let e=Date.parse(i);if(!Number.isNaN(e))return D(e-Date.now(),t)}}}}function we(e){return Math.random()*e}function Te(e,t,n=ge,r=!1,i=!1){return we(Math.min(e*(r?2:i?1.5:1)*2**t,n))}async function Ee(e,t){if(t?.aborted)throw new p(`Operation aborted`,`aborted`);await new Promise((n,r)=>{let i=()=>{clearTimeout(a),r(new p(`Operation aborted`,`aborted`))},a=setTimeout(()=>{t?.removeEventListener(`abort`,i),n()},e);t?.addEventListener(`abort`,i,{once:!0})})}function De(e,t,n,r){if(r?.aborted||e instanceof p&&!e.retryable)return!1;let i=n(e);return!(i!==void 0&&t.includes(i))}async function Oe(e){let{requestId:t,model:n,attempt:r,error:i,state:a,signal:o,providerName:s,maxRetries:c,baseDelayMs:l,middleware:u,middlewareTimeoutMs:d,logger:f,reportEvent:p,buildEventContext:m,extractStatus:h,normalizeError:g,emitEvent:_}=e,v=Ce(i),y=h(i),b=v??Te(l,r,void 0,y===429,y!==void 0&&y>=500&&y<=599),x=v!==void 0;f.warn(`[VernLLM:${t}] recovery attempt ${r}/${c}, waiting ${Math.ceil(b)}ms`+(x?` (honoring Retry-After)`:``)),_({kind:`retry`,requestId:t,provider:s,model:n,attempt:r,maxRetries:c,delayMs:b,retryAfterHonored:x,error:g(i,o)},m(t,n,r,o,a),p,u,d,f),await Ee(b,o)}async function ke(e){let{fn:t,maxRetries:n,signal:r,onAttempt:i,attempts:a,shouldRetryAttempt:o,recoverDelayForAttempt:s,normalizeError:c}=e,l,u;for(let e=0;e<=n;e++){u=void 0;try{return e>0&&await s(e,l),i?.(),await t(e,e=>{u=e})}catch(t){if(l=t,!(e<n&&o(t,r)))break;a?.push({index:e,error:c(t,r).toSnapshot(),request:u})}}throw l}var O=class{buckets;bucketStartMs=Date.now();bucketWidthMs;constructor(e){if(!Number.isFinite(e)||e<=0)throw RangeError(`RollingRatio: windowMs must be a finite number > 0, got ${e}`);this.bucketWidthMs=e/10,this.buckets=Array.from({length:10},()=>({total:0,failures:0}))}rotate(){let e=Date.now()-this.bucketStartMs,t=Math.floor(e/this.bucketWidthMs),n=Math.min(10,t);for(let e=0;e<n;e++)this.buckets.shift(),this.buckets.push({total:0,failures:0});t>0&&(this.bucketStartMs+=t*this.bucketWidthMs)}record(e){this.rotate();let t=this.buckets[9];t.total+=1,e&&(t.failures+=1)}getRatio(){this.rotate();let e=this.buckets.reduce((e,t)=>e+t.total,0),t=this.buckets.reduce((e,t)=>e+t.failures,0);return e===0?0:t/e}getCount(){return this.rotate(),this.buckets.reduce((e,t)=>e+t.total,0)}reset(){this.buckets=Array.from({length:10},()=>({total:0,failures:0})),this.bucketStartMs=Date.now()}};function Ae(e,t=`minCalls`){if(!Number.isInteger(e)||e<0)throw RangeError(`${t} must be a non-negative integer, got ${e}`)}function je(e,t){if(!Number.isFinite(e)||e<0||e>1)throw RangeError(`${t} must be a finite number between 0 and 1, got ${e}`)}function Me(e){if(e===void 0)return;if(typeof e==`function`)return e;let{multiplier:t,maxMs:n=1/0}=e;return(e,r)=>we(Math.min(r*t**e,n))}var Ne=class{threshold;failuresByKey=new Map;constructor(e){this.threshold=e}onSuccess(e){this.failuresByKey.set(e,0)}onFailure(e){let t=(this.failuresByKey.get(e)??0)+1;return this.failuresByKey.set(e,t),t>=this.threshold}reset(e){this.failuresByKey.set(e,0)}forget(e){this.failuresByKey.delete(e)}},Pe=class{windowMs;minCalls;failureRatio;ratiosByKey=new Map;constructor(e,t,n){this.windowMs=e,this.minCalls=t,this.failureRatio=n,new O(e),Ae(t),je(n,`failureRatio`)}ratioFor(e){let t=this.ratiosByKey.get(e);return t||(t=new O(this.windowMs),this.ratiosByKey.set(e,t)),t}onSuccess(e){this.ratioFor(e).record(!1)}onFailure(e){let t=this.ratioFor(e);return t.record(!0),t.getCount()>=this.minCalls&&t.getRatio()>=this.failureRatio}reset(e){this.ratiosByKey.delete(e)}forget(e){this.ratiosByKey.delete(e)}};function Fe(e){return`onFailure`in e?e:e.kind===`consecutive`?new Ne(e.threshold):new Pe(e.windowMs,e.minCalls,e.failureRatio)}function Ie(){return{state:`closed`,consecutiveFailures:0,openedAt:0,trial:null,reopenCount:0,failuresByReason:new Map,cooldownMsForOpen:0}}function k(e){return e??``}const A=new WeakMap;function Le(e,t){return!t||A.get(t.state)===e.trial}function Re(e,t){let n=t??`unknown`;e.failuresByReason.set(n,(e.failuresByReason.get(n)??0)+1)}var ze=class{cooldownMs;onStateChange;isolateByModel;halfOpenProbes;halfOpenSuccessRatio;cooldownBackoff;tripping;sharedBucket=Ie();bucketsByModel=new Map;constructor(e={}){let t=e.threshold??5;this.cooldownMs=e.cooldownMs??3e4,this.onStateChange=e.onStateChange,this.isolateByModel=e.isolateByModel??!1;let n=e.halfOpenProbes;this.halfOpenProbes=Number.isFinite(n)?Math.max(1,Math.floor(n)):1;let r=e.halfOpenSuccessRatio;this.halfOpenSuccessRatio=Number.isFinite(r)?Math.min(1,Math.max(0,r)):1,this.cooldownBackoff=Me(e.cooldownBackoff),this.tripping=Fe(e.tripping??{kind:`consecutive`,threshold:t})}assertClosed(e,t){let n=this.ensureBucketFor(e);if(n.state!==`closed`){if(n.state===`open`){let r=Date.now()-n.openedAt,i=n.cooldownMsForOpen;if(r<i)throw new p(`Circuit open, provider has failed ${n.consecutiveFailures} times in a row. Retry in ${Math.ceil((i-r)/1e3)}s.`,`circuit_open`,{code:`circuit_cooling_down`});n.trial={slotsRemaining:this.halfOpenProbes-1,successes:0,failures:0},t&&A.set(t.state,n.trial),this.transition(n,`half-open`,e,t);return}if(!n.trial||n.trial.slotsRemaining<=0)throw new p(`Circuit half-open. Every trial slot is already in flight. Try again shortly.`,`circuit_open`,{code:`circuit_trial_in_flight`});--n.trial.slotsRemaining,t&&A.set(t.state,n.trial)}}recordSuccess(e,t){let n=this.lookupBucket(e);if(n){if(n.state===`half-open`&&n.trial&&Le(n,t)){n.trial.successes+=1,this.settleTrialIfComplete(n,e,t);return}n.state!==`half-open`&&(n.consecutiveFailures=0,this.tripping.onSuccess(this.trippingKeyFor(e)),n.trial=null,n.reopenCount=0,n.failuresByReason.clear(),this.transition(n,`closed`,e,t),this.isolateByModel&&n.state===`closed`&&n.consecutiveFailures===0&&this.forgetModel(e))}}recordFailure(e,t,n){let r=this.ensureBucketFor(e);if(r.state===`half-open`&&r.trial&&Le(r,t)){r.trial.failures+=1,Re(r,n),this.settleTrialIfComplete(r,e,t);return}r.state!==`half-open`&&(r.consecutiveFailures+=1,Re(r,n),this.tripping.onFailure(this.trippingKeyFor(e))&&this.openBucket(r,e,t))}getState(e){return this.lookupBucket(e)?.state??`closed`}getFailureBreakdown(e){let t=this.lookupBucket(e);return t?Object.fromEntries(t.failuresByReason):{}}open(e,t){let n=this.ensureBucketFor(e);n.trial=null,this.openBucket(n,e,t)}close(e,t){let n=this.ensureBucketFor(e);n.consecutiveFailures=0,this.tripping.reset(this.trippingKeyFor(e)),n.trial=null,n.reopenCount=0,n.failuresByReason.clear(),this.transition(n,`closed`,e,t),this.isolateByModel&&n.state===`closed`&&n.consecutiveFailures===0&&this.forgetModel(e)}openBucket(e,t,n){e.openedAt=Date.now(),e.cooldownMsForOpen=this.computeCooldown(e),this.transition(e,`open`,t,n)}computeCooldown(e){if(!this.cooldownBackoff)return this.cooldownMs;let t=this.cooldownBackoff(e.reopenCount,this.cooldownMs);return Number.isNaN(t)?0:Math.max(0,t)}lookupBucket(e){return this.isolateByModel?this.bucketsByModel.get(k(e)):this.sharedBucket}trippingKeyFor(e){return this.isolateByModel?k(e):``}ensureBucketFor(e){if(!this.isolateByModel)return this.sharedBucket;let t=k(e),n=this.bucketsByModel.get(t);return n||(n=Ie(),this.bucketsByModel.set(t,n)),n}forgetModel(e){this.bucketsByModel.delete(k(e)),this.tripping.forget?.(this.trippingKeyFor(e))}transition(e,t,n,r){if(t===e.state)return;let i=e.state;e.state=t,this.onStateChange?.(i,t,e.consecutiveFailures,n,r)}settleTrialIfComplete(e,t,n){let r=e.trial;if(!r||r.successes+r.failures<this.halfOpenProbes)return;let i=r.successes/this.halfOpenProbes;if(e.trial=null,i>=this.halfOpenSuccessRatio){e.consecutiveFailures=0,this.tripping.onSuccess(this.trippingKeyFor(t)),e.reopenCount=0,e.failuresByReason.clear(),this.transition(e,`closed`,t,n),this.isolateByModel&&e.state===`closed`&&this.forgetModel(t);return}e.reopenCount+=1,this.openBucket(e,t,n)}},Be=class{options;ratio;constructor(e){this.options=e,this.ratio=new O(e.windowMs),Ae(e.minCalls),je(e.retryRatio,`retryRatio`)}assertAvailable(){if(this.ratio.getCount()>=this.options.minCalls&&this.ratio.getRatio()>=this.options.retryRatio)throw new p(`Retry budget exhausted`,`rate_limited`,{code:`retry_budget_exhausted`})}recordAttempt(e){this.ratio.record(e)}getSnapshot(){return{attempts:this.ratio.getCount(),retryRatio:this.ratio.getRatio()}}};function j(e){if(!e||typeof e!=`object`)return;let t=e;if(typeof t.status==`number`)return t.status;if(typeof t.statusCode==`number`)return t.statusCode;if(typeof t.$metadata?.httpStatusCode==`number`)return t.$metadata.httpStatusCode}const Ve=new Set([`ECONNREFUSED`,`ENOTFOUND`,`ECONNRESET`,`ETIMEDOUT`,`EAI_AGAIN`,`EPIPE`,`ECONNABORTED`,`EHOSTUNREACH`,`ENETUNREACH`]),He=new Set([`fetch failed`,`failed to fetch`,`load failed`,`networkerror when attempting to fetch resource.`]);function Ue(e){if(!e||typeof e!=`object`)return!1;let t=e;if(typeof t.code==`string`&&Ve.has(t.code)||typeof t.message==`string`&&He.has(t.message.toLowerCase()))return!0;if(t.cause&&typeof t.cause==`object`){let e=t.cause;if(typeof e.code==`string`&&Ve.has(e.code))return!0}return!1}function We(e){try{return JSON.stringify(e,null,2)??String(e)}catch{try{return String(e)}catch{return`[unprintable error]`}}}function Ge(e){if(e&&typeof e==`object`)try{let t=e;if(t.error!==void 0)return We(t.error);if(typeof t.message==`string`)return t.message}catch{}return We(e)}function Ke(e){switch(e){case 429:return`provider_rate_limited`;case 401:return`authentication`;case 403:return`authorization`;case 404:return`not_found`;case 413:return`payload_too_large`;default:return e>=500?`server_error`:void 0}}const qe=/\(no body\)/i;function Je(e){return Object.keys(e).length===0}function Ye(e){if(e&&typeof e==`object`){let{error:t,message:n}=e;if(t!=null){let e=typeof t==`string`&&t.trim().length===0,n=typeof t==`object`&&Je(t);if(!e&&!n)return!1}if(typeof n==`string`){let e=n.trim();return e.length===0||qe.test(e)}return!0}return!0}function M(e,t,n){if(t?.aborted)return new p(`LLM request aborted`,`aborted`,{attempts:n});if(e instanceof p)return e.code===void 0&&e.status!==void 0&&(e.code=Ke(e.status)),e.attempts===void 0&&n!==void 0&&(e.attempts=n),e;let r=j(e),i=Ce(e);if(r!==void 0){let t=Ge(e),a=Ke(r),o=a===void 0;return new p(Ye(e)?o?`LLM request failed with status ${r} and no error detail from the provider. This usually means a field or value in the request isn't supported by the specific model (for example, a reasoning/thinking parameter the model doesn't accept), rather than a transport or auth problem.`:`LLM request failed with status ${r} and no error detail from the provider.`:`LLM request failed: ${t}`,`api`,{status:r,cause:e,retryAfterMs:i,code:a,attempts:n})}return Ue(e)?new p(`LLM request failed`,`network`,{cause:e,retryAfterMs:i,code:`connection_failed`,attempts:n}):new p(`LLM request failed`,`unknown`,{cause:e,retryAfterMs:i,attempts:n})}function N(e,t){return e.name??`[${t}]`}function Xe(e,t,n){return t<=0?e():new Promise((r,i)=>{let a=setTimeout(()=>{i(new p(`middleware "${n}" timed out after ${t}ms`,`timeout`,{code:`middleware_timeout`}))},t);e().then(e=>{clearTimeout(a),r(e)},e=>{clearTimeout(a),i(e)})})}async function P(e,t,n,r,i){let{enabled:a}=e;if(a===void 0)return!0;if(typeof a==`boolean`)return a;let o=e.timeoutMs??r;try{return await Xe(async()=>a(t),o,n)}catch(e){return i.error(`[VernLLM] middleware "${n}".enabled threw or timed out, treating as disabled`,{message:e instanceof Error?e.message:`unknown`}),!1}}const Ze=[`temperature`,`max_tokens`,`reasoning_effort`,`budget_tokens`,`tool_choice`,`messages`,`addMessages`,`tools`,`addTools`];function Qe(e,t){let n=t,r=Ze.filter(e=>t[e]!==void 0);if(n.model!==void 0&&r.push(`model`),n.response_format!==void 0&&r.push(`response_format`),r.length===0)return{request:e,patchedFields:[]};let i={...e};return n.model!==void 0&&(i.model=n.model),n.response_format!==void 0&&(i.response_format=n.response_format),t.temperature!==void 0&&(i.temperature=t.temperature),t.max_tokens!==void 0&&(i.max_tokens=t.max_tokens),t.reasoning_effort!==void 0&&(i.reasoning_effort=t.reasoning_effort),t.budget_tokens!==void 0&&(i.budget_tokens=t.budget_tokens),t.tool_choice!==void 0&&(i.tool_choice=t.tool_choice),t.messages!==void 0&&(i.messages=t.messages),t.addMessages!==void 0&&t.addMessages.length>0&&(i.messages=[...i.messages,...t.addMessages]),t.tools!==void 0&&(i.tools=t.tools),t.addTools!==void 0&&t.addTools.length>0&&(i.tools=[...i.tools??[],...t.addTools]),{request:i,patchedFields:r}}function $e(e,t){if(!e.tools)return;let n=new Set,r=new Set;for(let t of e.tools)n.has(t.function.name)&&r.add(t.function.name),n.add(t.function.name);if(r.size>0)throw new p(`middleware "${t}" added tool(s) with duplicate name(s): [${[...r].join(`, `)}]. Tool names must be unique.`,`invalid_params`,{code:`duplicate_tool_names`,issues:{names:[...r]}})}function et(e,t,n){if(t.model!==e.model)throw new p(`middleware "${n}" changed \`model\` via transform, which isn't supported. Configure the target's model instead.`,`invalid_params`);if(JSON.stringify(t.response_format)!==JSON.stringify(e.response_format))throw new p(`middleware "${n}" changed \`response_format\` via transform, which isn't supported.`,`invalid_params`)}async function tt(e){let{request:t,requestId:n,attempt:r,signal:i,state:a,middleware:o,middlewareTimeoutMs:s,logger:c,reportEvent:l,buildContext:u}=e;if(o.length===0)return t;let d=t,f=t,p=o.map((e,t)=>({middlewareEntry:e,index:t})).sort((e,t)=>(e.middlewareEntry.priority??0)-(t.middlewareEntry.priority??0));for(let{middlewareEntry:e,index:t}of p){let d=N(e,t),p=u(r,i,a);if(!await P(e,p,d,s,c)){e.enabled!==void 0&&F({kind:`middleware`,requestId:n,middleware:d,hook:`enabled_skip`},p,l,o,s,c);continue}if(!e.transform)continue;let m=await nt(e,structuredClone(f),p,d,s),{request:h,patchedFields:g}=Qe(f,m);g.length>0&&((m.tools!==void 0||m.addTools?.length)&&$e(h,d),F({kind:`middleware`,requestId:n,middleware:d,hook:`transform`,patchedFields:g},p,l,o,s,c)),f=h}return et(d,f,`chain`),f}async function nt(e,t,n,r,i){if(!e.transform)return{};let a=e.timeoutMs??i;try{return await Xe(async()=>e.transform(t,n),a,r)}catch(e){throw rt(e,r,n.signal)}}function rt(e,t,n){let r=M(e,n);return r.type===`unknown`?new p(`middleware "${t}" threw: ${r.message}`,`invalid_params`,{code:`middleware_threw`,cause:e}):r}function F(e,t,n,r,i,a){n(e),r.length!==0&&it(r,e,t,i,a)}async function it(e,t,n,r,i){for(let a=0;a<e.length;a++){let o=e[a];if(!o.onEvent)continue;let s=N(o,a);if(await P(o,n,s,r,i))try{Promise.resolve(o.onEvent(t,{...n,own:{}})).catch(e=>{i.error(`[VernLLM] middleware "${s}".onEvent failed`,{message:e instanceof Error?e.message:`unknown`})})}catch(e){i.error(`[VernLLM] middleware "${s}".onEvent failed`,{message:e instanceof Error?e.message:`unknown`})}}}function I(e,t){return n=>{if(e)try{e(n)}catch(e){t.error(`[VernLLM] onEvent failed`,{message:e instanceof Error?e.message:`unknown`})}}}function L(e,t,n){let r=e[t];if(!r)throw RangeError(`${n}: no target at index ${t} (chain has ${e.length} target${e.length===1?``:`s`})`);return r}function R(e,t,n,r){t!==void 0&&!e&&r.warn(`[VernLLM] ${n}: \`model: '${t}'\` has no effect here. This target's circuitBreaker doesn't have isolateByModel on, so it only tracks one shared circuit regardless of \`model\`. Omit \`model\`, or set \`circuitBreaker.isolateByModel: true\` on this target if per-model tracking is what you want.`)}function at(e,t,n,r,i,a,o,s,c){if(!e)return;let l=typeof e==`object`?e:void 0,u=l?.onStateChange,d=I(r,i);return new ze({...l,onStateChange:(e,r,l,f,p)=>{let m={kind:`circuit_state`,provider:t,model:f??n,from:e,to:r,consecutiveFailures:l};if(p?F(m,{stage:`attempt`,requestId:p.requestId,requestedProvider:t,requestedModel:f??n,isFallbackAttempt:s,attempt:p.attempt??1,capabilities:{supportsJsonObjectMode:c},signal:p.signal,state:p.state,own:{}},d,a,o,i):d(m),u)try{u(e,r,l,f,p)}catch(e){i.error(`[VernLLM] circuitBreaker.onStateChange failed`,{message:e instanceof Error?e.message:`unknown`})}}})}function ot(e){let t=/^(?:(\d+(?:\.\d+)?)h)?(?:(\d+(?:\.\d+)?)m)?(?:(\d+(?:\.\d+)?)s)?(?:(\d+)ms)?$/.exec(e.trim());if(!t||t[0]===``)return;let[,n,r,i,a]=t,o=(n?parseFloat(n)*36e5:0)+(r?parseFloat(r)*6e4:0)+(i?parseFloat(i)*1e3:0)+(a?parseInt(a,10):0);return Number.isFinite(o)?o:void 0}function z(e,t){let n=e.get(t);if(n===null)return;let r=Number(n);return Number.isFinite(r)?r:void 0}function B(e){let t=e.get(`x-ratelimit-reset-requests`);return{limitRequests:z(e,`x-ratelimit-limit-requests`),remainingRequests:z(e,`x-ratelimit-remaining-requests`),resetAfterMs:t===null?void 0:ot(t)}}function st(e){let t=e.get(`anthropic-ratelimit-requests-reset`),n=t===null?NaN:Date.parse(t);return{limitRequests:z(e,`anthropic-ratelimit-requests-limit`),remainingRequests:z(e,`anthropic-ratelimit-requests-remaining`),resetAfterMs:Number.isFinite(n)?Math.max(0,n-Date.now()):void 0}}const ct=Symbol(`vernLLMProviderRateLimitHint`);function lt(e,t){t&&Object.defineProperty(e,ct,{value:t,enumerable:!1,configurable:!0})}function ut(e){if(!(!e||typeof e!=`object`))return e[ct]}function dt(e){let{breaker:t,requestId:n,model:r,providerName:i,isFallback:a,supportsJsonObjectMode:o}=e;function s(e,t,s){return{stage:`attempt`,requestId:n,requestedProvider:i,requestedModel:r,isFallbackAttempt:a,attempt:e+1,capabilities:{supportsJsonObjectMode:o},signal:t,state:s,own:{}}}function c(e,t,r){return{requestId:n,state:r,signal:t,attempt:e+1}}return{buildAttemptContext:s,buildCallContext:c,recordSuccess(e,n,i){t?.recordSuccess(r,c(e,n,i))},recordFailure(e,n,i,a){t?.recordFailure(r,c(e,n,i),a)}}}function ft(e){return e.map(e=>({type:`function`,function:{name:e.name,description:e.description,parameters:e.parameters}}))}function pt(e){return e.map(e=>({id:e.id,type:`function`,function:{name:e.name,arguments:JSON.stringify(e.arguments??{})}}))}function mt(e){return e.map(e=>{let t;try{t=e.function.arguments.trim()?JSON.parse(e.function.arguments):{}}catch{throw new p(`Invalid JSON arguments for tool call "${e.function.name}"`,`parse`,{code:`tool_arguments_parse_failed`})}return{id:e.id,name:e.function.name,arguments:t}})}function ht(e){return typeof e==`string`?e:JSON.stringify(e)}function gt(e,t){if(e&&e.length===0)throw new p("`tools` was an empty array. This is almost always a bug (e.g. a filtered tool list that ended up empty). An empty `tools` array still switches on tool-call mode (response shape, jsonMode default, wire format) with nothing for the model to call. Omit `tools` entirely for a normal call, or make sure the array is non-empty.",`invalid_params`);if(e){let t=new Set,n=new Set;for(let r of e)t.has(r.name)&&n.add(r.name),t.add(r.name);if(n.size)throw new p(`\`tools\` has duplicate name(s): [${[...n].join(`, `)}]. Tool names must be unique.`,`invalid_params`,{code:`duplicate_tool_names`,issues:{names:[...n]}})}if(t&&!e)throw new p("`toolChoice` was set without `tools`. There is nothing for it to choose between. Set `tools`, or remove `toolChoice`.",`invalid_params`);if(e&&typeof t==`object`&&!e.some(e=>e.name===t.name))throw new p(`toolChoice names "${t.name}", which is not in \`tools\` ([${e.map(e=>e.name).join(`, `)}]).`,`invalid_params`,{code:`unknown_tool_choice`,issues:{requested:t.name,available:e.map(e=>e.name)}})}function _t(e){let{jsonModeExplicit:t,hasTools:n,jsonSchema:r,hasSchema:i,supportsJsonObjectMode:a}=e,o=t??!n;if(!a&&!r&&t===!0)throw new p('jsonMode: true was set explicitly, but this client does not support `response_format: "json_object"` (see LLMClient.supportsJsonObjectMode). Neither Anthropic nor Bedrock has a field that mechanically guarantees JSON output for this mode. Use `jsonSchema` instead, which maps to a real constraint on both.',`invalid_params`);if(!a&&!r&&t===void 0&&i)throw new p('`schema` was provided, which requires JSON output to validate against, but this client does not support `response_format: "json_object"` (see LLMClient.supportsJsonObjectMode) and no `jsonSchema` was set. Neither Anthropic nor Bedrock has a field that mechanically guarantees JSON output without one. Use `jsonSchema` instead, which maps to a real constraint on both and still runs `schema` against its parsed result.',`invalid_params`);let s=(!a&&!r&&t===void 0?!1:o)||!!r;if(i&&!s)throw new p(`schema was provided but jsonMode: false disables JSON parsing, so nothing would validate it. Remove jsonMode: false, set jsonSchema, or remove schema.`,`invalid_params`);return s}function vt(e,t){return e?{type:`json_schema`,json_schema:{name:e.name,schema:e.schema,strict:e.strict??!0,description:e.description}}:t?{type:`json_object`}:void 0}function yt(e){return!e||e===`auto`?`auto`:e===`none`||e===`required`?e:{type:`function`,function:{name:e.name}}}function bt(e){return e.role===`tool`?(e.toolResults??[]).map(e=>({role:`tool`,tool_call_id:e.toolCallId,content:typeof e.content==`string`?e.content:JSON.stringify(e.content??null),...e.isError?{is_error:!0}:{}})):e.role===`assistant`&&e.toolCalls?.length?[{role:`assistant`,...e.content===void 0?{}:{content:ht(e.content)},tool_calls:pt(e.toolCalls)}]:e.role===`assistant`?[{role:`assistant`,content:ht(e.content===void 0?``:e.content)}]:[{role:e.role,content:e.content??``}]}function xt(e){let t;for(let[n,r]of e.entries()){if(r.role===`tool`){if(t?.role!==`assistant`||!t.toolCalls?.length)throw new p(`history[${n}] is a "tool" turn, but must immediately follow an "assistant" turn that requested tools`,`invalid_params`);if(!r.toolResults?.length)throw new p(`history[${n}] is a "tool" turn but has no toolResults`,`invalid_params`);let e=new Set(t.toolCalls.map(e=>e.id)),i=r.toolResults.map(e=>e.toolCallId),a=i.filter(t=>!e.has(t));if(a.length)throw new p(`history[${n}].toolResults references unknown toolCallId(s) [${a.join(`, `)}]`,`invalid_params`,{code:`unknown_tool_result_ids`,issues:{historyIndex:n,ids:a}});let o=new Set,s=new Set;for(let e of i)o.has(e)&&s.add(e),o.add(e);if(s.size)throw new p(`history[${n}].toolResults has duplicate toolCallId(s) [${[...s].join(`, `)}]`,`invalid_params`,{code:`duplicate_tool_result_ids`,issues:{historyIndex:n,ids:[...s]}});let c=[...e].filter(e=>!i.includes(e));if(c.length)throw new p(`history[${n}] is missing toolResults for toolCallId(s) [${c.join(`, `)}]`,`invalid_params`,{code:`missing_tool_results`,issues:{historyIndex:n,ids:c}})}else{if(r.role===t?.role)throw new p(`history must alternate user/assistant turns: consecutive "${r.role}" turns at history[${n-1}] and history[${n}]`,`invalid_params`);if(t?.role===`assistant`&&t.toolCalls?.length)throw new p(`history[${n}] follows an assistant tool request without tool results`,`invalid_params`)}t=r}if(t?.role===`assistant`&&t.toolCalls?.length)throw new p(`The last entry in history is an assistant tool request without tool results`,`invalid_params`);if(t?.role===`user`)throw new p(`The last entry in history is a "user" turn, which would collide with the current userContent turn.`,`invalid_params`)}var St=class{model;defaultMaxTokens;defaultTemperature;defaultReasoningEffort;defaultBudgetTokens;supportsJsonObjectMode;constructor(e){this.model=e.model,this.defaultMaxTokens=e.defaultMaxTokens,this.defaultTemperature=e.defaultTemperature,this.defaultReasoningEffort=e.defaultReasoningEffort,this.defaultBudgetTokens=e.defaultBudgetTokens,this.supportsJsonObjectMode=e.supportsJsonObjectMode}build(e){let{systemPrompt:t,userContent:n,history:r=[],maxTokens:i=this.defaultMaxTokens,model:a=this.model,jsonSchema:o,tools:s,toolChoice:c}=e,l=e.temperature===void 0?this.defaultTemperature:e.temperature,u=e.reasoningEffort===void 0?this.defaultReasoningEffort:e.reasoningEffort,d=e.budgetTokens===void 0?this.defaultBudgetTokens:e.budgetTokens;gt(s,c);let f=_t({jsonModeExplicit:e.jsonMode,hasTools:!!s,jsonSchema:o,hasSchema:!!e.schema,supportsJsonObjectMode:this.supportsJsonObjectMode}),p=vt(o,f);return xt(r),{useJson:f,model:a,request:{model:a,...l===null?{}:{temperature:l},max_tokens:i,...p?{response_format:p}:{},...u?{reasoning_effort:u}:{},...d==null?{}:{budget_tokens:d},...s?{tools:ft(s)}:{},...s?{tool_choice:yt(c)}:{},messages:[...t?[{role:`system`,content:t}]:[],...r.flatMap(e=>bt(e)),{role:`user`,content:n}]}}}};function Ct(e,t,n){let r;try{r=n(e)}catch{throw new p(`Invalid JSON response`,`parse`)}if(r==null)throw new p(`Invalid JSON response`,`parse`);if(!t)return r;let i=t.safeParse(r);if(!i.success)throw new p(`Schema validation failed`,`validation`,{issues:i.error});return i.data}function wt(e,t){let n=new Map(t.map(e=>[e.name,e])),r=new Set,i=[];for(let t of e)r.has(t.id)&&i.push({name:t.name,toolCallId:t.id,code:`duplicate_tool_call_id`}),r.add(t.id),n.has(t.name)||i.push({name:t.name,toolCallId:t.id,code:`unknown_tool`});if(i.length>0){let e=i.find(e=>e.code===`unknown_tool`),t=e?`Model requested tool "${e.name}", which was not in the tools offered ([${[...n.keys()].join(`, `)}]).`:`Duplicate tool call id "${i[0].toolCallId}" in the model's response.`;throw new p(i.length>1?`${t} (${i.length} tool call issues total, see error.issues.)`:t,`validation`,{code:e?`unknown_tool`:`duplicate_tool_call_id`,issues:i})}for(let t of e){let e=n.get(t.name);if(!e?.argumentsSchema)continue;let r=e.argumentsSchema.safeParse(t.arguments);if(!r.success)throw new p(`Arguments for tool call "${t.name}" failed validation`,`validation`,{issues:r.error})}}function Tt(e){let{rawContent:t,wireToolCalls:n,params:r,useJson:i,parseJson:a,requestId:o,logger:s,redactText:c}=e,l=t?.trim();if(!l&&!n?.length)throw new p(`Empty LLM response`,`api`,{code:`empty_response`});if(s.debug(`[VernLLM:${o}] output:\n`+c(l??`[${n?.length??0} tool call(s)]`).slice(0,800)),n?.length){if(!r.tools)throw new p("Provider returned tool_calls but no `tools` were sent with this call.",`validation`,{code:`unexpected_tool_calls`});if(r.toolChoice===`none`)throw new p(`Provider returned tool_calls despite toolChoice: 'none'.`,`validation`,{code:`tool_choice_none_violated`});let e=mt(n);return wt(e,r.tools),{type:`tool_calls`,toolCalls:e,...l?{content:l}:{}}}let u=l??``;if(!i)return r.tools?{type:`content`,content:u}:u;let d=Ct(u,r.schema,a);return r.tools?{type:`content`,content:d}:d}function Et(e,t,n,r,i,a,o,s,c){let{gateway:l,usageReporter:u,logger:d,redactText:f,parseJson:m,detectSoftFailure:h}=c;try{let g=Tt({rawContent:e,wireToolCalls:t,params:n,useJson:r,parseJson:m,requestId:a,logger:d,redactText:f}),_=Dt(h,g,{requestId:a,model:c.model,providerName:c.providerName,isFallback:c.isFallback,attempt:o+1},d);if(_!==void 0)throw new p(`Soft failure detected`,`api`,{code:_});return l.recordSuccess(o,n.signal,s),u.reportSuccess(i),g}catch(e){let t=M(e,n.signal);throw i&&t.type!==`aborted`&&u.reportFailure(i,t,o),t}}function Dt(e,t,n,r){if(e)try{return e(t,n)}catch(e){r.warn(`detectSoftFailure threw and was ignored, treated as no soft failure: ${e instanceof Error?e.message:String(e)}`);return}}function Ot(e){let{capacity:t,logger:n,label:r}=e,i=[],a=[],o=!1,s=!1,c,l=!1;function u(e){let o=a.shift();if(o){o.resolve({done:!1,value:e});return}i.push(e),i.length>t*2&&(l||(l=!0,n.warn(`[VernLLM] ${r} buffer exceeded cap (${t}), evicting ${i.length-t} oldest item(s); buffered=${i.length}. The iterable was never read (or fell far behind) for this channel.`)),i.splice(0,i.length-t))}function d(){o=!0;for(let e of a.splice(0))e.resolve({done:!0,value:void 0})}function f(e){o=!0,s=!0,c=e;for(let t of a.splice(0))t.reject(e)}return{push:u,finish:d,fail:f,iterable:{[Symbol.asyncIterator](){return{next(){return i.length?Promise.resolve({done:!1,value:i.shift()}):o?s?Promise.reject(c):Promise.resolve({done:!0,value:void 0}):new Promise((e,t)=>{a.push({resolve:e,reject:t})})}}}}}}const kt=1e4,At=1e6;function jt(){let e=new Map;function t(t){let n=e.get(t.index);if(!n){if(e.size>=kt)throw new p(`Stream contained more than ${kt} distinct tool calls`,`validation`);n={args:``}}let r=t.argumentsDelta??``;if(n.args.length+r.length>At)throw new p(`Tool call arguments exceeded the ${At}-character stream limit`,`validation`);return n.id??=t.id,n.name??=t.name,n.args+=r,e.set(t.index,n),{type:`tool_call_delta`,index:t.index,id:t.id,name:t.name,argsDelta:t.argumentsDelta,complete:t.complete}}function n(){if(e.size)return[...e.entries()].sort(([e],[t])=>e-t).map(([,e])=>({id:e.id??``,type:`function`,function:{name:e.name??``,arguments:e.args}}))}return{apply:t,toWireToolCalls:n}}async function Mt(e,t){try{await e.return?.()}catch{}t.abort()}function Nt(e,t,n){let{requestId:r,model:i,providerName:a,isFallback:o,chunkIdleTimeoutMs:s,streamController:c,logger:l,signal:u,onRateLimitHint:d}=n,f,p,m=new Promise((e,t)=>{f=e,p=t});m.catch(()=>{});let h=Ot({capacity:1e4,logger:l,label:`stream chunk`}),{push:g,finish:_,fail:v}=h,y=h.iterable,b=jt(),x=``,S;return(async()=>{try{let n=t;for(;!n.done;){let t=n.value;t.type===`ping`||(t.type===`rate_limit_hint`?d?.(t.hint):t.type===`text-delta`?(x+=t.delta,g({type:`text-delta`,delta:t.delta})):t.type===`tool_call_delta`?g(b.apply(t)):t.type===`usage`&&(S=se(t.usage,{requestId:r,model:i,providerName:a,isFallback:o}),g({type:`usage`,usage:S}))),n=await be(()=>e.next(),s,()=>c.abort(),l)}}catch(t){await Mt(e,c);let r=M(t,u);try{n.onStreamFailure(r,S)}catch{}v(r),p(r);return}_();try{n.onStreamSuccess(S)}catch{}try{let e=b.toWireToolCalls(),t=n.finalize(x,e,S);f(t)}catch(e){p(e)}})(),{chunks:y,finalResult:m}}function Pt(e){let{providerName:t,isFallback:n,maxRetries:r,onUsage:i,onUsageFailure:a,logger:o}=e;function s(e,r,i){if(e.usage)return se(e.usage,{requestId:r,model:i,providerName:t,isFallback:n})}function c(e){if(e)return e.totalTokens||e.promptTokens+e.completionTokens}function l(e){if(!(!e||!i))try{i(e)}catch(e){o.error(`[VernLLM] onUsage failed`,{message:e instanceof Error?e.message:`unknown`})}}function u(e,t,n,i=!1){let s=e.totalTokens||e.promptTokens+e.completionTokens,c=i?`mid-stream failure (terminal, no further attempts)`:`attempt ${n+1}/${r+1}`;if(o.warn(`[VernLLM:${e.requestId}] usage failure, ${c}: type=${t.type} tokens=${s}`),a)try{a(e,t)}catch(e){o.error(`[VernLLM] onUsageFailure failed`,{message:e instanceof Error?e.message:`unknown`})}}return{extract:s,actualTokensFor:c,reportSuccess:l,reportFailure:u}}async function Ft(e,t,n,r){if(!e)return{};let i=await e.acquire(e.estimate(t),n);return i.waitedMs>0&&r(i.waitedMs,i.reason??`rpm`),{release:i.release}}async function It(e){let{params:t,requestId:n,attempt:r,onRequest:i,middlewareState:a,gateway:o,requestBuilder:s,providerName:c,limiter:l,middleware:u,middlewareTimeoutMs:d,logger:p,reportEvent:m}=e,h=a??T(),g=s.build(t),{useJson:_,model:v}=g,y=await tt({request:g.request,requestId:n,attempt:r,signal:t.signal,state:h,middleware:u,middlewareTimeoutMs:d,logger:p,reportEvent:m,buildContext:(e,t,n)=>o.buildAttemptContext(e,t,n)});i?.(f(c,v,y,void 0,Date.now()));let{release:b}=await Ft(l,y,t.signal,(e,i)=>{F({kind:`rate_limited`,requestId:n,provider:c,model:v,waitedMs:e,reason:i},o.buildAttemptContext(r,t.signal,h),m,u,d,p)});return{request:y,model:v,useJson:_,state:h,release:b}}async function Lt(e){let{fn:t,requestId:n,model:r,providerName:i,isFallback:a,supportsJsonObjectMode:o,breaker:s,budget:c,maxRetries:l,baseDelayMs:u,nonRetryableStatus:d,signal:f,onAttempt:p,middleware:m,middlewareTimeoutMs:h,logger:g,reportEvent:_,logLabel:v,redactText:y,countsTowardBreaker:b}=e,x=e.state??T(),S=[],C=dt({breaker:s,requestId:n,model:r,providerName:i,isFallback:a,supportsJsonObjectMode:o}),ee;try{return await ke({fn:(e,n)=>(c?.recordAttempt(e>0),t(e,n,x,C)),maxRetries:l,signal:f,onAttempt:p,attempts:S,shouldRetryAttempt:(e,t)=>{if(!De(e,d,j,t))return!1;if(c)try{c.assertAvailable()}catch(e){return ee=e,!1}return!0},recoverDelayForAttempt:(e,t)=>Oe({requestId:n,model:r,attempt:e,error:t,state:x,signal:f,providerName:i,maxRetries:l,baseDelayMs:u,middleware:m,middlewareTimeoutMs:h,logger:g,reportEvent:_,buildEventContext:(e,t,n,r,i)=>C.buildAttemptContext(n,r,i),extractStatus:j,normalizeError:M,emitEvent:F}),normalizeError:M})}catch(e){let t=M(ee??e,f,S.length>0?S:void 0);throw b(t)&&C.recordFailure(S.length,f,x,t.code),g.debug(`[VernLLM:${n}] ${v}:\n${y(Ge(e))}`),t}}function Rt(e){try{return JSON.parse(e)}catch{return}}var zt=class{providerName;client;model;maxRetries;timeoutMs;chunkIdleTimeoutMs;baseDelayMs;nonRetryableStatus;parseJson;logger;redact;usageReporter;reportEvent;breaker;budget;limiter;isFallback;requestBuilder;middleware;middlewareTimeoutMs;supportsJsonObjectMode;detectSoftFailure;constructor(e,t,n,r){this.providerName=e,this.client=t,this.model=n,this.maxRetries=r.maxRetries,this.timeoutMs=r.timeoutMs,this.chunkIdleTimeoutMs=r.chunkIdleTimeoutMs,this.baseDelayMs=r.baseDelayMs,this.nonRetryableStatus=r.nonRetryableStatus,this.parseJson=r.parseJson??Rt,this.logger=r.logger,this.redact=r.redact,this.reportEvent=I(r.onEvent,this.logger),this.breaker=r.breaker,this.budget=r.budget,this.limiter=r.limiter,this.isFallback=r.isFallback??!1,this.middleware=r.middleware??[],this.middlewareTimeoutMs=r.middlewareTimeoutMs??5e3,this.supportsJsonObjectMode=t.supportsJsonObjectMode??!0,this.detectSoftFailure=r.detectSoftFailure,this.usageReporter=Pt({providerName:this.providerName,isFallback:this.isFallback,maxRetries:this.maxRetries,onUsage:r.onUsage,onUsageFailure:r.onUsageFailure,logger:this.logger}),this.requestBuilder=new St({model:n,defaultMaxTokens:r.defaultMaxTokens,defaultTemperature:r.defaultTemperature,defaultReasoningEffort:r.defaultReasoningEffort,defaultBudgetTokens:r.defaultBudgetTokens,supportsJsonObjectMode:this.supportsJsonObjectMode})}previewRequest(e){let{model:t,request:n}=this.requestBuilder.build(e);return{model:t,request:n}}get jsonObjectModeSupported(){return this.supportsJsonObjectMode}getCircuitState(e){return this.breaker?.getState(e)}getFailureBreakdown(e){return this.breaker?.getFailureBreakdown(e)}getRetryBudgetState(){return this.budget?.getSnapshot()}get isolateByModel(){return this.breaker?.isolateByModel??!1}openCircuit(e,t){this.breaker?.open(e,t)}closeCircuit(e,t){this.breaker?.close(e,t)}assertBreakerClosed(e,t){this.breaker?.assertClosed(e??this.model,t)}async run(e,t,n,r){return Lt({fn:(n,r,i,a)=>this.executeCall(e,t,n,r,i,a),requestId:t,model:e.model??this.model,providerName:this.providerName,isFallback:this.isFallback,supportsJsonObjectMode:this.supportsJsonObjectMode,breaker:this.breaker,budget:this.budget,maxRetries:this.maxRetries,baseDelayMs:this.baseDelayMs,nonRetryableStatus:this.nonRetryableStatus,signal:e.signal,onAttempt:n,state:r,middleware:this.middleware,middlewareTimeoutMs:this.middlewareTimeoutMs,logger:this.logger,reportEvent:this.reportEvent,logLabel:`error`,redactText:e=>this.redactText(e),countsTowardBreaker:e=>this.countsTowardBreaker(e)})}async runStream(e,t,n,r){return Lt({fn:(n,r,i,a)=>this.executeStreamCall(e,t,n,r,i,a),requestId:t,model:e.model??this.model,providerName:this.providerName,isFallback:this.isFallback,supportsJsonObjectMode:this.supportsJsonObjectMode,breaker:this.breaker,budget:this.budget,maxRetries:this.maxRetries,baseDelayMs:this.baseDelayMs,nonRetryableStatus:this.nonRetryableStatus,signal:e.signal,onAttempt:n,state:r,middleware:this.middleware,middlewareTimeoutMs:this.middlewareTimeoutMs,logger:this.logger,reportEvent:this.reportEvent,logLabel:`stream-open error`,redactText:e=>this.redactText(e),countsTowardBreaker:e=>this.countsTowardBreaker(e)})}async executeCall(e,t,n,r,i,a){let{request:o,model:s,useJson:c,state:l,release:u}=await It({params:e,requestId:t,attempt:n,onRequest:r,middlewareState:i,gateway:a,requestBuilder:this.requestBuilder,providerName:this.providerName,limiter:this.limiter,middleware:this.middleware,middlewareTimeoutMs:this.middlewareTimeoutMs,logger:this.logger,reportEvent:this.reportEvent}),d=u;try{let r;try{r=await ye(e=>this.client.chat.completions.create(o,{signal:e}),this.timeoutMs,e.signal)}catch(e){throw this.reactToRateLimitError(e),e}this.limiter?.reactToRateLimitHint(ut(r));let i=this.usageReporter.extract(r,t,s),u=this.usageReporter.actualTokensFor(i),f=r.choices?.[0]?.message?.content,p=r.choices?.[0]?.message?.tool_calls,m;try{m=Et(f,p,e,c,i,t,n,l,{gateway:a,usageReporter:this.usageReporter,logger:this.logger,redactText:e=>this.redactText(e),parseJson:this.parseJson,detectSoftFailure:this.detectSoftFailure,providerName:this.providerName,isFallback:this.isFallback,model:s})}catch(e){throw d?.(u),d=void 0,e}return d?.(u,!0),d=void 0,m}finally{d?.()}}redactText(e){return this.redact?this.redact(e):e}reactToRateLimitError(e){this.limiter&&j(e)===429&&this.limiter.signalRateLimit()}async executeStreamCall(e,t,n,r,i,a){let o=this.client.chat.completions;if(!o.createStream)throw new p(`stream: true requires a client/adapter with createStream`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`createStream`}});let s=o.createStream.bind(o),{request:c,model:l,useJson:u,state:d,release:f}=await It({params:e,requestId:t,attempt:n,onRequest:r,middlewareState:i,gateway:a,requestBuilder:this.requestBuilder,providerName:this.providerName,limiter:this.limiter,middleware:this.middleware,middlewareTimeoutMs:this.middlewareTimeoutMs,logger:this.logger,reportEvent:this.reportEvent}),m=f,h=new AbortController,g=e.signal?AbortSignal.any([e.signal,h.signal]):h.signal;try{let{iterator:r,first:i}=await(async()=>{try{return await ye(async e=>{let t=s(c,{signal:e})[Symbol.asyncIterator]();return{iterator:t,first:await t.next()}},this.timeoutMs,g)}catch(e){throw this.reactToRateLimitError(e),e}})();if(i.done)throw new p(`Empty LLM response`,`api`);let o=m,f=Nt(r,i,{requestId:t,model:l,providerName:this.providerName,isFallback:this.isFallback,chunkIdleTimeoutMs:e.chunkIdleTimeoutMs??this.chunkIdleTimeoutMs,streamController:h,logger:this.logger,signal:e.signal,onRateLimitHint:e=>{this.limiter?.reactToRateLimitHint(e)},onStreamSuccess:e=>{},onStreamFailure:(t,r)=>{this.reactToRateLimitError(t),t.type===`timeout`&&a.recordFailure(n,e.signal,d,t.code),r&&t.type!==`aborted`&&this.usageReporter.reportFailure(r,t,n,!0),o?.(this.usageReporter.actualTokensFor(r))},finalize:(r,i,s)=>{let c=this.usageReporter.actualTokensFor(s);try{let f=Et(r,i,e,u,s,t,n,d,{gateway:a,usageReporter:this.usageReporter,logger:this.logger,redactText:e=>this.redactText(e),parseJson:this.parseJson,detectSoftFailure:this.detectSoftFailure,providerName:this.providerName,isFallback:this.isFallback,model:l});return o?.(c,!0),f}catch(t){throw o?.(c),t instanceof p&&this.countsTowardBreaker(t)&&a.recordFailure(n,e.signal,d,t.code),t}}});return m=void 0,f}finally{m?.()}}countsTowardBreaker(e){return e.countsTowardBreaker}};async function Bt(e,t,n,r,i,a=!1){let o=[];for(let s=0;s<e.executors.length;s++){let c=e.executors[s],l=Date.now(),u=0;try{return s===0&&a||c.assertBreakerClosed(t.model,{requestId:n,state:r,signal:t.signal}),{result:await i(c,()=>{u+=1}),executor:c,index:s,attemptCount:u}}catch(i){let a=M(i,t.signal);o.push({index:s-1,provider:c.providerName,model:t.model??c.model,error:a.toSnapshot()});let d=s===e.executors.length-1,f=e.fallbackOn(a,{isLastTarget:d});if((d?`stop`:f)===`stop`)throw o.length>1?new w(o):a;let p=e.executors[s+1],m=t.model??c.model,h={stage:`attempt`,requestId:n,requestedProvider:c.providerName,requestedModel:m,isFallbackAttempt:s>0,attempt:u||1,capabilities:{supportsJsonObjectMode:c.jsonObjectModeSupported},signal:t.signal,state:r,own:{}};F({kind:`fallback`,requestId:n,from:c.providerName,to:p.providerName,fromIndex:s-1,toIndex:s,error:a,elapsedMs:Date.now()-l},h,e.reportEvent,e.middleware,e.middlewareTimeoutMs,e.logger)}}throw new p(`No provider targets configured`,`invalid_params`)}function Vt(e,t){let n={provider:e.executor.providerName,model:t.model??e.executor.model,fallbackIndex:e.index-1,usedFallback:e.index>0,attempts:e.attemptCount};return t.meta&&(t.meta.current=n),{value:e.result,meta:n}}async function Ht(e,t,n,r,i){return Vt(await Bt(e,t,n,i,(e,r)=>e.run(t,n,r,i),r),t)}async function Ut(e,t,n,r,i){return Vt(await Bt(e,t,n,i,(e,r)=>e.runStream(t,n,r,i),r),t)}function Wt(e){let t=!1,n,r;function i(){return r||(t=!0,r=e().then(e=>(n=e,e)),r.catch(()=>{}),r)}return{call:i,wasCalled:()=>t,resolvedValue:()=>n}}async function V(e,t,n,r,i,a=!1){if(e.middleware.length===0||a)return i();let o=e.primaryExecutor,{model:s,request:c}=o.previewRequest(t),l=i;for(let i=e.middleware.length-1;i>=0;i--){let a=e.middleware[i],u=N(a,i),d=l;l=async()=>{let i={stage:`pre-dispatch`,requestId:n,primaryProvider:o.providerName,primaryModel:s,capabilities:{supportsJsonObjectMode:o.jsonObjectModeSupported},signal:t.signal,state:r,own:{}};if(!await P(a,i,u,e.middlewareTimeoutMs,e.logger))return a.enabled!==void 0&&F({kind:`middleware`,requestId:n,middleware:u,hook:`enabled_skip`},i,e.reportEvent,e.middleware,e.middlewareTimeoutMs,e.logger),d();if(!a.wrap)return d();let l=Wt(d);try{let t=await a.wrap(c,l.call,i);return l.wasCalled()||F({kind:`middleware`,requestId:n,middleware:u,hook:`wrap_short_circuit`},i,e.reportEvent,e.middleware,e.middlewareTimeoutMs,e.logger),t}catch(t){let n=l.resolvedValue();if(n!==void 0)return e.logger.error(`[VernLLM] middleware "${u}".wrap threw after next() resolved; keeping the original result`,{message:t instanceof Error?t.message:`unknown`}),n;throw rt(t,u,i.signal)}}}return l()}const H=Symbol(`deadlineExceeded`);function Gt(e,t){if(e===void 0)return{signal:t,timer:void 0};let n=new AbortController;if(e<=0)return n.abort(H),{signal:t?AbortSignal.any([t,n.signal]):n.signal,timer:void 0};let r=setTimeout(()=>n.abort(H),e);return{signal:t?AbortSignal.any([t,n.signal]):n.signal,timer:r}}function Kt(e,t){return e instanceof p&&e.type===`aborted`&&e.code===void 0&&t?.reason===H&&(e.code=`deadline_exceeded`),e}var qt=class{capacity;refillPerMs;available;lastRefill=Date.now();constructor(e,t){this.capacity=e,this.refillPerMs=t,this.available=e}refill(){if(this.refillPerMs===0)return;let e=Date.now(),t=e-this.lastRefill;this.available=Math.min(this.capacity,this.available+Math.max(0,t)*this.refillPerMs),this.lastRefill=e}tryTake(e){return this.refill(),this.available<e?!1:(this.available-=e,!0)}msUntilAvailable(e){return this.refill(),this.available>=e?0:this.refillPerMs===0?1/0:(e-this.available)/this.refillPerMs}give(e){this.available=Math.min(this.capacity,this.available+e)}getCapacity(){return this.capacity}resize(e){this.refill(),this.refillPerMs>0&&(this.refillPerMs=this.refillPerMs/this.capacity*e),this.capacity=e,this.available=Math.min(this.available,e)}};function Jt(e){let t=e.messages.reduce((e,t)=>{let n=t.content;if(typeof n==`string`)return e+n.length;if(n==null)return e;try{return e+JSON.stringify(n).length}catch{return e}},0);return Math.ceil(t/4)+(e.max_tokens??0)}function Yt(e){if(!e)return;let t=t=>{if(!Number.isFinite(e[t]))throw new p(`aimd.${t} (${e[t]}) must be a finite number.`,`invalid_params`)};for(let e of[`minCapacity`,`maxCapacity`,`increaseBy`,`decreaseFactor`])t(e);if(e.minCapacity<1||e.maxCapacity<1)throw new p(`aimd.minCapacity (${e.minCapacity}) and aimd.maxCapacity (${e.maxCapacity}) must both be at least 1, since the requests bucket always takes 1 per acquire; a capacity below 1 could never be satisfied.`,`invalid_params`);if(e.minCapacity>e.maxCapacity)throw new p(`aimd.minCapacity (${e.minCapacity}) must not exceed aimd.maxCapacity (${e.maxCapacity}).`,`invalid_params`);return{increaseBy:Math.max(0,e.increaseBy),decreaseFactor:Math.min(1,Math.max(Number.MIN_VALUE,e.decreaseFactor)),minCapacity:e.minCapacity,maxCapacity:e.maxCapacity,proactiveFloor:e.proactiveFloor??0}}function Xt(e,t){return e===`tpm`?t:1}function Zt(e){if(e)return new qt(e,e/6e4)}var Qt=class{requests;tokens;concurrency;buckets;maxQueueMs;maxQueueSize;estimateTokensFn;aimd;queue=[];wakeTimer;constructor(e){this.requests=Zt(e.requestsPerMinute),this.tokens=Zt(e.tokensPerMinute),e.maxConcurrent&&(this.concurrency=new qt(e.maxConcurrent,0)),this.buckets=[{reason:`concurrency`,bucket:this.concurrency},{reason:`rpm`,bucket:this.requests},{reason:`tpm`,bucket:this.tokens}].flatMap(({reason:e,bucket:t})=>t?[{reason:e,bucket:t}]:[]),this.maxQueueMs=e.maxQueueMs??3e4,this.maxQueueSize=e.maxQueueSize??0,this.estimateTokensFn=e.estimateTokens??Jt,this.aimd=this.requests?Yt(e.aimd):void 0}estimate(e){return this.estimateTokensFn(e)}async acquire(e,t){if(t?.aborted)throw new p(`LLM request aborted`,`aborted`);if(!Number.isFinite(e)||e<0)throw new p(`estimatedTokens must be a finite, non-negative number, got ${String(e)}`,`invalid_params`);if(this.tokens&&e>this.tokens.getCapacity())throw new p(`estimatedTokens (${e}) exceeds the configured tokensPerMinute capacity (${this.tokens.getCapacity()}); this call could never acquire capacity.`,`rate_limited`,{code:`rate_limit_capacity_exceeded`});if(this.queue.length===0){let n=this.tryAcquireBuckets(e);return n.ok?{release:this.makeRelease(e),waitedMs:0}:this.enqueue(e,n.reason,t)}if(this.maxQueueSize>0&&this.queue.length>=this.maxQueueSize)throw this.queueFullError();return this.enqueue(e,void 0,t)}queueFullError(){return new p(`Rate limit queue is full`,`rate_limited`,{code:`rate_limit_queue_full`})}enqueue(e,t,n){return new Promise((r,i)=>{let a={estimatedTokens:e,enqueuedAt:Date.now(),lastReason:t,resolve:e=>{c(),r(e)},reject:e=>{c(),this.wakeTimer&&=(clearTimeout(this.wakeTimer),void 0),this.drain(),i(e)}},o,s=()=>{a.reject(new p(`LLM request aborted`,`aborted`))},c=()=>{o&&clearTimeout(o),n?.removeEventListener(`abort`,s);let e=this.queue.indexOf(a);e!==-1&&this.queue.splice(e,1)};this.maxQueueMs>0&&(o=setTimeout(()=>{a.reject(new p(`Rate limit queue timed out before capacity was available`,`rate_limited`,{code:`rate_limit_queue_timeout`}))},this.maxQueueMs)),n?.addEventListener(`abort`,s,{once:!0}),this.queue.push(a),this.drain()})}tryAcquireBuckets(e){let t=[];for(let{reason:n,bucket:r}of this.buckets){let i=Xt(n,e);if(!r.tryTake(i)){for(let e of t)e.bucket.give(e.amount);return{ok:!1,reason:n}}t.push({bucket:r,amount:i})}return{ok:!0}}drain(){for(;this.queue.length>0;){let e=this.queue[0],t=this.tryAcquireBuckets(e.estimatedTokens);if(!t.ok){e.lastReason=t.reason,this.scheduleWake(t.reason,e.estimatedTokens);return}let n=Date.now()-e.enqueuedAt;e.resolve({release:this.makeRelease(e.estimatedTokens),waitedMs:n,reason:e.lastReason})}}scheduleWake(e,t){if(this.wakeTimer||e===`concurrency`)return;let n=(this.buckets.find(t=>t.reason===e)?.bucket)?.msUntilAvailable(Xt(e,t));if(n===void 0||!Number.isFinite(n))return;let r=Math.min(Math.max(1,Math.ceil(n)),2147483647);this.wakeTimer=setTimeout(()=>{this.wakeTimer=void 0,this.drain()},r)}makeRelease(e){let t=!1;return(n,r=!1)=>{t||(t=!0,this.concurrency?.give(1),this.tokens&&n!==void 0&&Number.isFinite(n)&&this.tokens.give(e-n),r&&this.growOnSuccess(),this.drain())}}resizeRequestsCeiling(e){!this.aimd||!this.requests||this.requests.resize(e(this.aimd,this.requests.getCapacity()))}growOnSuccess(){this.resizeRequestsCeiling((e,t)=>Math.min(t+e.increaseBy,e.maxCapacity))}signalRateLimit(){this.resizeRequestsCeiling((e,t)=>Math.max(t*e.decreaseFactor,e.minCapacity))}reactToRateLimitHint(e){if(!this.aimd||!e||e.remainingRequests===void 0)return;let t=this.aimd.proactiveFloor??0;t>0&&e.remainingRequests<=t&&this.signalRateLimit()}};const U=[`estimate`,`acquire`,`signalRateLimit`,`reactToRateLimitHint`];function $t(e){let t=e;return U.every(e=>typeof t[e]==`function`)}function en(e){let t=e;return U.some(e=>typeof t[e]==`function`)}function tn(e){if(e!==void 0){if($t(e))return e;if(en(e))throw new p(`rateLimit looks like a RateLimiterAdapter but is missing: ${U.filter(t=>typeof e[t]!=`function`).join(`, `)}. All four methods (${U.join(`, `)}) are required.`,`invalid_params`);return new Qt(e)}}function nn(e,t,n){return[e,...t].map((e,t)=>{let r=t>0,i=e.name??(r?`fallback[${t-1}]`:n.providerName),a=at(e.circuitBreaker,i,e.model,n.onEvent,n.logger,n.middleware,n.middlewareTimeoutMs,r,e.client.supportsJsonObjectMode??!0),o=e.retryBudget?new Be(e.retryBudget):void 0;return new zt(i,e.client,e.model,{maxRetries:e.maxRetries??n.maxRetries??1,timeoutMs:e.timeoutMs??n.timeoutMs??25e3,chunkIdleTimeoutMs:e.chunkIdleTimeoutMs??n.chunkIdleTimeoutMs??3e4,baseDelayMs:e.baseDelayMs??n.baseDelayMs??500,defaultMaxTokens:e.defaultMaxTokens??n.defaultMaxTokens??1e3,defaultTemperature:e.defaultTemperature===void 0?n.primaryDefaultTemperature:e.defaultTemperature,defaultReasoningEffort:e.defaultReasoningEffort===void 0?n.primaryDefaultReasoningEffort:e.defaultReasoningEffort,defaultBudgetTokens:e.defaultBudgetTokens===void 0?n.primaryDefaultBudgetTokens:e.defaultBudgetTokens,nonRetryableStatus:e.nonRetryableStatus??n.nonRetryableStatus??[400,401,403,404,422],parseJson:n.parseJson,logger:n.logger,redact:n.redact,onUsage:n.onUsage,onUsageFailure:n.onUsageFailure,onEvent:n.onEvent,breaker:a,budget:o,limiter:tn(e.rateLimit),isFallback:r,middleware:n.middleware,middlewareTimeoutMs:n.middlewareTimeoutMs,detectSoftFailure:e.detectSoftFailure??n.detectSoftFailure})})}function rn(e){let t=e;return typeof t.get==`function`&&typeof t.set==`function`}function an(e){if(e===void 0)return new b;if(rn(e))return e;if(e.maxSize!==void 0&&(!Number.isFinite(e.maxSize)||e.maxSize<0||!Number.isInteger(e.maxSize)))throw RangeError(`cache.maxSize must be a finite, non-negative integer, got ${String(e.maxSize)}`);return new b(e.maxSize,e.eviction)}function on(e){return{debug:W(e,`debug`),warn:W(e,`warn`),error:W(e,`error`)}}function W(e,t){let n=e[t].bind(e);return((...e)=>{try{cn(n(...e))}catch{}})}function sn(e){return typeof e?.then==`function`}function cn(e){sn(e)&&Promise.resolve(e).catch(()=>{})}var ln=class{debugEnabled;constructor(e){this.debugEnabled=e}debug(e){this.debugEnabled&&console.debug(e)}warn(e){console.warn(e)}error(e,t){console.error(e,t??``)}},un=class{logger;executors;fallbackOn;reportEvent;cacheOrchestrator;middleware;middlewareTimeoutMs;cachedCallInnerParams=new WeakMap;cachedCallMeta=new Map;constructor(e){this.logger=on(e.logger??new ln(e.debug??!1));let t=e.name??`primary`;this.cacheOrchestrator=new he(an(e.cache),this.logger),this.fallbackOn=e.fallbackOn??ne,this.reportEvent=I(e.onEvent,this.logger),this.middleware=[...e.middleware??[]].sort((e,t)=>(e.priority??0)-(t.priority??0)),this.middlewareTimeoutMs=e.middlewareTimeoutMs??5e3;let n=e.defaultTemperature===void 0?.2:e.defaultTemperature,r=e.defaultReasoningEffort,i=e.defaultBudgetTokens,a={client:e.client,model:e.model,name:t,maxRetries:e.maxRetries,timeoutMs:e.timeoutMs,chunkIdleTimeoutMs:e.chunkIdleTimeoutMs,baseDelayMs:e.baseDelayMs,defaultMaxTokens:e.defaultMaxTokens,defaultTemperature:n,defaultReasoningEffort:r,defaultBudgetTokens:i,nonRetryableStatus:e.nonRetryableStatus,circuitBreaker:e.circuitBreaker,rateLimit:e.rateLimit,retryBudget:e.retryBudget,detectSoftFailure:e.detectSoftFailure},o=Array.isArray(e.fallback)?e.fallback:e.fallback?[e.fallback]:[];this.executors=nn(a,o,{providerName:t,primaryDefaultTemperature:n,primaryDefaultReasoningEffort:r,primaryDefaultBudgetTokens:i,maxRetries:e.maxRetries,timeoutMs:e.timeoutMs,chunkIdleTimeoutMs:e.chunkIdleTimeoutMs,baseDelayMs:e.baseDelayMs,defaultMaxTokens:e.defaultMaxTokens,nonRetryableStatus:e.nonRetryableStatus,parseJson:e.parseJson,redact:e.redact,onUsage:e.onUsage,onUsageFailure:e.onUsageFailure,onEvent:e.onEvent,logger:this.logger,middleware:this.middleware,middlewareTimeoutMs:this.middlewareTimeoutMs,detectSoftFailure:e.detectSoftFailure})}logRefundError(e,t){this.logger.error(e,{message:t instanceof Error?t.message:`unknown`})}get logicalCallDependencies(){return{executors:this.executors,fallbackOn:this.fallbackOn,reportEvent:this.reportEvent,middleware:this.middleware,middlewareTimeoutMs:this.middlewareTimeoutMs,logger:this.logger}}get runOperationDependencies(){return{middleware:this.middleware,primaryExecutor:this.executors[0],middlewareTimeoutMs:this.middlewareTimeoutMs,logger:this.logger,reportEvent:this.reportEvent}}async call(e){if(e.signal?.aborted)throw new p(`LLM request aborted`,`aborted`);let t=e.requestId??globalThis.crypto.randomUUID(),n=this.cachedCallInnerParams.get(e),r=n!==void 0,{signal:i,timer:a}=Gt(e.deadlineMs,e.signal),o=i===e.signal?e:{...e,signal:i},s=n??T();try{let e=this.executors.length===1,n=await V(this.runOperationDependencies,o,t,s,async()=>{if(e){if(o.signal?.aborted)throw new p(`LLM request aborted`,`aborted`);this.executors[0].assertBreakerClosed(o.model,{requestId:t,state:s,signal:o.signal})}let n,r=async e=>{let t=await e;return n=t.meta,t.value};return o.stream?{value:await ue(o,()=>r(Ut(this.logicalCallDependencies,o,t,e,s)),o.signal,(e,t)=>this.logRefundError(e,t)),meta:n}:{value:await E(o,!1,()=>r(Ht(this.logicalCallDependencies,o,t,e,s)),o.signal,(e,t)=>this.logRefundError(e,t)),meta:n}},r);return o.stream,n.value}catch(e){throw Kt(e,i)}finally{clearTimeout(a)}}runCached(e){return this.cacheOrchestrator.runCached(e)}async deleteCache(e){await this.cacheOrchestrator.deleteCache(e)}async cachedCall(e){let{call:t,...n}=e,r=t;if(r.reserveUsage||r.refundUsage)throw new p("`reserveUsage`/`refundUsage` were set inside `call`, where cachedCall ignores them. Move them to the top level of the cachedCall() params, alongside cacheKey/ttl, instead.",`invalid_params`);let i=r.requestId??globalThis.crypto.randomUUID(),a=T(),o=await this.cacheOrchestrator.resolveCacheKey(n.cacheKey),s=this.cachedCallMeta.get(o),c=!1;s||(s={},c=!0,this.cachedCallMeta.set(o,s));let l=()=>{c&&this.cachedCallMeta.delete(o)},u=r.meta,d=e=>{let t={...e,meta:s};return this.cachedCallInnerParams.set(t,a),this.call(t).finally(()=>{u&&(u.current=s.current),this.cachedCallInnerParams.delete(t)})};if(r.stream){let e={...r,requestId:i},t;try{t=await V(this.runOperationDependencies,e,i,a,async()=>({value:await this.cacheOrchestrator.runCachedStream({...n,openStream:()=>d(e)},!!r.tools),meta:s.current}))}catch(e){throw l(),e}let o=t.value,c=()=>{u&&(u.current=s.current),l()};return o.finalResult.then(c,c),o}let f={...r,requestId:i};try{let e=await V(this.runOperationDependencies,f,i,a,async()=>({value:await this.runCached({...n,fn:()=>d(f)}),meta:s.current}));return u&&(u.current=s.current),e.value}finally{l()}}getCircuitState(e){let t=L(this.executors,e?.index??0,`getCircuitState`);return R(t.isolateByModel,e?.model,`getCircuitState`,this.logger),t.getCircuitState(e?.model??t.model)}getFailureBreakdown(e){let t=L(this.executors,e?.index??0,`getFailureBreakdown`);return R(t.isolateByModel,e?.model,`getFailureBreakdown`,this.logger),t.getFailureBreakdown(e?.model??t.model)}getRetryBudgetState(e){return L(this.executors,e?.index??0,`getRetryBudgetState`).getRetryBudgetState()}getCircuitStates(e){return this.executors.map((t,n)=>({provider:t.providerName,index:n,isFallback:n>0,isolateByModel:t.isolateByModel,state:t.getCircuitState(e??t.model)}))}openCircuit(e){let t=L(this.executors,e?.index??0,`openCircuit`);R(t.isolateByModel,e?.model,`openCircuit`,this.logger),t.openCircuit(e?.model??t.model,{requestId:globalThis.crypto.randomUUID(),state:T()})}closeCircuit(e){let t=L(this.executors,e?.index??0,`closeCircuit`);R(t.isolateByModel,e?.model,`closeCircuit`,this.logger),t.closeCircuit(e?.model??t.model,{requestId:globalThis.crypto.randomUUID(),state:T()})}};function dn(e){return e}function fn(e){return e}async function*pn(e){let t=new TextDecoder(`utf-8`,{fatal:!0}),n=``;for await(let r of e){let e;try{e=typeof r==`string`?r:t.decode(r,{stream:!0})}catch(e){throw new p(`Invalid UTF-8 in SSE stream`,`parse`,{cause:e})}n=(n+e).replace(/\r\n/g,`
|
|
1
|
+
Object.defineProperty(exports,Symbol.toStringTag,{value:`Module`});const e=new Set([`unknown_tool`,`duplicate_tool_call_id`,`tool_choice_none_violated`,`unexpected_tool_calls`]),t=new Set([`rate_limit_queue_full`,`rate_limit_queue_timeout`,`rate_limit_capacity_exceeded`,`retry_budget_exhausted`]),n=new Set([`middleware_timeout`]),r=new Set([`parse`,`validation`,`invalid_params`,`aborted`]);function i(i,a){return!(r.has(i)||a&&e.has(a)||a&&t.has(a)||a&&n.has(a))}const a=new Set([`quota_exceeded`]);function o(e,t){return!(!i(e,t)||a.has(e))}function s(e){if(e!==void 0)try{return JSON.stringify(e),e}catch{return`[Unserializable: issues contained a circular reference]`}}function c(e){if(e!==void 0)try{return JSON.parse(JSON.stringify(e))}catch{return`[Unserializable: request body contained a circular reference]`}}const l=new Set([`authorization`,`x-api-key`,`x-goog-api-key`,`api-key`]);function u(e){if(e===void 0)return;let t={};for(let[n,r]of Object.entries(e))l.has(n.toLowerCase())||(t[n]=r);return t}function d(e,t=0){if(e!==void 0)return t>=20?[]:e.map(e=>({...e,error:{...e.error,issues:s(e.error.issues),attempts:d(e.error.attempts,t+1)},request:e.request&&{...e.request,body:c(e.request.body),headers:u(e.request.headers)}}))}function f(e,t,n,r,i=Date.now()){return{provider:e,model:t,body:c(n),headers:u(r),startedAt:i}}var p=class extends Error{type;status;issues;cause;retryAfterMs;code;attempts;constructor(e,t,n={}){super(e),this.type=t,this.name=`LLMError`,this.status=n.status,this.issues=n.issues,this.cause=n.cause,this.retryAfterMs=n.retryAfterMs,this.code=n.code,this.attempts=n.attempts}get retryable(){return i(this.type,this.code)}get countsTowardBreaker(){return o(this.type,this.code)}toSnapshot(){return{message:this.message,type:this.type,status:this.status,issues:s(this.issues),retryAfterMs:this.retryAfterMs,code:this.code,retryable:this.retryable,attempts:d(this.attempts)}}toJSON(){return{name:this.name,message:this.message,type:this.type,status:this.status,issues:s(this.issues),retryAfterMs:this.retryAfterMs,code:this.code,retryable:this.retryable,attempts:d(this.attempts)}}};function m(e){return e instanceof p}function h(e,t){return e.code===t&&e.issues!==void 0}const g={onAccess(){},onInsert(){},getEvictee:e=>e.keys().next().value};function _(e,t){if(!e.has(t))return;let n=e.get(t);e.delete(t),e.set(t,n)}const v={onAccess:_,onInsert:_,getEvictee:e=>e.keys().next().value};function y(e){return e===`lru`?v:g}var b=class{maxSize;store=new Map;eviction;constructor(e=1e3,t=`fifo`){this.maxSize=e,this.eviction=y(t)}async get(e){let t=this.store.get(e);return t?Date.now()>=t.expiresAt?(this.store.delete(e),{hit:!1,value:null}):(this.eviction.onAccess(this.store,e),{hit:!0,value:t.value}):{hit:!1,value:null}}async set(e,t,n){this.cleanupExpiredEntries(),this.store.set(e,{value:t,expiresAt:Date.now()+n*1e3}),this.eviction.onInsert(this.store,e),this.enforceSizeLimit()}async delete(e){this.store.delete(e)}cleanupExpiredEntries(){let e=Date.now();for(let[t,n]of this.store)e>=n.expiresAt&&this.store.delete(t)}enforceSizeLimit(){for(;this.store.size>this.maxSize;){let e=this.eviction.getEvictee(this.store);if(e===void 0)break;this.store.delete(e)}}},x=class{inner;constructor(e=new b){this.inner=e}normalize(e){return e.toLowerCase().trim().replace(/[^\p{L}\p{N}\s]/gu,` `).replace(/\s+/g,` `).trim()}async resolveKey(e){return this.normalize(e)}async get(e){return this.inner.get(this.normalize(e))}async set(e,t,n){await this.inner.set(this.normalize(e),t,n)}async delete(e){await this.inner.delete?.(this.normalize(e))}},S=class{l1;l2;l1Ttl;constructor(e,t,n){this.l1=e,this.l2=t,this.l1Ttl=n}async resolveKey(e){return this.l1.resolveKey?this.l1.resolveKey(e):this.l2.resolveKey?this.l2.resolveKey(e):e}async get(e){let t=await this.l1.get(e);if(t.hit)return t;let n=await this.l2.get(e);return n.hit&&await this.l1.set(e,n.value,this.l1Ttl??60),n}async set(e,t,n){await Promise.all([this.l1.set(e,t,this.l1Ttl??n),this.l2.set(e,t,n)])}async delete(e){await Promise.all([this.l1.delete?.(e),this.l2.delete?.(e)])}};function C(e){return e}function ee(e){if(typeof e!=`object`||!e)return!1;let t=e;return t.type===`tool_calls`&&Array.isArray(t.toolCalls)}function te(e){return typeof e==`object`&&!!e&&typeof e.then==`function`}function ne(e){return typeof e==`object`&&!!e&&typeof e[Symbol.asyncIterator]==`function`}function re(e){if(typeof e!=`object`||!e)return!1;let t=e;return ne(t.chunks)&&te(t.finalResult)}const ie=new Set([`unknown_tool`,`duplicate_tool_call_id`,`tool_choice_none_violated`,`unexpected_tool_calls`]),ae=e=>e.type===`parse`||e.type===`validation`||e.type===`aborted`||e.type===`quota_exceeded`||e.code&&ie.has(e.code)?`stop`:`next`;var w=class extends p{attempts;constructor(e){let t=e[e.length-1]?.error;super(`${e.length} provider${e.length===1?``:`s`} attempted and failed: ${e.map(e=>`${e.provider}(${e.error.type})`).join(` then `)}`,`fallback_exhausted`,{status:t?.status,cause:t,retryAfterMs:t?.retryAfterMs,code:`fallback_exhausted`,attempts:e}),this.attempts=e}get retryable(){let e=this.attempts[this.attempts.length-1]?.error;return e?e.retryable:super.retryable}};function oe(e){return e instanceof w}function se(){return{}}function ce(e){return{debugName:e}}function T(){let e=new Map;return{get(t){return e.get(t)},set(t,n){e.set(t,n)}}}function le(e){let{onError:t,...n}=e;return t?{...n,wrap:async(e,n,r)=>{try{return await n()}catch(e){try{await t(e,r)}catch{}throw e}}}:n}function ue(e,t){let n=e.completion_tokens_details?.reasoning_tokens;return{promptTokens:e.prompt_tokens??0,completionTokens:e.completion_tokens??0,totalTokens:e.total_tokens??0,...n===void 0?{}:{reasoningTokens:n},requestId:t.requestId,model:t.model,provider:t.providerName,usedFallback:t.isFallback}}async function de(e,t,n){if(!e.reserveUsage)return!1;try{return await e.reserveUsage({coalesced:t,signal:n}),!0}catch(e){throw n?.aborted?new p(`LLM request aborted`,`aborted`):new p(e instanceof Error?e.message:`Usage reservation failed`,`quota_exceeded`,{cause:e})}}function fe(e,t,n,r){return async i=>{try{await e.refundUsage?.({coalesced:t,signal:n})}catch(e){r(i,e)}}}async function E(e,t,n,r,i){if(r?.aborted)throw new p(`LLM request aborted`,`aborted`);let a=await de(e,t,r),o=fe(e,t,r,i);if(r?.aborted)throw a&&await o(`[VernLLM] refundUsage failed after abort`),new p(`LLM request aborted`,`aborted`);let s;try{s=await n()}catch(e){throw a&&await o(`[VernLLM] refundUsage failed`),e}if(r?.aborted)throw a&&await o(`[VernLLM] refundUsage failed after abort`),new p(`LLM request aborted`,`aborted`);return s}async function pe(e,t,n,r){if(n?.aborted)throw new p(`LLM request aborted`,`aborted`);let i=await de(e,!1,n),a=fe(e,!1,n,r);if(n?.aborted)throw i&&await a(`[VernLLM] refundUsage failed after abort`),new p(`LLM request aborted`,`aborted`);let o;try{o=await t()}catch(e){throw i&&await a(`[VernLLM] refundUsage failed after stream-open failure`),e}let s=o.finalResult.then(e=>e,async e=>{throw i&&await a(`[VernLLM] refundUsage failed after stream error`),e});return s.catch(()=>{}),{chunks:o.chunks,finalResult:s}}function me(){let e=new Map;function t(t){return e.get(t)}function n(t,n){return e.set(t,n),n.catch(()=>{}).finally(()=>{e.get(t)===n&&e.delete(t)}),n}return{get:t,track:n}}function he(e){return typeof e==`string`?e:JSON.stringify(e)??``}function ge(e,t){let n=[];if(t){let t=e;t.type===`tool_calls`?(t.toolCalls.forEach((e,t)=>{n.push({type:`tool_call_delta`,index:t,id:e.id,name:e.name,argsDelta:JSON.stringify(e.arguments??{}),complete:!0})}),t.content&&n.push({type:`text-delta`,delta:t.content})):n.push({type:`text-delta`,delta:he(t.content)})}else n.push({type:`text-delta`,delta:he(e)});return{async*[Symbol.asyncIterator](){for(let e of n)yield e}}}function _e(e,t){return{async*[Symbol.asyncIterator](){yield*ge(await e,t)}}}var ve=class{cache;logger;inFlight=me();constructor(e,t){this.cache=e,this.logger=t}async resolveCacheKey(e){return this.cache.resolveKey?await this.cache.resolveKey(e):e}async deleteCache(e){if(this.cache.delete)try{await this.cache.delete(await this.resolveCacheKey(e))}catch(e){this.logger.warn(`[VernLLM] cache delete failed: ${e instanceof Error?e.message:`unknown`}`)}}logRefundError(e,t){this.logger.error(e,{message:t instanceof Error?t.message:`unknown`})}async getCached(e){try{return await this.cache.get(e)}catch(e){return this.logger.warn(`[VernLLM] cache read failed: ${e instanceof Error?e.message:`unknown`}`),{hit:!1}}}async resolveKeyAndReadCache(e){let t=await this.resolveCacheKey(e.cacheKey);return{resolvedParams:t===e.cacheKey?e:{...e,cacheKey:t},cached:await this.getCached(t)}}joinInFlight(e,t){return E(e,!0,()=>t,e.signal,(e,t)=>this.logRefundError(e,t))}async runCached(e){let{resolvedParams:t,cached:n}=await this.resolveKeyAndReadCache(e);if(n.hit)return n.value;let r=this.inFlight.get(t.cacheKey);return r?this.joinInFlight(t,r):this.registerTrigger(t)}registerTrigger(e){let t=E(e,!1,()=>this.runAndCache(e),e.signal,(e,t)=>this.logRefundError(e,t));return this.inFlight.track(e.cacheKey,t),t}async runAndCache(e){let t=await e.fn();try{await this.cache.set(e.cacheKey,t,e.ttl)}catch(e){this.logger.warn(`[VernLLM] cache write failed: ${e instanceof Error?e.message:`unknown`}`)}return t}async runCachedStream(e,t){let{resolvedParams:n,cached:r}=await this.resolveKeyAndReadCache(e);if(r.hit){let e=r.value;return{chunks:ge(e,t),finalResult:Promise.resolve(e)}}let i=this.inFlight.get(n.cacheKey);if(i){let e=this.joinInFlight(n,i);return e.catch(()=>{}),{chunks:_e(e,t),finalResult:e}}return this.registerStreamTrigger(n)}registerStreamTrigger(e){let t,n,r=new Promise((e,r)=>{t=e,n=r});this.inFlight.track(e.cacheKey,r);let i=pe(e,async()=>{let t=await e.openStream(),n=t.finalResult.then(async t=>{try{await this.cache.set(e.cacheKey,t,e.ttl)}catch(e){this.logger.warn(`[VernLLM] cache write failed: ${e instanceof Error?e.message:`unknown`}`)}return t},e=>{throw e});return{chunks:t.chunks,finalResult:n}},e.signal,(e,t)=>this.logRefundError(e,t));return i.then(e=>{e.finalResult.then(t,n)},e=>{n(e)}),i}};const ye=1e4;function be(e){return!e||e<=0||e===1/0?void 0:e}function xe(e){return Math.min(e,2147483647)}async function Se(e,t,n){let r=new AbortController,i=be(t),a=i===void 0?void 0:setTimeout(()=>{r.abort()},xe(i)),o=n?AbortSignal.any([n,r.signal]):r.signal;try{return await e(o)}catch(e){throw r.signal.aborted&&!n?.aborted&&e instanceof DOMException&&e.name===`AbortError`?new p(`Request timed out`,`timeout`,{code:`request_timeout`}):e}finally{clearTimeout(a)}}function Ce(e,t,n,r){let i=be(t);if(i===void 0)return e();let a=!1;return new Promise((t,o)=>{let s=setTimeout(()=>{a=!0,n?.(),o(new p(`No stream chunk received for ${i}ms (idle timeout)`,`timeout`,{code:`idle_timeout`}))},xe(i));e().then(e=>{if(clearTimeout(s),a){r?.debug(`[VernLLM] chunk resolved after idle timeout already fired; discarding`);return}a=!0,t(e)},e=>{if(clearTimeout(s),a){r?.debug(`[VernLLM] chunk rejection arrived after idle timeout already fired; discarding`);return}a=!0,o(e)})})}const we=[{name:`Retry-After-Ms`,kind:`ms`},{name:`X-Retry-After-Ms`,kind:`ms`},{name:`Retry-After`,kind:`seconds`},{name:`Retry-After`,kind:`date`}];function Te(e,t){let n=e;if(typeof n.get==`function`)return n.get(t)??void 0;let r=Object.entries(e).find(([e])=>e.toLowerCase()===t.toLowerCase())?.[1];return typeof r==`string`?r:void 0}function D(e,t){return e<0?void 0:Math.min(e,t)}function Ee(e,t=ye){if(!e||typeof e!=`object`)return;let n=e,r=n.headers??n.response?.headers;if(!(!r||typeof r!=`object`))for(let{name:e,kind:n}of we){let i=Te(r,e)?.trim();if(i){if(n===`ms`&&/^\d+$/.test(i))return D(Number(i),t);if(n===`seconds`&&/^\d+(\.\d+)?$/.test(i))return D(Number(i)*1e3,t);if(n===`date`){let e=Date.parse(i);if(!Number.isNaN(e))return D(e-Date.now(),t)}}}}function De(e){return Math.random()*e}function Oe(e,t,n=ye,r=!1,i=!1){return De(Math.min(e*(r?2:i?1.5:1)*2**t,n))}async function ke(e,t){if(t?.aborted)throw new p(`Operation aborted`,`aborted`);await new Promise((n,r)=>{let i=()=>{clearTimeout(a),r(new p(`Operation aborted`,`aborted`))},a=setTimeout(()=>{t?.removeEventListener(`abort`,i),n()},e);t?.addEventListener(`abort`,i,{once:!0})})}function Ae(e,t,n,r){if(r?.aborted||e instanceof p&&!e.retryable)return!1;let i=n(e);return!(i!==void 0&&t.includes(i))}async function je(e){let{requestId:t,model:n,attempt:r,error:i,state:a,signal:o,providerName:s,maxRetries:c,baseDelayMs:l,middleware:u,middlewareTimeoutMs:d,logger:f,reportEvent:p,buildEventContext:m,extractStatus:h,normalizeError:g,emitEvent:_}=e,v=Ee(i),y=h(i),b=v??Oe(l,r,void 0,y===429,y!==void 0&&y>=500&&y<=599),x=v!==void 0;f.warn(`[VernLLM:${t}] recovery attempt ${r}/${c}, waiting ${Math.ceil(b)}ms`+(x?` (honoring Retry-After)`:``)),_({kind:`retry`,requestId:t,provider:s,model:n,attempt:r,maxRetries:c,delayMs:b,retryAfterHonored:x,error:g(i,o)},m(t,n,r,o,a),p,u,d,f),await ke(b,o)}async function Me(e){let{fn:t,maxRetries:n,signal:r,onAttempt:i,attempts:a,shouldRetryAttempt:o,recoverDelayForAttempt:s,normalizeError:c}=e,l,u;for(let e=0;e<=n;e++){u=void 0;try{return e>0&&await s(e,l),i?.(),await t(e,e=>{u=e})}catch(t){if(l=t,!(e<n&&o(t,r)))break;a?.push({index:e,error:c(t,r).toSnapshot(),request:u})}}throw l}var O=class{buckets;bucketStartMs=Date.now();bucketWidthMs;constructor(e){if(!Number.isFinite(e)||e<=0)throw RangeError(`RollingRatio: windowMs must be a finite number > 0, got ${e}`);this.bucketWidthMs=e/10,this.buckets=Array.from({length:10},()=>({total:0,failures:0}))}rotate(){let e=Date.now()-this.bucketStartMs,t=Math.floor(e/this.bucketWidthMs),n=Math.min(10,t);for(let e=0;e<n;e++)this.buckets.shift(),this.buckets.push({total:0,failures:0});t>0&&(this.bucketStartMs+=t*this.bucketWidthMs)}record(e){this.rotate();let t=this.buckets[9];t.total+=1,e&&(t.failures+=1)}getRatio(){this.rotate();let e=this.buckets.reduce((e,t)=>e+t.total,0),t=this.buckets.reduce((e,t)=>e+t.failures,0);return e===0?0:t/e}getCount(){return this.rotate(),this.buckets.reduce((e,t)=>e+t.total,0)}reset(){this.buckets=Array.from({length:10},()=>({total:0,failures:0})),this.bucketStartMs=Date.now()}};function Ne(e,t=`minCalls`){if(!Number.isInteger(e)||e<0)throw RangeError(`${t} must be a non-negative integer, got ${e}`)}function Pe(e,t){if(!Number.isFinite(e)||e<0||e>1)throw RangeError(`${t} must be a finite number between 0 and 1, got ${e}`)}function Fe(e){if(e===void 0)return;if(typeof e==`function`)return e;let{multiplier:t,maxMs:n=1/0}=e;return(e,r)=>De(Math.min(r*t**e,n))}var Ie=class{threshold;failuresByKey=new Map;constructor(e){this.threshold=e}onSuccess(e){this.failuresByKey.set(e,0)}onFailure(e){let t=(this.failuresByKey.get(e)??0)+1;return this.failuresByKey.set(e,t),t>=this.threshold}reset(e){this.failuresByKey.set(e,0)}forget(e){this.failuresByKey.delete(e)}},Le=class{windowMs;minCalls;failureRatio;ratiosByKey=new Map;constructor(e,t,n){this.windowMs=e,this.minCalls=t,this.failureRatio=n,new O(e),Ne(t),Pe(n,`failureRatio`)}ratioFor(e){let t=this.ratiosByKey.get(e);return t||(t=new O(this.windowMs),this.ratiosByKey.set(e,t)),t}onSuccess(e){this.ratioFor(e).record(!1)}onFailure(e){let t=this.ratioFor(e);return t.record(!0),t.getCount()>=this.minCalls&&t.getRatio()>=this.failureRatio}reset(e){this.ratiosByKey.delete(e)}forget(e){this.ratiosByKey.delete(e)}};function Re(e){return`onFailure`in e?e:e.kind===`consecutive`?new Ie(e.threshold):new Le(e.windowMs,e.minCalls,e.failureRatio)}function ze(){return{state:`closed`,consecutiveFailures:0,openedAt:0,trial:null,reopenCount:0,failuresByReason:new Map,cooldownMsForOpen:0}}function k(e){return e??``}const A=new WeakMap;function Be(e,t){return!t||A.get(t.state)===e.trial}function Ve(e,t){let n=t??`unknown`;e.failuresByReason.set(n,(e.failuresByReason.get(n)??0)+1)}var He=class{cooldownMs;onStateChange;isolateByModel;halfOpenProbes;halfOpenSuccessRatio;cooldownBackoff;tripping;sharedBucket=ze();bucketsByModel=new Map;constructor(e={}){let t=e.threshold??5;this.cooldownMs=e.cooldownMs??3e4,this.onStateChange=e.onStateChange,this.isolateByModel=e.isolateByModel??!1;let n=e.halfOpenProbes;this.halfOpenProbes=Number.isFinite(n)?Math.max(1,Math.floor(n)):1;let r=e.halfOpenSuccessRatio;this.halfOpenSuccessRatio=Number.isFinite(r)?Math.min(1,Math.max(0,r)):1,this.cooldownBackoff=Fe(e.cooldownBackoff),this.tripping=Re(e.tripping??{kind:`consecutive`,threshold:t})}assertClosed(e,t){let n=this.ensureBucketFor(e);if(n.state!==`closed`){if(n.state===`open`){let r=Date.now()-n.openedAt,i=n.cooldownMsForOpen;if(r<i)throw new p(`Circuit open, provider has failed ${n.consecutiveFailures} times in a row. Retry in ${Math.ceil((i-r)/1e3)}s.`,`circuit_open`,{code:`circuit_cooling_down`});n.trial={slotsRemaining:this.halfOpenProbes-1,successes:0,failures:0},t&&A.set(t.state,n.trial),this.transition(n,`half-open`,e,t);return}if(!n.trial||n.trial.slotsRemaining<=0)throw new p(`Circuit half-open. Every trial slot is already in flight. Try again shortly.`,`circuit_open`,{code:`circuit_trial_in_flight`});--n.trial.slotsRemaining,t&&A.set(t.state,n.trial)}}recordSuccess(e,t){let n=this.lookupBucket(e);if(n){if(n.state===`half-open`&&n.trial&&Be(n,t)){n.trial.successes+=1,this.settleTrialIfComplete(n,e,t);return}n.state!==`half-open`&&(n.consecutiveFailures=0,this.tripping.onSuccess(this.trippingKeyFor(e)),n.trial=null,n.reopenCount=0,n.failuresByReason.clear(),this.transition(n,`closed`,e,t),this.isolateByModel&&n.state===`closed`&&n.consecutiveFailures===0&&this.forgetModel(e))}}recordFailure(e,t,n){let r=this.ensureBucketFor(e);if(r.state===`half-open`&&r.trial&&Be(r,t)){r.trial.failures+=1,Ve(r,n),this.settleTrialIfComplete(r,e,t);return}r.state!==`half-open`&&(r.consecutiveFailures+=1,Ve(r,n),this.tripping.onFailure(this.trippingKeyFor(e))&&this.openBucket(r,e,t))}getState(e){return this.lookupBucket(e)?.state??`closed`}getFailureBreakdown(e){let t=this.lookupBucket(e);return t?Object.fromEntries(t.failuresByReason):{}}open(e,t){let n=this.ensureBucketFor(e);n.trial=null,this.openBucket(n,e,t)}close(e,t){let n=this.ensureBucketFor(e);n.consecutiveFailures=0,this.tripping.reset(this.trippingKeyFor(e)),n.trial=null,n.reopenCount=0,n.failuresByReason.clear(),this.transition(n,`closed`,e,t),this.isolateByModel&&n.state===`closed`&&n.consecutiveFailures===0&&this.forgetModel(e)}openBucket(e,t,n){e.openedAt=Date.now(),e.cooldownMsForOpen=this.computeCooldown(e),this.transition(e,`open`,t,n)}computeCooldown(e){if(!this.cooldownBackoff)return this.cooldownMs;let t=this.cooldownBackoff(e.reopenCount,this.cooldownMs);return Number.isNaN(t)?0:Math.max(0,t)}lookupBucket(e){return this.isolateByModel?this.bucketsByModel.get(k(e)):this.sharedBucket}trippingKeyFor(e){return this.isolateByModel?k(e):``}ensureBucketFor(e){if(!this.isolateByModel)return this.sharedBucket;let t=k(e),n=this.bucketsByModel.get(t);return n||(n=ze(),this.bucketsByModel.set(t,n)),n}forgetModel(e){this.bucketsByModel.delete(k(e)),this.tripping.forget?.(this.trippingKeyFor(e))}transition(e,t,n,r){if(t===e.state)return;let i=e.state;e.state=t,this.onStateChange?.(i,t,e.consecutiveFailures,n,r)}settleTrialIfComplete(e,t,n){let r=e.trial;if(!r||r.successes+r.failures<this.halfOpenProbes)return;let i=r.successes/this.halfOpenProbes;if(e.trial=null,i>=this.halfOpenSuccessRatio){e.consecutiveFailures=0,this.tripping.onSuccess(this.trippingKeyFor(t)),e.reopenCount=0,e.failuresByReason.clear(),this.transition(e,`closed`,t,n),this.isolateByModel&&e.state===`closed`&&this.forgetModel(t);return}e.reopenCount+=1,this.openBucket(e,t,n)}},Ue=class{options;ratio;constructor(e){this.options=e,this.ratio=new O(e.windowMs),Ne(e.minCalls),Pe(e.retryRatio,`retryRatio`)}assertAvailable(){if(this.ratio.getCount()>=this.options.minCalls&&this.ratio.getRatio()>=this.options.retryRatio)throw new p(`Retry budget exhausted`,`rate_limited`,{code:`retry_budget_exhausted`})}recordAttempt(e){this.ratio.record(e)}getSnapshot(){return{attempts:this.ratio.getCount(),retryRatio:this.ratio.getRatio()}}};function j(e){if(!e||typeof e!=`object`)return;let t=e;if(typeof t.status==`number`)return t.status;if(typeof t.statusCode==`number`)return t.statusCode;if(typeof t.$metadata?.httpStatusCode==`number`)return t.$metadata.httpStatusCode}const We=new Set([`ECONNREFUSED`,`ENOTFOUND`,`ECONNRESET`,`ETIMEDOUT`,`EAI_AGAIN`,`EPIPE`,`ECONNABORTED`,`EHOSTUNREACH`,`ENETUNREACH`]),Ge=new Set([`fetch failed`,`failed to fetch`,`load failed`,`networkerror when attempting to fetch resource.`]);function Ke(e){if(!e||typeof e!=`object`)return!1;let t=e;if(typeof t.code==`string`&&We.has(t.code)||typeof t.message==`string`&&Ge.has(t.message.toLowerCase()))return!0;if(t.cause&&typeof t.cause==`object`){let e=t.cause;if(typeof e.code==`string`&&We.has(e.code))return!0}return!1}function qe(e){try{return JSON.stringify(e,null,2)??String(e)}catch{try{return String(e)}catch{return`[unprintable error]`}}}function Je(e){if(e&&typeof e==`object`)try{let t=e;if(t.error!==void 0)return qe(t.error);if(typeof t.message==`string`)return t.message}catch{}return qe(e)}function Ye(e){switch(e){case 429:return`provider_rate_limited`;case 401:return`authentication`;case 403:return`authorization`;case 404:return`not_found`;case 413:return`payload_too_large`;default:return e>=500?`server_error`:void 0}}const Xe=/\(no body\)/i;function Ze(e){return Object.keys(e).length===0}function Qe(e){if(e&&typeof e==`object`){let{error:t,message:n}=e;if(t!=null){let e=typeof t==`string`&&t.trim().length===0,n=typeof t==`object`&&Ze(t);if(!e&&!n)return!1}if(typeof n==`string`){let e=n.trim();return e.length===0||Xe.test(e)}return!0}return!0}function M(e,t,n){if(t?.aborted)return new p(`LLM request aborted`,`aborted`,{attempts:n});if(e instanceof p)return e.code===void 0&&e.status!==void 0&&(e.code=Ye(e.status)),e.attempts===void 0&&n!==void 0&&(e.attempts=n),e;let r=j(e),i=Ee(e);if(r!==void 0){let t=Je(e),a=Ye(r),o=a===void 0;return new p(Qe(e)?o?`LLM request failed with status ${r} and no error detail from the provider. This usually means a field or value in the request isn't supported by the specific model (for example, a reasoning/thinking parameter the model doesn't accept), rather than a transport or auth problem.`:`LLM request failed with status ${r} and no error detail from the provider.`:`LLM request failed: ${t}`,`api`,{status:r,cause:e,retryAfterMs:i,code:a,attempts:n})}return Ke(e)?new p(`LLM request failed`,`network`,{cause:e,retryAfterMs:i,code:`connection_failed`,attempts:n}):new p(`LLM request failed`,`unknown`,{cause:e,retryAfterMs:i,attempts:n})}function N(e,t){return e.name??`[${t}]`}function $e(e,t,n){return t<=0?e():new Promise((r,i)=>{let a=setTimeout(()=>{i(new p(`middleware "${n}" timed out after ${t}ms`,`timeout`,{code:`middleware_timeout`}))},t);e().then(e=>{clearTimeout(a),r(e)},e=>{clearTimeout(a),i(e)})})}async function P(e,t,n,r,i){let{enabled:a}=e;if(a===void 0)return!0;if(typeof a==`boolean`)return a;let o=e.timeoutMs??r;try{return await $e(async()=>a(t),o,n)}catch(e){return i.error(`[VernLLM] middleware "${n}".enabled threw or timed out, treating as disabled`,{message:e instanceof Error?e.message:`unknown`}),!1}}const et=[`temperature`,`max_tokens`,`reasoning_effort`,`budget_tokens`,`tool_choice`,`messages`,`addMessages`,`tools`,`addTools`];function tt(e,t){let n=t,r=et.filter(e=>t[e]!==void 0);if(n.model!==void 0&&r.push(`model`),n.response_format!==void 0&&r.push(`response_format`),r.length===0)return{request:e,patchedFields:[]};let i={...e};return n.model!==void 0&&(i.model=n.model),n.response_format!==void 0&&(i.response_format=n.response_format),t.temperature!==void 0&&(i.temperature=t.temperature),t.max_tokens!==void 0&&(i.max_tokens=t.max_tokens),t.reasoning_effort!==void 0&&(i.reasoning_effort=t.reasoning_effort),t.budget_tokens!==void 0&&(i.budget_tokens=t.budget_tokens),t.tool_choice!==void 0&&(i.tool_choice=t.tool_choice),t.messages!==void 0&&(i.messages=t.messages),t.addMessages!==void 0&&t.addMessages.length>0&&(i.messages=[...i.messages,...t.addMessages]),t.tools!==void 0&&(i.tools=t.tools),t.addTools!==void 0&&t.addTools.length>0&&(i.tools=[...i.tools??[],...t.addTools]),{request:i,patchedFields:r}}function nt(e,t){if(!e.tools)return;let n=new Set,r=new Set;for(let t of e.tools)n.has(t.function.name)&&r.add(t.function.name),n.add(t.function.name);if(r.size>0)throw new p(`middleware "${t}" added tool(s) with duplicate name(s): [${[...r].join(`, `)}]. Tool names must be unique.`,`invalid_params`,{code:`duplicate_tool_names`,issues:{names:[...r]}})}function rt(e,t,n){if(t.model!==e.model)throw new p(`middleware "${n}" changed \`model\` via transform, which isn't supported. Configure the target's model instead.`,`invalid_params`);if(JSON.stringify(t.response_format)!==JSON.stringify(e.response_format))throw new p(`middleware "${n}" changed \`response_format\` via transform, which isn't supported.`,`invalid_params`)}async function it(e){let{request:t,requestId:n,attempt:r,signal:i,state:a,middleware:o,middlewareTimeoutMs:s,logger:c,reportEvent:l,buildContext:u}=e;if(o.length===0)return t;let d=t,f=t,p=o.map((e,t)=>({middlewareEntry:e,index:t})).sort((e,t)=>(e.middlewareEntry.priority??0)-(t.middlewareEntry.priority??0));for(let{middlewareEntry:e,index:t}of p){let d=N(e,t),p=u(r,i,a);if(!await P(e,p,d,s,c)){e.enabled!==void 0&&F({kind:`middleware`,requestId:n,middleware:d,hook:`enabled_skip`},p,l,o,s,c);continue}if(!e.transform)continue;let m=await at(e,structuredClone(f),p,d,s),{request:h,patchedFields:g}=tt(f,m);g.length>0&&((m.tools!==void 0||m.addTools?.length)&&nt(h,d),F({kind:`middleware`,requestId:n,middleware:d,hook:`transform`,patchedFields:g},p,l,o,s,c)),f=h}return rt(d,f,`chain`),f}async function at(e,t,n,r,i){if(!e.transform)return{};let a=e.timeoutMs??i;try{return await $e(async()=>e.transform(t,n),a,r)}catch(e){throw ot(e,r,n.signal)}}function ot(e,t,n){let r=M(e,n);return r.type===`unknown`?new p(`middleware "${t}" threw: ${r.message}`,`invalid_params`,{code:`middleware_threw`,cause:e}):r}function F(e,t,n,r,i,a){n(e),r.length!==0&&st(r,e,t,i,a)}async function st(e,t,n,r,i){for(let a=0;a<e.length;a++){let o=e[a];if(!o.onEvent)continue;let s=N(o,a);if(await P(o,n,s,r,i))try{Promise.resolve(o.onEvent(t,{...n,own:{}})).catch(e=>{i.error(`[VernLLM] middleware "${s}".onEvent failed`,{message:e instanceof Error?e.message:`unknown`})})}catch(e){i.error(`[VernLLM] middleware "${s}".onEvent failed`,{message:e instanceof Error?e.message:`unknown`})}}}function I(e,t){return n=>{if(e)try{e(n)}catch(e){t.error(`[VernLLM] onEvent failed`,{message:e instanceof Error?e.message:`unknown`})}}}function L(e,t,n){let r=e[t];if(!r)throw RangeError(`${n}: no target at index ${t} (chain has ${e.length} target${e.length===1?``:`s`})`);return r}function R(e,t,n,r){t!==void 0&&!e&&r.warn(`[VernLLM] ${n}: \`model: '${t}'\` has no effect here. This target's circuitBreaker doesn't have isolateByModel on, so it only tracks one shared circuit regardless of \`model\`. Omit \`model\`, or set \`circuitBreaker.isolateByModel: true\` on this target if per-model tracking is what you want.`)}function ct(e,t,n,r,i,a,o,s,c){if(!e)return;let l=typeof e==`object`?e:void 0,u=l?.onStateChange,d=I(r,i);return new He({...l,onStateChange:(e,r,l,f,p)=>{let m={kind:`circuit_state`,provider:t,model:f??n,from:e,to:r,consecutiveFailures:l};if(p?F(m,{stage:`attempt`,requestId:p.requestId,requestedProvider:t,requestedModel:f??n,isFallbackAttempt:s,attempt:p.attempt??1,capabilities:{supportsJsonObjectMode:c},signal:p.signal,state:p.state,own:{}},d,a,o,i):d(m),u)try{u(e,r,l,f,p)}catch(e){i.error(`[VernLLM] circuitBreaker.onStateChange failed`,{message:e instanceof Error?e.message:`unknown`})}}})}function lt(e){let t=/^(?:(\d+(?:\.\d+)?)h)?(?:(\d+(?:\.\d+)?)m)?(?:(\d+(?:\.\d+)?)s)?(?:(\d+)ms)?$/.exec(e.trim());if(!t||t[0]===``)return;let[,n,r,i,a]=t,o=(n?parseFloat(n)*36e5:0)+(r?parseFloat(r)*6e4:0)+(i?parseFloat(i)*1e3:0)+(a?parseInt(a,10):0);return Number.isFinite(o)?o:void 0}function z(e,t){let n=e.get(t);if(n===null)return;let r=Number(n);return Number.isFinite(r)?r:void 0}function B(e){let t=e.get(`x-ratelimit-reset-requests`);return{limitRequests:z(e,`x-ratelimit-limit-requests`),remainingRequests:z(e,`x-ratelimit-remaining-requests`),resetAfterMs:t===null?void 0:lt(t)}}function ut(e){let t=e.get(`anthropic-ratelimit-requests-reset`),n=t===null?NaN:Date.parse(t);return{limitRequests:z(e,`anthropic-ratelimit-requests-limit`),remainingRequests:z(e,`anthropic-ratelimit-requests-remaining`),resetAfterMs:Number.isFinite(n)?Math.max(0,n-Date.now()):void 0}}const dt=Symbol(`vernLLMProviderRateLimitHint`);function ft(e,t){t&&Object.defineProperty(e,dt,{value:t,enumerable:!1,configurable:!0})}function pt(e){if(!(!e||typeof e!=`object`))return e[dt]}function mt(e){let{breaker:t,requestId:n,model:r,providerName:i,isFallback:a,supportsJsonObjectMode:o}=e;function s(e,t,s){return{stage:`attempt`,requestId:n,requestedProvider:i,requestedModel:r,isFallbackAttempt:a,attempt:e+1,capabilities:{supportsJsonObjectMode:o},signal:t,state:s,own:{}}}function c(e,t,r){return{requestId:n,state:r,signal:t,attempt:e+1}}return{buildAttemptContext:s,buildCallContext:c,recordSuccess(e,n,i){t?.recordSuccess(r,c(e,n,i))},recordFailure(e,n,i,a){t?.recordFailure(r,c(e,n,i),a)}}}function ht(e){return e.map(e=>({type:`function`,function:{name:e.name,description:e.description,parameters:e.parameters}}))}function gt(e){return e.map(e=>({id:e.id,type:`function`,function:{name:e.name,arguments:JSON.stringify(e.arguments??{})}}))}function _t(e){return e.map(e=>{let t;try{t=e.function.arguments.trim()?JSON.parse(e.function.arguments):{}}catch{throw new p(`Invalid JSON arguments for tool call "${e.function.name}"`,`parse`,{code:`tool_arguments_parse_failed`})}return{id:e.id,name:e.function.name,arguments:t}})}function vt(e){return typeof e==`string`?e:JSON.stringify(e)}function yt(e,t){if(e&&e.length===0)throw new p("`tools` was an empty array. This is almost always a bug (e.g. a filtered tool list that ended up empty). An empty `tools` array still switches on tool-call mode (response shape, jsonMode default, wire format) with nothing for the model to call. Omit `tools` entirely for a normal call, or make sure the array is non-empty.",`invalid_params`);if(e){let t=new Set,n=new Set;for(let r of e)t.has(r.name)&&n.add(r.name),t.add(r.name);if(n.size)throw new p(`\`tools\` has duplicate name(s): [${[...n].join(`, `)}]. Tool names must be unique.`,`invalid_params`,{code:`duplicate_tool_names`,issues:{names:[...n]}})}if(t&&!e)throw new p("`toolChoice` was set without `tools`. There is nothing for it to choose between. Set `tools`, or remove `toolChoice`.",`invalid_params`);if(e&&typeof t==`object`&&!e.some(e=>e.name===t.name))throw new p(`toolChoice names "${t.name}", which is not in \`tools\` ([${e.map(e=>e.name).join(`, `)}]).`,`invalid_params`,{code:`unknown_tool_choice`,issues:{requested:t.name,available:e.map(e=>e.name)}})}function bt(e){let{jsonModeExplicit:t,hasTools:n,jsonSchema:r,hasSchema:i,supportsJsonObjectMode:a}=e,o=t??!n;if(!a&&!r&&t===!0)throw new p('jsonMode: true was set explicitly, but this client does not support `response_format: "json_object"` (see LLMClient.supportsJsonObjectMode). Neither Anthropic nor Bedrock has a field that mechanically guarantees JSON output for this mode. Use `jsonSchema` instead, which maps to a real constraint on both.',`invalid_params`);if(!a&&!r&&t===void 0&&i)throw new p('`schema` was provided, which requires JSON output to validate against, but this client does not support `response_format: "json_object"` (see LLMClient.supportsJsonObjectMode) and no `jsonSchema` was set. Neither Anthropic nor Bedrock has a field that mechanically guarantees JSON output without one. Use `jsonSchema` instead, which maps to a real constraint on both and still runs `schema` against its parsed result.',`invalid_params`);let s=(!a&&!r&&t===void 0?!1:o)||!!r;if(i&&!s)throw new p(`schema was provided but jsonMode: false disables JSON parsing, so nothing would validate it. Remove jsonMode: false, set jsonSchema, or remove schema.`,`invalid_params`);return s}function xt(e,t){return e?{type:`json_schema`,json_schema:{name:e.name,schema:e.schema,strict:e.strict??!0,description:e.description}}:t?{type:`json_object`}:void 0}function St(e){return!e||e===`auto`?`auto`:e===`none`||e===`required`?e:{type:`function`,function:{name:e.name}}}function Ct(e){return e.role===`tool`?(e.toolResults??[]).map(e=>({role:`tool`,tool_call_id:e.toolCallId,content:typeof e.content==`string`?e.content:JSON.stringify(e.content??null),...e.isError?{is_error:!0}:{}})):e.role===`assistant`&&e.toolCalls?.length?[{role:`assistant`,...e.content===void 0?{}:{content:vt(e.content)},tool_calls:gt(e.toolCalls)}]:e.role===`assistant`?[{role:`assistant`,content:vt(e.content===void 0?``:e.content)}]:[{role:e.role,content:e.content??``}]}function wt(e){let t;for(let[n,r]of e.entries()){if(r.role===`tool`){if(t?.role!==`assistant`||!t.toolCalls?.length)throw new p(`history[${n}] is a "tool" turn, but must immediately follow an "assistant" turn that requested tools`,`invalid_params`);if(!r.toolResults?.length)throw new p(`history[${n}] is a "tool" turn but has no toolResults`,`invalid_params`);let e=new Set(t.toolCalls.map(e=>e.id)),i=r.toolResults.map(e=>e.toolCallId),a=i.filter(t=>!e.has(t));if(a.length)throw new p(`history[${n}].toolResults references unknown toolCallId(s) [${a.join(`, `)}]`,`invalid_params`,{code:`unknown_tool_result_ids`,issues:{historyIndex:n,ids:a}});let o=new Set,s=new Set;for(let e of i)o.has(e)&&s.add(e),o.add(e);if(s.size)throw new p(`history[${n}].toolResults has duplicate toolCallId(s) [${[...s].join(`, `)}]`,`invalid_params`,{code:`duplicate_tool_result_ids`,issues:{historyIndex:n,ids:[...s]}});let c=[...e].filter(e=>!i.includes(e));if(c.length)throw new p(`history[${n}] is missing toolResults for toolCallId(s) [${c.join(`, `)}]`,`invalid_params`,{code:`missing_tool_results`,issues:{historyIndex:n,ids:c}})}else{if(r.role===t?.role)throw new p(`history must alternate user/assistant turns: consecutive "${r.role}" turns at history[${n-1}] and history[${n}]`,`invalid_params`);if(t?.role===`assistant`&&t.toolCalls?.length)throw new p(`history[${n}] follows an assistant tool request without tool results`,`invalid_params`)}t=r}if(t?.role===`assistant`&&t.toolCalls?.length)throw new p(`The last entry in history is an assistant tool request without tool results`,`invalid_params`);if(t?.role===`user`)throw new p(`The last entry in history is a "user" turn, which would collide with the current userContent turn.`,`invalid_params`)}var Tt=class{model;defaultMaxTokens;defaultTemperature;defaultReasoningEffort;defaultBudgetTokens;supportsJsonObjectMode;constructor(e){this.model=e.model,this.defaultMaxTokens=e.defaultMaxTokens,this.defaultTemperature=e.defaultTemperature,this.defaultReasoningEffort=e.defaultReasoningEffort,this.defaultBudgetTokens=e.defaultBudgetTokens,this.supportsJsonObjectMode=e.supportsJsonObjectMode}build(e){let{systemPrompt:t,userContent:n,history:r=[],maxTokens:i=this.defaultMaxTokens,model:a=this.model,jsonSchema:o,tools:s,toolChoice:c}=e,l=e.temperature===void 0?this.defaultTemperature:e.temperature,u=e.reasoningEffort===void 0?this.defaultReasoningEffort:e.reasoningEffort,d=e.budgetTokens===void 0?this.defaultBudgetTokens:e.budgetTokens;yt(s,c);let f=bt({jsonModeExplicit:e.jsonMode,hasTools:!!s,jsonSchema:o,hasSchema:!!e.schema,supportsJsonObjectMode:this.supportsJsonObjectMode}),p=xt(o,f);return wt(r),{useJson:f,model:a,request:{model:a,...l===null?{}:{temperature:l},max_tokens:i,...p?{response_format:p}:{},...u?{reasoning_effort:u}:{},...d==null?{}:{budget_tokens:d},...s?{tools:ht(s)}:{},...s?{tool_choice:St(c)}:{},messages:[...t?[{role:`system`,content:t}]:[],...r.flatMap(e=>Ct(e)),{role:`user`,content:n}]}}}};function Et(e,t,n){let r;try{r=n(e)}catch{throw new p(`Invalid JSON response`,`parse`)}if(r==null)throw new p(`Invalid JSON response`,`parse`);if(!t)return r;let i=t.safeParse(r);if(!i.success)throw new p(`Schema validation failed`,`validation`,{issues:i.error});return i.data}function Dt(e,t){let n=new Map(t.map(e=>[e.name,e])),r=new Set,i=[];for(let t of e)r.has(t.id)&&i.push({name:t.name,toolCallId:t.id,code:`duplicate_tool_call_id`}),r.add(t.id),n.has(t.name)||i.push({name:t.name,toolCallId:t.id,code:`unknown_tool`});if(i.length>0){let e=i.find(e=>e.code===`unknown_tool`),t=e?`Model requested tool "${e.name}", which was not in the tools offered ([${[...n.keys()].join(`, `)}]).`:`Duplicate tool call id "${i[0].toolCallId}" in the model's response.`;throw new p(i.length>1?`${t} (${i.length} tool call issues total, see error.issues.)`:t,`validation`,{code:e?`unknown_tool`:`duplicate_tool_call_id`,issues:i})}for(let t of e){let e=n.get(t.name);if(!e?.argumentsSchema)continue;let r=e.argumentsSchema.safeParse(t.arguments);if(!r.success)throw new p(`Arguments for tool call "${t.name}" failed validation`,`validation`,{issues:r.error})}}function Ot(e){let{rawContent:t,wireToolCalls:n,params:r,useJson:i,parseJson:a,requestId:o,logger:s,redactText:c}=e,l=t?.trim();if(!l&&!n?.length)throw new p(`Empty LLM response`,`api`,{code:`empty_response`});if(s.debug(`[VernLLM:${o}] output:\n`+c(l??`[${n?.length??0} tool call(s)]`).slice(0,800)),n?.length){if(!r.tools)throw new p("Provider returned tool_calls but no `tools` were sent with this call.",`validation`,{code:`unexpected_tool_calls`});if(r.toolChoice===`none`)throw new p(`Provider returned tool_calls despite toolChoice: 'none'.`,`validation`,{code:`tool_choice_none_violated`});let e=_t(n);return Dt(e,r.tools),{type:`tool_calls`,toolCalls:e,...l?{content:l}:{}}}let u=l??``;if(!i)return r.tools?{type:`content`,content:u}:u;let d=Et(u,r.schema,a);return r.tools?{type:`content`,content:d}:d}function kt(e,t,n,r,i,a,o,s,c){let{gateway:l,usageReporter:u,logger:d,redactText:f,parseJson:m,detectSoftFailure:h}=c;try{let g=Ot({rawContent:e,wireToolCalls:t,params:n,useJson:r,parseJson:m,requestId:a,logger:d,redactText:f}),_=At(h,g,{requestId:a,model:c.model,providerName:c.providerName,isFallback:c.isFallback,attempt:o+1},d);if(_!==void 0)throw new p(`Soft failure detected`,`api`,{code:_});return l.recordSuccess(o,n.signal,s),u.reportSuccess(i),g}catch(e){let t=M(e,n.signal);throw i&&t.type!==`aborted`&&u.reportFailure(i,t,o),t}}function At(e,t,n,r){if(e)try{return e(t,n)}catch(e){r.warn(`detectSoftFailure threw and was ignored, treated as no soft failure: ${e instanceof Error?e.message:String(e)}`);return}}function jt(e){let{capacity:t,logger:n,label:r}=e,i=[],a=[],o=!1,s=!1,c,l=!1;function u(e){let o=a.shift();if(o){o.resolve({done:!1,value:e});return}i.push(e),i.length>t*2&&(l||(l=!0,n.warn(`[VernLLM] ${r} buffer exceeded cap (${t}), evicting ${i.length-t} oldest item(s); buffered=${i.length}. The iterable was never read (or fell far behind) for this channel.`)),i.splice(0,i.length-t))}function d(){o=!0;for(let e of a.splice(0))e.resolve({done:!0,value:void 0})}function f(e){o=!0,s=!0,c=e;for(let t of a.splice(0))t.reject(e)}return{push:u,finish:d,fail:f,iterable:{[Symbol.asyncIterator](){return{next(){return i.length?Promise.resolve({done:!1,value:i.shift()}):o?s?Promise.reject(c):Promise.resolve({done:!0,value:void 0}):new Promise((e,t)=>{a.push({resolve:e,reject:t})})}}}}}}const Mt=1e4,Nt=1e6;function Pt(){let e=new Map;function t(t){let n=e.get(t.index);if(!n){if(e.size>=Mt)throw new p(`Stream contained more than ${Mt} distinct tool calls`,`validation`);n={args:``}}let r=t.argumentsDelta??``;if(n.args.length+r.length>Nt)throw new p(`Tool call arguments exceeded the ${Nt}-character stream limit`,`validation`);return n.id??=t.id,n.name??=t.name,n.args+=r,e.set(t.index,n),{type:`tool_call_delta`,index:t.index,id:t.id,name:t.name,argsDelta:t.argumentsDelta,complete:t.complete}}function n(){if(e.size)return[...e.entries()].sort(([e],[t])=>e-t).map(([,e])=>({id:e.id??``,type:`function`,function:{name:e.name??``,arguments:e.args}}))}return{apply:t,toWireToolCalls:n}}async function Ft(e,t){try{await e.return?.()}catch{}t.abort()}function It(e,t,n){let{requestId:r,model:i,providerName:a,isFallback:o,chunkIdleTimeoutMs:s,streamController:c,logger:l,signal:u,onRateLimitHint:d}=n,f,p,m=new Promise((e,t)=>{f=e,p=t});m.catch(()=>{});let h=jt({capacity:1e4,logger:l,label:`stream chunk`}),{push:g,finish:_,fail:v}=h,y=h.iterable,b=Pt(),x=``,S;return(async()=>{try{let n=t;for(;!n.done;){let t=n.value;t.type===`ping`||(t.type===`rate_limit_hint`?d?.(t.hint):t.type===`text-delta`?(x+=t.delta,g({type:`text-delta`,delta:t.delta})):t.type===`tool_call_delta`?g(b.apply(t)):t.type===`usage`&&(S=ue(t.usage,{requestId:r,model:i,providerName:a,isFallback:o}),g({type:`usage`,usage:S}))),n=await Ce(()=>e.next(),s,()=>c.abort(),l)}}catch(t){await Ft(e,c);let r=M(t,u);try{n.onStreamFailure(r,S)}catch{}v(r),p(r);return}_();try{n.onStreamSuccess(S)}catch{}try{let e=b.toWireToolCalls(),t=n.finalize(x,e,S);f(t)}catch(e){p(e)}})(),{chunks:y,finalResult:m}}function Lt(e){let{providerName:t,isFallback:n,maxRetries:r,onUsage:i,onUsageFailure:a,logger:o}=e;function s(e,r,i){if(e.usage)return ue(e.usage,{requestId:r,model:i,providerName:t,isFallback:n})}function c(e){if(e)return e.totalTokens||e.promptTokens+e.completionTokens}function l(e){if(!(!e||!i))try{i(e)}catch(e){o.error(`[VernLLM] onUsage failed`,{message:e instanceof Error?e.message:`unknown`})}}function u(e,t,n,i=!1){let s=e.totalTokens||e.promptTokens+e.completionTokens,c=i?`mid-stream failure (terminal, no further attempts)`:`attempt ${n+1}/${r+1}`;if(o.warn(`[VernLLM:${e.requestId}] usage failure, ${c}: type=${t.type} tokens=${s}`),a)try{a(e,t)}catch(e){o.error(`[VernLLM] onUsageFailure failed`,{message:e instanceof Error?e.message:`unknown`})}}return{extract:s,actualTokensFor:c,reportSuccess:l,reportFailure:u}}async function Rt(e,t,n,r){if(!e)return{};let i=await e.acquire(e.estimate(t),n);return i.waitedMs>0&&r(i.waitedMs,i.reason??`rpm`),{release:i.release}}async function zt(e){let{params:t,requestId:n,attempt:r,onRequest:i,middlewareState:a,gateway:o,requestBuilder:s,providerName:c,limiter:l,middleware:u,middlewareTimeoutMs:d,logger:p,reportEvent:m}=e,h=a??T(),g=s.build(t),{useJson:_,model:v}=g,y=await it({request:g.request,requestId:n,attempt:r,signal:t.signal,state:h,middleware:u,middlewareTimeoutMs:d,logger:p,reportEvent:m,buildContext:(e,t,n)=>o.buildAttemptContext(e,t,n)});i?.(f(c,v,y,void 0,Date.now()));let{release:b}=await Rt(l,y,t.signal,(e,i)=>{F({kind:`rate_limited`,requestId:n,provider:c,model:v,waitedMs:e,reason:i},o.buildAttemptContext(r,t.signal,h),m,u,d,p)});return{request:y,model:v,useJson:_,state:h,release:b}}async function Bt(e){let{fn:t,requestId:n,model:r,providerName:i,isFallback:a,supportsJsonObjectMode:o,breaker:s,budget:c,maxRetries:l,baseDelayMs:u,nonRetryableStatus:d,signal:f,onAttempt:p,middleware:m,middlewareTimeoutMs:h,logger:g,reportEvent:_,logLabel:v,redactText:y,countsTowardBreaker:b}=e,x=e.state??T(),S=[],C=mt({breaker:s,requestId:n,model:r,providerName:i,isFallback:a,supportsJsonObjectMode:o}),ee;try{return await Me({fn:(e,n)=>(c?.recordAttempt(e>0),t(e,n,x,C)),maxRetries:l,signal:f,onAttempt:p,attempts:S,shouldRetryAttempt:(e,t)=>{if(!Ae(e,d,j,t))return!1;if(c)try{c.assertAvailable()}catch(e){return ee=e,!1}return!0},recoverDelayForAttempt:(e,t)=>je({requestId:n,model:r,attempt:e,error:t,state:x,signal:f,providerName:i,maxRetries:l,baseDelayMs:u,middleware:m,middlewareTimeoutMs:h,logger:g,reportEvent:_,buildEventContext:(e,t,n,r,i)=>C.buildAttemptContext(n,r,i),extractStatus:j,normalizeError:M,emitEvent:F}),normalizeError:M})}catch(e){let t=M(ee??e,f,S.length>0?S:void 0);throw b(t)&&C.recordFailure(S.length,f,x,t.code),g.debug(`[VernLLM:${n}] ${v}:\n${y(Je(e))}`),t}}function Vt(e){try{return JSON.parse(e)}catch{return}}var Ht=class{providerName;client;model;maxRetries;timeoutMs;chunkIdleTimeoutMs;baseDelayMs;nonRetryableStatus;parseJson;logger;redact;usageReporter;reportEvent;breaker;budget;limiter;isFallback;requestBuilder;middleware;middlewareTimeoutMs;supportsJsonObjectMode;detectSoftFailure;constructor(e,t,n,r){this.providerName=e,this.client=t,this.model=n,this.maxRetries=r.maxRetries,this.timeoutMs=r.timeoutMs,this.chunkIdleTimeoutMs=r.chunkIdleTimeoutMs,this.baseDelayMs=r.baseDelayMs,this.nonRetryableStatus=r.nonRetryableStatus,this.parseJson=r.parseJson??Vt,this.logger=r.logger,this.redact=r.redact,this.reportEvent=I(r.onEvent,this.logger),this.breaker=r.breaker,this.budget=r.budget,this.limiter=r.limiter,this.isFallback=r.isFallback??!1,this.middleware=r.middleware??[],this.middlewareTimeoutMs=r.middlewareTimeoutMs??5e3,this.supportsJsonObjectMode=t.supportsJsonObjectMode??!0,this.detectSoftFailure=r.detectSoftFailure,this.usageReporter=Lt({providerName:this.providerName,isFallback:this.isFallback,maxRetries:this.maxRetries,onUsage:r.onUsage,onUsageFailure:r.onUsageFailure,logger:this.logger}),this.requestBuilder=new Tt({model:n,defaultMaxTokens:r.defaultMaxTokens,defaultTemperature:r.defaultTemperature,defaultReasoningEffort:r.defaultReasoningEffort,defaultBudgetTokens:r.defaultBudgetTokens,supportsJsonObjectMode:this.supportsJsonObjectMode})}previewRequest(e){let{model:t,request:n}=this.requestBuilder.build(e);return{model:t,request:n}}get jsonObjectModeSupported(){return this.supportsJsonObjectMode}getCircuitState(e){return this.breaker?.getState(e)}getFailureBreakdown(e){return this.breaker?.getFailureBreakdown(e)}getRetryBudgetState(){return this.budget?.getSnapshot()}get isolateByModel(){return this.breaker?.isolateByModel??!1}openCircuit(e,t){this.breaker?.open(e,t)}closeCircuit(e,t){this.breaker?.close(e,t)}assertBreakerClosed(e,t){this.breaker?.assertClosed(e??this.model,t)}async run(e,t,n,r){return Bt({fn:(n,r,i,a)=>this.executeCall(e,t,n,r,i,a),requestId:t,model:e.model??this.model,providerName:this.providerName,isFallback:this.isFallback,supportsJsonObjectMode:this.supportsJsonObjectMode,breaker:this.breaker,budget:this.budget,maxRetries:this.maxRetries,baseDelayMs:this.baseDelayMs,nonRetryableStatus:this.nonRetryableStatus,signal:e.signal,onAttempt:n,state:r,middleware:this.middleware,middlewareTimeoutMs:this.middlewareTimeoutMs,logger:this.logger,reportEvent:this.reportEvent,logLabel:`error`,redactText:e=>this.redactText(e),countsTowardBreaker:e=>this.countsTowardBreaker(e)})}async runStream(e,t,n,r){return Bt({fn:(n,r,i,a)=>this.executeStreamCall(e,t,n,r,i,a),requestId:t,model:e.model??this.model,providerName:this.providerName,isFallback:this.isFallback,supportsJsonObjectMode:this.supportsJsonObjectMode,breaker:this.breaker,budget:this.budget,maxRetries:this.maxRetries,baseDelayMs:this.baseDelayMs,nonRetryableStatus:this.nonRetryableStatus,signal:e.signal,onAttempt:n,state:r,middleware:this.middleware,middlewareTimeoutMs:this.middlewareTimeoutMs,logger:this.logger,reportEvent:this.reportEvent,logLabel:`stream-open error`,redactText:e=>this.redactText(e),countsTowardBreaker:e=>this.countsTowardBreaker(e)})}async executeCall(e,t,n,r,i,a){let{request:o,model:s,useJson:c,state:l,release:u}=await zt({params:e,requestId:t,attempt:n,onRequest:r,middlewareState:i,gateway:a,requestBuilder:this.requestBuilder,providerName:this.providerName,limiter:this.limiter,middleware:this.middleware,middlewareTimeoutMs:this.middlewareTimeoutMs,logger:this.logger,reportEvent:this.reportEvent}),d=u;try{let r;try{r=await Se(e=>this.client.chat.completions.create(o,{signal:e}),this.timeoutMs,e.signal)}catch(e){throw this.reactToRateLimitError(e),e}this.limiter?.reactToRateLimitHint(pt(r));let i=this.usageReporter.extract(r,t,s),u=this.usageReporter.actualTokensFor(i),f=r.choices?.[0]?.message?.content,p=r.choices?.[0]?.message?.tool_calls,m;try{m=kt(f,p,e,c,i,t,n,l,{gateway:a,usageReporter:this.usageReporter,logger:this.logger,redactText:e=>this.redactText(e),parseJson:this.parseJson,detectSoftFailure:this.detectSoftFailure,providerName:this.providerName,isFallback:this.isFallback,model:s})}catch(e){throw d?.(u),d=void 0,e}return d?.(u,!0),d=void 0,m}finally{d?.()}}redactText(e){return this.redact?this.redact(e):e}reactToRateLimitError(e){this.limiter&&j(e)===429&&this.limiter.signalRateLimit()}async executeStreamCall(e,t,n,r,i,a){let o=this.client.chat.completions;if(!o.createStream)throw new p(`stream: true requires a client/adapter with createStream`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`createStream`}});let s=o.createStream.bind(o),{request:c,model:l,useJson:u,state:d,release:f}=await zt({params:e,requestId:t,attempt:n,onRequest:r,middlewareState:i,gateway:a,requestBuilder:this.requestBuilder,providerName:this.providerName,limiter:this.limiter,middleware:this.middleware,middlewareTimeoutMs:this.middlewareTimeoutMs,logger:this.logger,reportEvent:this.reportEvent}),m=f,h=new AbortController,g=e.signal?AbortSignal.any([e.signal,h.signal]):h.signal;try{let{iterator:r,first:i}=await(async()=>{try{return await Se(async e=>{let t=s(c,{signal:e})[Symbol.asyncIterator]();return{iterator:t,first:await t.next()}},this.timeoutMs,g)}catch(e){throw this.reactToRateLimitError(e),e}})();if(i.done)throw new p(`Empty LLM response`,`api`);let o=m,f=It(r,i,{requestId:t,model:l,providerName:this.providerName,isFallback:this.isFallback,chunkIdleTimeoutMs:e.chunkIdleTimeoutMs??this.chunkIdleTimeoutMs,streamController:h,logger:this.logger,signal:e.signal,onRateLimitHint:e=>{this.limiter?.reactToRateLimitHint(e)},onStreamSuccess:e=>{},onStreamFailure:(t,r)=>{this.reactToRateLimitError(t),t.type===`timeout`&&a.recordFailure(n,e.signal,d,t.code),r&&t.type!==`aborted`&&this.usageReporter.reportFailure(r,t,n,!0),o?.(this.usageReporter.actualTokensFor(r))},finalize:(r,i,s)=>{let c=this.usageReporter.actualTokensFor(s);try{let f=kt(r,i,e,u,s,t,n,d,{gateway:a,usageReporter:this.usageReporter,logger:this.logger,redactText:e=>this.redactText(e),parseJson:this.parseJson,detectSoftFailure:this.detectSoftFailure,providerName:this.providerName,isFallback:this.isFallback,model:l});return o?.(c,!0),f}catch(t){throw o?.(c),t instanceof p&&this.countsTowardBreaker(t)&&a.recordFailure(n,e.signal,d,t.code),t}}});return m=void 0,f}finally{m?.()}}countsTowardBreaker(e){return e.countsTowardBreaker}};async function Ut(e,t,n,r,i,a=!1){let o=[];for(let s=0;s<e.executors.length;s++){let c=e.executors[s],l=Date.now(),u=0;try{return s===0&&a||c.assertBreakerClosed(t.model,{requestId:n,state:r,signal:t.signal}),{result:await i(c,()=>{u+=1}),executor:c,index:s,attemptCount:u}}catch(i){let a=M(i,t.signal);o.push({index:s-1,provider:c.providerName,model:t.model??c.model,error:a.toSnapshot()});let d=s===e.executors.length-1,f=e.fallbackOn(a,{isLastTarget:d});if((d?`stop`:f)===`stop`)throw o.length>1?new w(o):a;let p=e.executors[s+1],m=t.model??c.model,h={stage:`attempt`,requestId:n,requestedProvider:c.providerName,requestedModel:m,isFallbackAttempt:s>0,attempt:u||1,capabilities:{supportsJsonObjectMode:c.jsonObjectModeSupported},signal:t.signal,state:r,own:{}};F({kind:`fallback`,requestId:n,from:c.providerName,to:p.providerName,fromIndex:s-1,toIndex:s,error:a,elapsedMs:Date.now()-l},h,e.reportEvent,e.middleware,e.middlewareTimeoutMs,e.logger)}}throw new p(`No provider targets configured`,`invalid_params`)}function Wt(e,t){let n={provider:e.executor.providerName,model:t.model??e.executor.model,fallbackIndex:e.index-1,usedFallback:e.index>0,attempts:e.attemptCount};return t.meta&&(t.meta.current=n),{value:e.result,meta:n}}async function Gt(e,t,n,r,i){return Wt(await Ut(e,t,n,i,(e,r)=>e.run(t,n,r,i),r),t)}async function Kt(e,t,n,r,i){return Wt(await Ut(e,t,n,i,(e,r)=>e.runStream(t,n,r,i),r),t)}function qt(e){let t=!1,n,r;function i(){return r||(t=!0,r=e().then(e=>(n=e,e)),r.catch(()=>{}),r)}return{call:i,wasCalled:()=>t,resolvedValue:()=>n}}async function V(e,t,n,r,i,a=!1){if(e.middleware.length===0||a)return i();let o=e.primaryExecutor,{model:s,request:c}=o.previewRequest(t),l=i;for(let i=e.middleware.length-1;i>=0;i--){let a=e.middleware[i],u=N(a,i),d=l;l=async()=>{let i={stage:`pre-dispatch`,requestId:n,primaryProvider:o.providerName,primaryModel:s,capabilities:{supportsJsonObjectMode:o.jsonObjectModeSupported},signal:t.signal,state:r,own:{}};if(!await P(a,i,u,e.middlewareTimeoutMs,e.logger))return a.enabled!==void 0&&F({kind:`middleware`,requestId:n,middleware:u,hook:`enabled_skip`},i,e.reportEvent,e.middleware,e.middlewareTimeoutMs,e.logger),d();if(!a.wrap)return d();let l=qt(d);try{let t=await a.wrap(c,l.call,i);return l.wasCalled()||F({kind:`middleware`,requestId:n,middleware:u,hook:`wrap_short_circuit`},i,e.reportEvent,e.middleware,e.middlewareTimeoutMs,e.logger),t}catch(t){let n=l.resolvedValue();if(n!==void 0)return e.logger.error(`[VernLLM] middleware "${u}".wrap threw after next() resolved; keeping the original result`,{message:t instanceof Error?t.message:`unknown`}),n;throw ot(t,u,i.signal)}}}return l()}const H=Symbol(`deadlineExceeded`);function Jt(e,t){if(e===void 0)return{signal:t,timer:void 0};let n=new AbortController;if(e<=0)return n.abort(H),{signal:t?AbortSignal.any([t,n.signal]):n.signal,timer:void 0};let r=setTimeout(()=>n.abort(H),e);return{signal:t?AbortSignal.any([t,n.signal]):n.signal,timer:r}}function Yt(e,t){return e instanceof p&&e.type===`aborted`&&e.code===void 0&&t?.reason===H&&(e.code=`deadline_exceeded`),e}var Xt=class{capacity;refillPerMs;available;lastRefill=Date.now();constructor(e,t){this.capacity=e,this.refillPerMs=t,this.available=e}refill(){if(this.refillPerMs===0)return;let e=Date.now(),t=e-this.lastRefill;this.available=Math.min(this.capacity,this.available+Math.max(0,t)*this.refillPerMs),this.lastRefill=e}tryTake(e){return this.refill(),this.available<e?!1:(this.available-=e,!0)}msUntilAvailable(e){return this.refill(),this.available>=e?0:this.refillPerMs===0?1/0:(e-this.available)/this.refillPerMs}give(e){this.available=Math.min(this.capacity,this.available+e)}getCapacity(){return this.capacity}resize(e){this.refill(),this.refillPerMs>0&&(this.refillPerMs=this.refillPerMs/this.capacity*e),this.capacity=e,this.available=Math.min(this.available,e)}};function Zt(e){let t=e.messages.reduce((e,t)=>{let n=t.content;if(typeof n==`string`)return e+n.length;if(n==null)return e;try{return e+JSON.stringify(n).length}catch{return e}},0);return Math.ceil(t/4)+(e.max_tokens??0)}function Qt(e){if(!e)return;let t=t=>{if(!Number.isFinite(e[t]))throw new p(`aimd.${t} (${e[t]}) must be a finite number.`,`invalid_params`)};for(let e of[`minCapacity`,`maxCapacity`,`increaseBy`,`decreaseFactor`])t(e);if(e.minCapacity<1||e.maxCapacity<1)throw new p(`aimd.minCapacity (${e.minCapacity}) and aimd.maxCapacity (${e.maxCapacity}) must both be at least 1, since the requests bucket always takes 1 per acquire; a capacity below 1 could never be satisfied.`,`invalid_params`);if(e.minCapacity>e.maxCapacity)throw new p(`aimd.minCapacity (${e.minCapacity}) must not exceed aimd.maxCapacity (${e.maxCapacity}).`,`invalid_params`);return{increaseBy:Math.max(0,e.increaseBy),decreaseFactor:Math.min(1,Math.max(Number.MIN_VALUE,e.decreaseFactor)),minCapacity:e.minCapacity,maxCapacity:e.maxCapacity,proactiveFloor:e.proactiveFloor??0}}function $t(e,t){return e===`tpm`?t:1}function en(e){if(e)return new Xt(e,e/6e4)}var tn=class{requests;tokens;concurrency;buckets;maxQueueMs;maxQueueSize;estimateTokensFn;aimd;queue=[];wakeTimer;constructor(e){this.requests=en(e.requestsPerMinute),this.tokens=en(e.tokensPerMinute),e.maxConcurrent&&(this.concurrency=new Xt(e.maxConcurrent,0)),this.buckets=[{reason:`concurrency`,bucket:this.concurrency},{reason:`rpm`,bucket:this.requests},{reason:`tpm`,bucket:this.tokens}].flatMap(({reason:e,bucket:t})=>t?[{reason:e,bucket:t}]:[]),this.maxQueueMs=e.maxQueueMs??3e4,this.maxQueueSize=e.maxQueueSize??0,this.estimateTokensFn=e.estimateTokens??Zt,this.aimd=this.requests?Qt(e.aimd):void 0}estimate(e){return this.estimateTokensFn(e)}async acquire(e,t){if(t?.aborted)throw new p(`LLM request aborted`,`aborted`);if(!Number.isFinite(e)||e<0)throw new p(`estimatedTokens must be a finite, non-negative number, got ${String(e)}`,`invalid_params`);if(this.tokens&&e>this.tokens.getCapacity())throw new p(`estimatedTokens (${e}) exceeds the configured tokensPerMinute capacity (${this.tokens.getCapacity()}); this call could never acquire capacity.`,`rate_limited`,{code:`rate_limit_capacity_exceeded`});if(this.queue.length===0){let n=this.tryAcquireBuckets(e);return n.ok?{release:this.makeRelease(e),waitedMs:0}:this.enqueue(e,n.reason,t)}if(this.maxQueueSize>0&&this.queue.length>=this.maxQueueSize)throw this.queueFullError();return this.enqueue(e,void 0,t)}queueFullError(){return new p(`Rate limit queue is full`,`rate_limited`,{code:`rate_limit_queue_full`})}enqueue(e,t,n){return new Promise((r,i)=>{let a={estimatedTokens:e,enqueuedAt:Date.now(),lastReason:t,resolve:e=>{c(),r(e)},reject:e=>{c(),this.wakeTimer&&=(clearTimeout(this.wakeTimer),void 0),this.drain(),i(e)}},o,s=()=>{a.reject(new p(`LLM request aborted`,`aborted`))},c=()=>{o&&clearTimeout(o),n?.removeEventListener(`abort`,s);let e=this.queue.indexOf(a);e!==-1&&this.queue.splice(e,1)};this.maxQueueMs>0&&(o=setTimeout(()=>{a.reject(new p(`Rate limit queue timed out before capacity was available`,`rate_limited`,{code:`rate_limit_queue_timeout`}))},this.maxQueueMs)),n?.addEventListener(`abort`,s,{once:!0}),this.queue.push(a),this.drain()})}tryAcquireBuckets(e){let t=[];for(let{reason:n,bucket:r}of this.buckets){let i=$t(n,e);if(!r.tryTake(i)){for(let e of t)e.bucket.give(e.amount);return{ok:!1,reason:n}}t.push({bucket:r,amount:i})}return{ok:!0}}drain(){for(;this.queue.length>0;){let e=this.queue[0],t=this.tryAcquireBuckets(e.estimatedTokens);if(!t.ok){e.lastReason=t.reason,this.scheduleWake(t.reason,e.estimatedTokens);return}let n=Date.now()-e.enqueuedAt;e.resolve({release:this.makeRelease(e.estimatedTokens),waitedMs:n,reason:e.lastReason})}}scheduleWake(e,t){if(this.wakeTimer||e===`concurrency`)return;let n=(this.buckets.find(t=>t.reason===e)?.bucket)?.msUntilAvailable($t(e,t));if(n===void 0||!Number.isFinite(n))return;let r=Math.min(Math.max(1,Math.ceil(n)),2147483647);this.wakeTimer=setTimeout(()=>{this.wakeTimer=void 0,this.drain()},r)}makeRelease(e){let t=!1;return(n,r=!1)=>{t||(t=!0,this.concurrency?.give(1),this.tokens&&n!==void 0&&Number.isFinite(n)&&this.tokens.give(e-n),r&&this.growOnSuccess(),this.drain())}}resizeRequestsCeiling(e){!this.aimd||!this.requests||this.requests.resize(e(this.aimd,this.requests.getCapacity()))}growOnSuccess(){this.resizeRequestsCeiling((e,t)=>Math.min(t+e.increaseBy,e.maxCapacity))}signalRateLimit(){this.resizeRequestsCeiling((e,t)=>Math.max(t*e.decreaseFactor,e.minCapacity))}reactToRateLimitHint(e){if(!this.aimd||!e||e.remainingRequests===void 0)return;let t=this.aimd.proactiveFloor??0;t>0&&e.remainingRequests<=t&&this.signalRateLimit()}};const U=[`estimate`,`acquire`,`signalRateLimit`,`reactToRateLimitHint`];function nn(e){let t=e;return U.every(e=>typeof t[e]==`function`)}function rn(e){let t=e;return U.some(e=>typeof t[e]==`function`)}function an(e){if(e!==void 0){if(nn(e))return e;if(rn(e))throw new p(`rateLimit looks like a RateLimiterAdapter but is missing: ${U.filter(t=>typeof e[t]!=`function`).join(`, `)}. All four methods (${U.join(`, `)}) are required.`,`invalid_params`);return new tn(e)}}function on(e,t,n){return[e,...t].map((e,t)=>{let r=t>0,i=e.name??(r?`fallback[${t-1}]`:n.providerName),a=ct(e.circuitBreaker,i,e.model,n.onEvent,n.logger,n.middleware,n.middlewareTimeoutMs,r,e.client.supportsJsonObjectMode??!0),o=e.retryBudget?new Ue(e.retryBudget):void 0;return new Ht(i,e.client,e.model,{maxRetries:e.maxRetries??n.maxRetries??1,timeoutMs:e.timeoutMs??n.timeoutMs??25e3,chunkIdleTimeoutMs:e.chunkIdleTimeoutMs??n.chunkIdleTimeoutMs??3e4,baseDelayMs:e.baseDelayMs??n.baseDelayMs??500,defaultMaxTokens:e.defaultMaxTokens??n.defaultMaxTokens??1e3,defaultTemperature:e.defaultTemperature===void 0?n.primaryDefaultTemperature:e.defaultTemperature,defaultReasoningEffort:e.defaultReasoningEffort===void 0?n.primaryDefaultReasoningEffort:e.defaultReasoningEffort,defaultBudgetTokens:e.defaultBudgetTokens===void 0?n.primaryDefaultBudgetTokens:e.defaultBudgetTokens,nonRetryableStatus:e.nonRetryableStatus??n.nonRetryableStatus??[400,401,403,404,422],parseJson:n.parseJson,logger:n.logger,redact:n.redact,onUsage:n.onUsage,onUsageFailure:n.onUsageFailure,onEvent:n.onEvent,breaker:a,budget:o,limiter:an(e.rateLimit),isFallback:r,middleware:n.middleware,middlewareTimeoutMs:n.middlewareTimeoutMs,detectSoftFailure:e.detectSoftFailure??n.detectSoftFailure})})}function sn(e){let t=e;return typeof t.get==`function`&&typeof t.set==`function`}function cn(e){if(e===void 0)return new b;if(sn(e))return e;if(e.maxSize!==void 0&&(!Number.isFinite(e.maxSize)||e.maxSize<0||!Number.isInteger(e.maxSize)))throw RangeError(`cache.maxSize must be a finite, non-negative integer, got ${String(e.maxSize)}`);return new b(e.maxSize,e.eviction)}function ln(e){return{debug:W(e,`debug`),warn:W(e,`warn`),error:W(e,`error`)}}function W(e,t){let n=e[t].bind(e);return((...e)=>{try{dn(n(...e))}catch{}})}function un(e){return typeof e?.then==`function`}function dn(e){un(e)&&Promise.resolve(e).catch(()=>{})}var fn=class{debugEnabled;constructor(e){this.debugEnabled=e}debug(e){this.debugEnabled&&console.debug(e)}warn(e){console.warn(e)}error(e,t){console.error(e,t??``)}},pn=class{logger;executors;fallbackOn;reportEvent;cacheOrchestrator;middleware;middlewareTimeoutMs;cachedCallInnerParams=new WeakMap;cachedCallMeta=new Map;constructor(e){this.logger=ln(e.logger??new fn(e.debug??!1));let t=e.name??`primary`;this.cacheOrchestrator=new ve(cn(e.cache),this.logger),this.fallbackOn=e.fallbackOn??ae,this.reportEvent=I(e.onEvent,this.logger),this.middleware=[...e.middleware??[]].sort((e,t)=>(e.priority??0)-(t.priority??0)),this.middlewareTimeoutMs=e.middlewareTimeoutMs??5e3;let n=e.defaultTemperature===void 0?.2:e.defaultTemperature,r=e.defaultReasoningEffort,i=e.defaultBudgetTokens,a={client:e.client,model:e.model,name:t,maxRetries:e.maxRetries,timeoutMs:e.timeoutMs,chunkIdleTimeoutMs:e.chunkIdleTimeoutMs,baseDelayMs:e.baseDelayMs,defaultMaxTokens:e.defaultMaxTokens,defaultTemperature:n,defaultReasoningEffort:r,defaultBudgetTokens:i,nonRetryableStatus:e.nonRetryableStatus,circuitBreaker:e.circuitBreaker,rateLimit:e.rateLimit,retryBudget:e.retryBudget,detectSoftFailure:e.detectSoftFailure},o=Array.isArray(e.fallback)?e.fallback:e.fallback?[e.fallback]:[];this.executors=on(a,o,{providerName:t,primaryDefaultTemperature:n,primaryDefaultReasoningEffort:r,primaryDefaultBudgetTokens:i,maxRetries:e.maxRetries,timeoutMs:e.timeoutMs,chunkIdleTimeoutMs:e.chunkIdleTimeoutMs,baseDelayMs:e.baseDelayMs,defaultMaxTokens:e.defaultMaxTokens,nonRetryableStatus:e.nonRetryableStatus,parseJson:e.parseJson,redact:e.redact,onUsage:e.onUsage,onUsageFailure:e.onUsageFailure,onEvent:e.onEvent,logger:this.logger,middleware:this.middleware,middlewareTimeoutMs:this.middlewareTimeoutMs,detectSoftFailure:e.detectSoftFailure})}logRefundError(e,t){this.logger.error(e,{message:t instanceof Error?t.message:`unknown`})}get logicalCallDependencies(){return{executors:this.executors,fallbackOn:this.fallbackOn,reportEvent:this.reportEvent,middleware:this.middleware,middlewareTimeoutMs:this.middlewareTimeoutMs,logger:this.logger}}get runOperationDependencies(){return{middleware:this.middleware,primaryExecutor:this.executors[0],middlewareTimeoutMs:this.middlewareTimeoutMs,logger:this.logger,reportEvent:this.reportEvent}}async call(e){if(e.signal?.aborted)throw new p(`LLM request aborted`,`aborted`);let t=e.requestId??globalThis.crypto.randomUUID(),n=this.cachedCallInnerParams.get(e),r=n!==void 0,{signal:i,timer:a}=Jt(e.deadlineMs,e.signal),o=i===e.signal?e:{...e,signal:i},s=n??T();try{let e=this.executors.length===1,n=await V(this.runOperationDependencies,o,t,s,async()=>{if(e){if(o.signal?.aborted)throw new p(`LLM request aborted`,`aborted`);this.executors[0].assertBreakerClosed(o.model,{requestId:t,state:s,signal:o.signal})}let n,r=async e=>{let t=await e;return n=t.meta,t.value};return o.stream?{value:await pe(o,()=>r(Kt(this.logicalCallDependencies,o,t,e,s)),o.signal,(e,t)=>this.logRefundError(e,t)),meta:n}:{value:await E(o,!1,()=>r(Gt(this.logicalCallDependencies,o,t,e,s)),o.signal,(e,t)=>this.logRefundError(e,t)),meta:n}},r);return o.stream,n.value}catch(e){throw Yt(e,i)}finally{clearTimeout(a)}}runCached(e){return this.cacheOrchestrator.runCached(e)}async deleteCache(e){await this.cacheOrchestrator.deleteCache(e)}async cachedCall(e){let{call:t,...n}=e,r=t;if(r.reserveUsage||r.refundUsage)throw new p("`reserveUsage`/`refundUsage` were set inside `call`, where cachedCall ignores them. Move them to the top level of the cachedCall() params, alongside cacheKey/ttl, instead.",`invalid_params`);let i=r.requestId??globalThis.crypto.randomUUID(),a=T(),o=await this.cacheOrchestrator.resolveCacheKey(n.cacheKey),s=this.cachedCallMeta.get(o),c=!1;s||(s={},c=!0,this.cachedCallMeta.set(o,s));let l=()=>{c&&this.cachedCallMeta.delete(o)},u=r.meta,d=e=>{let t={...e,meta:s};return this.cachedCallInnerParams.set(t,a),this.call(t).finally(()=>{u&&(u.current=s.current),this.cachedCallInnerParams.delete(t)})};if(r.stream){let e={...r,requestId:i},t;try{t=await V(this.runOperationDependencies,e,i,a,async()=>({value:await this.cacheOrchestrator.runCachedStream({...n,openStream:()=>d(e)},!!r.tools),meta:s.current}))}catch(e){throw l(),e}let o=t.value,c=()=>{u&&(u.current=s.current),l()};return o.finalResult.then(c,c),o}let f={...r,requestId:i};try{let e=await V(this.runOperationDependencies,f,i,a,async()=>({value:await this.runCached({...n,fn:()=>d(f)}),meta:s.current}));return u&&(u.current=s.current),e.value}finally{l()}}getCircuitState(e){let t=L(this.executors,e?.index??0,`getCircuitState`);return R(t.isolateByModel,e?.model,`getCircuitState`,this.logger),t.getCircuitState(e?.model??t.model)}getFailureBreakdown(e){let t=L(this.executors,e?.index??0,`getFailureBreakdown`);return R(t.isolateByModel,e?.model,`getFailureBreakdown`,this.logger),t.getFailureBreakdown(e?.model??t.model)}getRetryBudgetState(e){return L(this.executors,e?.index??0,`getRetryBudgetState`).getRetryBudgetState()}getCircuitStates(e){return this.executors.map((t,n)=>({provider:t.providerName,index:n,isFallback:n>0,isolateByModel:t.isolateByModel,state:t.getCircuitState(e??t.model)}))}openCircuit(e){let t=L(this.executors,e?.index??0,`openCircuit`);R(t.isolateByModel,e?.model,`openCircuit`,this.logger),t.openCircuit(e?.model??t.model,{requestId:globalThis.crypto.randomUUID(),state:T()})}closeCircuit(e){let t=L(this.executors,e?.index??0,`closeCircuit`);R(t.isolateByModel,e?.model,`closeCircuit`,this.logger),t.closeCircuit(e?.model??t.model,{requestId:globalThis.crypto.randomUUID(),state:T()})}};function mn(e){return e}function hn(e){return e}async function*gn(e){let t=new TextDecoder(`utf-8`,{fatal:!0}),n=``;for await(let r of e){let e;try{e=typeof r==`string`?r:t.decode(r,{stream:!0})}catch(e){throw new p(`Invalid UTF-8 in SSE stream`,`parse`,{cause:e})}n=(n+e).replace(/\r\n/g,`
|
|
2
2
|
`).replace(/\r(?!$)/g,`
|
|
3
3
|
`);let i=n.indexOf(`
|
|
4
4
|
|
|
@@ -11,5 +11,5 @@ Object.defineProperty(exports,Symbol.toStringTag,{value:`Module`});const e=new S
|
|
|
11
11
|
|
|
12
12
|
`)}let i=n.trim();if(i){let e=J(i);e!==G&&e!==K&&(yield e)}}const G=Symbol(`sse-stream-done`),K=Symbol(`sse-frame-no-data`),q=Symbol(`sse-frame-ping`);function J(e){let t=[],n=!1;for(let r of e.split(`
|
|
13
13
|
`)){if(r.startsWith(`:`)){n=!0;continue}r.startsWith(`data:`)&&t.push(r.startsWith(`data: `)?r.slice(6):r.slice(5))}if(!t.length)return n?q:K;let r=t.join(`
|
|
14
|
-
`);if(r===`[DONE]`)return G;try{return JSON.parse(r)}catch(e){throw new p(`Invalid JSON in SSE frame: ${r.slice(0,200)}`,`parse`,{cause:e,code:`stream_frame_invalid`})}}const mn=[`image/png`,`image/jpeg`,`image/gif`,`image/webp`];function Y(e){if(mn.includes(e))return e;throw new p(`Unsupported image mimeType "${e}": expected one of ${mn.join(`, `)}`,`invalid_params`)}function hn(e,t){return t?Array.isArray(t)?t.includes(e):t(e):!1}const X={minimal:1024,low:4096,medium:16e3,high:32e3};function Z(e){if(!e)return X;let t={...X,...e};if(!(t.minimal<t.low&&t.low<t.medium&&t.medium<t.high))throw new p(`reasoningEffortTokens must keep tiers in strictly ascending order (minimal < low < medium < high), got ${JSON.stringify(t)}. An out-of-order override doesn't just misrank tiers, it can make some of them unreachable.`,`invalid_params`);return t}function gn(e,t=X){return t[e]}function Q(e,t=X){return e<=t.minimal?`minimal`:e<=t.low?`low`:e<=t.medium?`medium`:`high`}function _n(e){let t=/opus-(\d+)(?:-(\d+))?/.exec(e);if(!t)return null;let n=t[2],r=n===void 0||n.length>=8?0:Number(n);return[Number(t[1]),r]}function vn(e){let t=_n(e);if(t){let[e,n]=t;return e>4||e===4&&n>=7}return[`sonnet-5`,`fable-5`,`mythos`].some(t=>e.includes(t))}function yn(e,t){return vn(e)?!0:t?Array.isArray(t)?t.includes(e):t(e):!1}function bn(e,t){return!yn(e,t)}function xn(e,t){if(e<1024)throw new p(`budgetTokens (${e}) is below Anthropic's minimum of 1024. Raise budgetTokens, or use a reasoningEffort tier of 'low' or above with the default conversion table.`,`invalid_params`);if(e>=t)throw new p(`budgetTokens (${e}) must be less than maxTokens (${t}); the thinking budget and the reply share the same max_tokens ceiling on Anthropic. Raise maxTokens, or lower budgetTokens/reasoningEffort.`,`invalid_params`)}function Sn(e){if(e)throw new p(`budgetTokens/reasoningEffort was set alongside ${e}. Anthropic rejects thinking combined with a tool_choice that forces tool use, the model has to be able to reply with plain text for thinking to run. Use toolChoice: 'auto' (or omit toolChoice) for this call, or drop budgetTokens/reasoningEffort for it.`,`invalid_params`)}function Cn(e){return e===`minimal`?`low`:e}function wn(e,t){return En(t,e.toUpperCase())}function Tn(e){let t=/gemini-\d+\.(\d+)/.exec(e);return t?Number(t[1]):0}function En(e,t){if(!e.includes(`pro`))return t;let n=Dn(e);return n===null||n<3?t:Tn(e)===0?t===`HIGH`?`HIGH`:`LOW`:t===`MINIMAL`?`LOW`:t}function Dn(e){let t=/gemini-(\d+)/.exec(e);return t?Number(t[1]):null}function On(e){let t=Dn(e);return t!==null&&t>=3}function kn(e,t){return On(e)?!0:t?Array.isArray(t)?t.includes(e):t(e):!1}function An(e){return e.map(e=>e.type===`image`?{type:`image`,source:{type:`base64`,media_type:Y(e.mimeType),data:e.data}}:{type:`text`,text:e.text})}function jn(e,t){if(e.type!==`object`)throw new p(`Tool "${t}"'s schema must have "type": "object" (Anthropic requires object-shaped tool parameters).`,`validation`);return e}function Mn(e){return!e||e===`auto`?{type:`auto`}:e===`none`?{type:`none`}:e===`required`?{type:`any`}:{type:`tool`,name:e.function.name}}function Nn(e,t){return{tools:e.map(e=>({name:e.function.name,description:e.function.description,input_schema:jn(e.function.parameters,e.function.name)})),toolChoice:Mn(t)}}function Pn(e,t,n,r){let i=e.messages.find(e=>e.role===`system`),a=e.messages.filter(e=>e.role===`user`||e.role===`assistant`||e.role===`tool`),o=e.response_format?.type===`json_schema`?e.response_format.json_schema:void 0,s=o?.name.trim();if(o&&!s)throw new p(`json_schema.name must not be empty.`,`validation`);let c=!!o&&hn(e.model,t);if(o&&e.tools?.length&&!c)throw new p(`Anthropic model "${e.model}" is not covered by nativeStructuredOutputModels, so \`jsonSchema\` is emulated as a forced single tool call there, which collides with the \`tools\` you also provided. Either drop \`tools\` or \`jsonSchema\` for this call, or pass this model in fromAnthropic's \`nativeStructuredOutputModels\` option once you've confirmed it supports Anthropic's \`output_config.format\`.`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`tools_with_json_schema`}});if(e.response_format?.type===`json_object`)throw new p('response_format: "json_object" is not supported on Anthropic. Unlike OpenAI, Anthropic has no API-level field that mechanically guarantees valid JSON output for this mode, so it used to be emulated by injecting a "respond with JSON only" instruction into the system prompt, a guarantee this adapter can no longer make. Use `jsonSchema` instead, which maps to a real API-level constraint (Anthropic\'s native output_config.format on covered models, or a forced single tool call otherwise).',`validation`);let l,u,d,f;if(o&&c)u={type:`json_schema`,schema:o.schema},e.tools?.length&&({tools:d,toolChoice:f}=Nn(e.tools,e.tool_choice));else if(o&&s){let{schema:e,description:t,strict:n}=o;l=s,d=[{name:l,description:t,input_schema:jn(e,l),strict:n}],f={type:`tool`,name:l}}!o&&e.tools?.length&&({tools:d,toolChoice:f}=Nn(e.tools,e.tool_choice));let m,h;if(e.budget_tokens!==void 0||e.reasoning_effort!==void 0){if(Sn(f?.type===`tool`?`toolChoice forcing the "${f.name}" tool`:f?.type===`any`?`toolChoice: 'required' (Anthropic's "any" tool_choice)`:void 0),bn(e.model,r)){let t=e.budget_tokens??gn(e.reasoning_effort,n);xn(t,e.max_tokens),m={type:`enabled`,budget_tokens:t}}else{let t=e.reasoning_effort??Q(e.budget_tokens,n);m={type:`adaptive`},h=Cn(t)}}let g=i?.content,_=m?void 0:e.temperature;return{body:{model:e.model,max_tokens:e.max_tokens,..._===void 0?{}:{temperature:_},system:g||void 0,messages:In(a.map(e=>Ln(e))),...d?{tools:d,tool_choice:f}:{},...u||h?{output_config:{...u?{format:u}:{},...h?{effort:h}:{}}}:{},...m?{thinking:m}:{}},toolName:l}}function Fn(e,t){let n=t?.nativeStructuredOutputModels,r=Z(t?.reasoningEffortTokens),i=t?.adaptiveOnlyModels,a=t?.supportsWithResponse??!1,o=e.messages.create.bind(e.messages);return{supportsJsonObjectMode:!1,chat:{completions:{async create(t,o){let{body:s,toolName:c}=Pn(t,n,r,i),l,u=a?await(async()=>{let{data:t,response:n}=await e.messages.create(s,o).withResponse();return l=n.headers,t})():await e.messages.create(s,o),d,f;if(c){let e=u.content.find(e=>e.type===`tool_use`&&e.name===c);if(!e)throw new p(`Anthropic did not return the required structured output tool "${c}".`,`validation`);if(!e.input||typeof e.input!=`object`||Array.isArray(e.input))throw new p(`Anthropic returned invalid structured output for tool "${c}". Expected an object.`,`validation`);d=JSON.stringify(e.input)}else{d=u.content.filter(e=>e.type===`text`).map(e=>e.text??``).join(``);let e=u.content.filter(e=>e.type===`tool_use`);e.length&&(f=e.map(e=>({id:e.id,type:`function`,function:{name:e.name,arguments:JSON.stringify(e.input??{})}})))}let m={choices:[{message:{content:d,...f?{tool_calls:f}:{}}}],usage:{prompt_tokens:u.usage?.input_tokens,completion_tokens:u.usage?.output_tokens,total_tokens:(u.usage?.input_tokens??0)+(u.usage?.output_tokens??0),...u.usage?.output_tokens_details?.thinking_tokens===void 0?{}:{completion_tokens_details:{reasoning_tokens:u.usage.output_tokens_details.thinking_tokens}}}};return l&<(m,st(l)),m},async*createStream(t,s){let{body:c,toolName:l}=Pn(t,n,r,i),u;if(!a)u=await o({...c,stream:!0},s);else{let{data:t,response:n}=await e.messages.create({...c,stream:!0},s).withResponse();u=t;let r=st(n.headers);(r.remainingRequests!==void 0||r.limitRequests!==void 0)&&(yield{type:`rate_limit_hint`,hint:r})}let d=new Map,f=0,m=!1;for await(let e of u)if(e.type===`message_start`)f=e.message.usage?.input_tokens??0;else if(e.type===`content_block_start`){if(e.content_block.type===`tool_use`){let t=e.content_block.name===l?`json-tool`:`tool_use`;d.set(e.index,t),t===`json-tool`?m=!0:l||(yield{type:`tool_call_delta`,index:e.index,id:e.content_block.id,name:e.content_block.name})}else d.set(e.index,`text`)}else if(e.type===`content_block_delta`)e.delta.type===`text_delta`?l||(yield{type:`text-delta`,delta:e.delta.text}):e.delta.type===`input_json_delta`&&(d.get(e.index)===`json-tool`?yield{type:`text-delta`,delta:e.delta.partial_json}:l||(yield{type:`tool_call_delta`,index:e.index,argumentsDelta:e.delta.partial_json}));else if(e.type===`message_delta`){let t=e.usage?.output_tokens??0,n=e.usage?.output_tokens_details?.thinking_tokens;yield{type:`usage`,usage:{prompt_tokens:f,completion_tokens:t,total_tokens:f+t,...n===void 0?{}:{completion_tokens_details:{reasoning_tokens:n}}}}}else e.type===`ping`&&(yield{type:`ping`});if(l&&!m)throw new p(`Anthropic did not return the required structured output tool "${l}".`,`validation`)}}}}}function In(e){let t=e=>e.role===`user`&&Array.isArray(e.content)&&e.content.length>0&&e.content.every(e=>e.type===`tool_result`),n=[];for(let r of e){let e=n.at(-1);t(r)&&e&&t(e)?e.content.push(...r.content):n.push(r)}return n}function Ln(e){if(e.role===`tool`)return{role:`user`,content:[{type:`tool_result`,tool_use_id:e.tool_call_id,content:e.content,...e.is_error?{is_error:!0}:{}}]};if(e.role===`assistant`&&e.tool_calls?.length){let t=[];e.content&&t.push({type:`text`,text:e.content});for(let n of e.tool_calls){let e;try{e=n.function.arguments.trim()?JSON.parse(n.function.arguments):{}}catch(e){throw new p(`Assistant tool call "${n.function.name}" (${n.id}) has arguments that are not valid JSON.`,`validation`,{cause:e})}if(e===null||Array.isArray(e)||typeof e!=`object`)throw new p(`Assistant tool call "${n.function.name}" (${n.id}) arguments must be a JSON object.`,`validation`);t.push({type:`tool_use`,id:n.id,name:n.function.name,input:e})}return{role:`assistant`,content:t}}return{role:e.role,content:Array.isArray(e.content)?An(e.content):e.content??``}}function Rn(e){return e.map(e=>e.type===`image`?{inlineData:{mimeType:Y(e.mimeType),data:e.data}}:{text:e.text})}function zn(e,t){return`${e}#${t}`}function Bn(e){return!e||e===`auto`?{functionCallingConfig:{mode:`AUTO`}}:e===`none`?{functionCallingConfig:{mode:`NONE`}}:e===`required`?{functionCallingConfig:{mode:`ANY`}}:{functionCallingConfig:{mode:`ANY`,allowedFunctionNames:[e.function.name]}}}function Vn(e){let t=new Map;for(let n of e)if(n.role===`assistant`)for(let e of n.tool_calls??[])t.set(e.id,e.function.name);return t}function Hn(e,t){if(e.role===`tool`)return{role:`user`,parts:[{functionResponse:{id:e.tool_call_id,name:t.get(e.tool_call_id)??e.tool_call_id,response:Wn(e.content)}}]};if(e.role===`assistant`&&e.tool_calls?.length){let t=[];return typeof e.content==`string`&&e.content&&t.push({text:e.content}),t.push(...e.tool_calls.map(e=>({functionCall:{id:e.id,name:e.function.name,args:Un(e.function.arguments,e.function.name)}}))),{role:`model`,parts:t}}return{role:e.role===`assistant`?`model`:`user`,parts:Array.isArray(e.content)?Rn(e.content):[{text:e.content??``}]}}function Un(e,t){let n;try{n=e.trim()?JSON.parse(e):{}}catch(e){throw new p(`Tool call "${t}" arguments are not valid JSON.`,`parse`,{cause:e,code:`tool_arguments_parse_failed`})}if(!n||Array.isArray(n)||typeof n!=`object`)throw new p(`Tool call "${t}" arguments must be a JSON object.`,`validation`);return n}function Wn(e){let t;try{t=e.trim()?JSON.parse(e):``}catch{t=e}return t&&!Array.isArray(t)&&typeof t==`object`?t:{output:t}}function Gn(e){let t=e=>e.role===`user`&&e.parts.length>0&&e.parts.every(e=>`functionResponse`in e),n=[];for(let r of e){let e=n.at(-1);t(r)&&e&&t(e)?e.parts.push(...r.parts):n.push(r)}return n}function Kn(e,t,n){let r=e.messages.find(e=>e.role===`system`),i=e.messages.filter(e=>e.role===`user`||e.role===`assistant`||e.role===`tool`),a=!!e.response_format,o={...e.temperature===void 0?{}:{temperature:e.temperature},maxOutputTokens:e.max_tokens,...r?{systemInstruction:{parts:[{text:r.content}]}}:{}};if(a&&(o.responseMimeType=`application/json`),e.response_format?.type===`json_schema`){let{schema:t,description:n}=e.response_format.json_schema;o.responseSchema={...t,...n?{description:n}:{}}}if(e.tools?.length&&(o.tools=[{functionDeclarations:e.tools.map(e=>({name:e.function.name,description:e.function.description,parameters:e.function.parameters}))}],o.toolConfig=Bn(e.tool_choice)),kn(e.model,n)){let n=e.reasoning_effort??(e.budget_tokens===void 0?void 0:Q(e.budget_tokens,t));n!==void 0&&(o.thinkingConfig={thinkingLevel:wn(n,e.model)})}else{let n=e.budget_tokens??(e.reasoning_effort?gn(e.reasoning_effort,t):void 0);n!==void 0&&(o.thinkingConfig={thinkingBudget:n})}let s=Vn(i);return{model:e.model,contents:Gn(i.map(e=>Hn(e,s))),config:o}}function qn(e,t){let n=Z(t?.reasoningEffortTokens),r=t?.thinkingLevelModels,i=e.models??e;if(typeof i.generateContent!=`function`)throw new p(`fromGemini requires a client with generateContent: pass ai.models, or the whole ai client (fromGemini(ai)).`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`generateContent`}});let a=i.generateContent.bind(i),o=typeof i.generateContentStream==`function`?i.generateContentStream.bind(i):void 0;return{chat:{completions:{async create(e,t){let i=Kn(e,n,r);i.config={...i.config,abortSignal:t.signal};let o=await a(i),s=o.candidates?.[0]?.content?.parts??[],c=s.map(e=>e.text??``).join(``),l=s.filter(e=>e.functionCall),u;if(l.length){let e=new Map;u=l.map(t=>{let n=t.functionCall.name,r=t.functionCall.id,i=e.get(n)??0;return e.set(n,i+1),{id:r??zn(n,i),type:`function`,function:{name:n,arguments:JSON.stringify(t.functionCall.args??{})}}})}return{choices:[{message:{content:c,...u?{tool_calls:u}:{}}}],usage:{prompt_tokens:o.usageMetadata?.promptTokenCount,completion_tokens:o.usageMetadata?.candidatesTokenCount,total_tokens:o.usageMetadata?.totalTokenCount,...o.usageMetadata?.thoughtsTokenCount===void 0?{}:{completion_tokens_details:{reasoning_tokens:o.usageMetadata.thoughtsTokenCount}}}}},async*createStream(e,t){if(!o)throw new p(`stream: true requires a Gemini client with generateContentStream`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`generateContentStream`}});let i=Kn(e,n,r);i.config={...i.config,abortSignal:t.signal};let a=await o(i),s=0,c=new Map,l;for await(let e of a){let t=e.candidates?.[0]?.content?.parts??[];for(let e of t)if(e.text&&(yield{type:`text-delta`,delta:e.text}),e.functionCall){let t=e.functionCall.name,n=e.functionCall.id,r=t?c.get(t)??0:0;t&&c.set(t,r+1),yield{type:`tool_call_delta`,index:s,id:n??(t?zn(t,r):void 0),name:e.functionCall.name,argumentsDelta:JSON.stringify(e.functionCall.args??{}),complete:!0},s++}e.usageMetadata&&(l=e.usageMetadata)}l&&(yield{type:`usage`,usage:{prompt_tokens:l.promptTokenCount,completion_tokens:l.candidatesTokenCount,total_tokens:l.totalTokenCount,...l.thoughtsTokenCount===void 0?{}:{completion_tokens_details:{reasoning_tokens:l.thoughtsTokenCount}}}})}}}}}function Jn(e){return e.includes(`claude`)}function Yn(e){switch(Y(e)){case`image/png`:return`png`;case`image/jpeg`:return`jpeg`;case`image/gif`:return`gif`;case`image/webp`:return`webp`}}function Xn(e){return new Uint8Array(Buffer.from(e,`base64`))}function Zn(e){return e.map(e=>e.type===`image`?{image:{format:Yn(e.mimeType),source:{bytes:Xn(e.data)}}}:{text:e.text})}function Qn(e,t){return{tools:e.map(e=>({toolSpec:{name:e.function.name,description:e.function.description,inputSchema:{json:e.function.parameters}}})),toolChoice:ar(t)}}function $n(e,t,n,r,i){let a=e.messages.find(e=>e.role===`system`),o=e.messages.filter(e=>e.role===`user`||e.role===`assistant`||e.role===`tool`),s=e.response_format?.type===`json_schema`?e.response_format.json_schema:void 0,c=s?.name.trim();if(s&&!c)throw new p(`json_schema.name must not be empty.`,`validation`);let l=!!s&&hn(e.model,n);if(s&&e.tools?.length&&!l)throw new p(`Bedrock model "${e.model}" is not covered by nativeStructuredOutputModels, so \`jsonSchema\` is emulated as a forced single tool call there (via \`toolConfig\`), which collides with the \`tools\` you also provided. Either drop \`tools\` or \`jsonSchema\` for this call, or pass this model in fromBedrock's \`nativeStructuredOutputModels\` option once you've confirmed it supports Converse's \`outputConfig.textFormat\`.`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`tools_with_json_schema`}});if(e.response_format?.type===`json_object`)throw new p('response_format: "json_object" is not supported on Bedrock. Converse has no field that mechanically guarantees valid JSON output for this mode, so it used to be emulated by injecting a "respond with JSON only" instruction into the system prompt, a guarantee this adapter can no longer make. Use `jsonSchema` instead, which maps to a real constraint (Converse\'s native outputConfig.textFormat on covered models, or a forced tool call otherwise).',`validation`);let u,d,f;if(s&&l){let{schema:e,description:t}=s;f={textFormat:{type:`json_schema`,structure:{jsonSchema:{schema:JSON.stringify(e),name:c,description:t}}}}}else if(s&&c){let{schema:e,description:t,strict:n}=s;u=c,d={tools:[{toolSpec:{name:u,description:t,inputSchema:{json:e},strict:n}}],toolChoice:{tool:{name:u}}}}if(e.tools?.length&&!u&&(d=Qn(e.tools,e.tool_choice)),s&&d&&t&&!(Array.isArray(t)?t.includes(e.model):t(e.model)))throw new p(`Bedrock model "${e.model}" is not listed in toolUseSupportedModels, but this call requires Converse tool use (either jsonSchema emulated as a forced tool call, or real \`tools\` sent alongside native structured output).`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`toolUseSupportedModels`}});let m,h;if(Jn(e.model)&&(e.budget_tokens!==void 0||e.reasoning_effort!==void 0)){let t=d?.toolChoice;if(Sn(t&&`tool`in t?`toolChoice forcing the "${t.tool?.name}" tool`:t&&`any`in t?`toolChoice: 'required' (Converse's "any" tool_choice)`:void 0),bn(e.model,i)){let t=e.budget_tokens??gn(e.reasoning_effort,r);xn(t,e.max_tokens),m={thinking:{type:`enabled`,budget_tokens:t}}}else{let t=e.reasoning_effort??Q(e.budget_tokens,r);m={thinking:{type:`adaptive`}},h=Cn(t)}}let g=m?void 0:e.temperature;return{request:{modelId:e.model,messages:sr(o.map(e=>or(e))),system:a?.content?[{text:a.content}]:void 0,inferenceConfig:{...g===void 0?{}:{temperature:g},maxTokens:e.max_tokens},...d?{toolConfig:d}:{},...f||h?{outputConfig:{...f??{},...h?{effort:h}:{}}}:{},...m?{additionalModelRequestFields:m}:{}},toolName:u}}function er(e){return typeof e.send==`function`}function tr(e){if(`messageStart`in e)return{messageStart:e.messageStart};if(`contentBlockStart`in e)return{contentBlockStart:e.contentBlockStart};if(`contentBlockDelta`in e)return{contentBlockDelta:e.contentBlockDelta};if(`contentBlockStop`in e)return{contentBlockStop:e.contentBlockStop};if(`messageStop`in e)return{messageStop:e.messageStop};if(`metadata`in e)return{metadata:e.metadata};if(`internalServerException`in e)return{internalServerException:e.internalServerException};if(`modelStreamErrorException`in e)return{modelStreamErrorException:e.modelStreamErrorException};if(`validationException`in e)return{validationException:e.validationException};if(`throttlingException`in e)return{throttlingException:e.throttlingException};if(`serviceUnavailableException`in e)return{serviceUnavailableException:e.serviceUnavailableException}}async function*nr(e){for await(let t of e){let e=tr(t);e&&(yield e)}}function rr(e){let t;function n(){return t??=import(`@aws-sdk/client-bedrock-runtime`).then(e=>e,e=>{throw t=void 0,new p(`fromBedrock requires "@aws-sdk/client-bedrock-runtime" to be installed to use a raw AWS SDK client (it is not a dependency of vern-llm itself). Install it, or pass your own object with .converse()/.converseStream() methods instead.`,`validation`,{cause:e})}),t}return{converse:async(t,r)=>{let{ConverseCommand:i}=await n();return e.send(new i(t),{abortSignal:r.signal})},converseStream:async(t,r)=>{let{ConverseStreamCommand:i}=await n(),a=await e.send(new i(t),{abortSignal:r.signal});if(!a.stream)throw new p(`Bedrock ConverseStreamCommand response did not include a stream. This can happen if the request or the model doesn't actually support Converse streaming.`,`api`,{code:`server_error`});return{stream:nr(a.stream)}}}}function ir(e,t){let n=er(e)?rr(e):e,r=t?.toolUseSupportedModels,i=t?.nativeStructuredOutputModels,a=Z(t?.reasoningEffortTokens),o=t?.adaptiveOnlyModels;return{supportsJsonObjectMode:!1,chat:{completions:{async create(e,t){let{request:s,toolName:c}=$n(e,r,i,a,o),l=await n.converse(s,t),u,d;if(c){let e=l.output?.message?.content?.find(e=>e.toolUse?.name===c);u=e?.toolUse?JSON.stringify(e.toolUse.input):``}else{let e=l.output?.message?.content??[];u=e.map(e=>e.text??``).join(``);let t=e.filter(e=>!!e.toolUse);t.length&&(d=t.map((e,t)=>{let n=e.toolUse;if(!n.name)throw new p(`Bedrock returned a toolUse block without a name at index ${t}.`,`validation`);return{id:n.toolUseId??`${n.name}_${t}`,type:`function`,function:{name:n.name,arguments:JSON.stringify(n.input??{})}}}))}return{choices:[{message:{content:u,...d?{tool_calls:d}:{}}}],usage:{prompt_tokens:l.usage?.inputTokens,completion_tokens:l.usage?.outputTokens,total_tokens:l.usage?.totalTokens}}},async*createStream(e,t){if(!n.converseStream)throw new p(`stream: true requires a Bedrock client with converseStream`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`converseStream`}});let{request:s,toolName:c}=$n(e,r,i,a,o),{stream:l}=await n.converseStream(s,t),u=new Map;for await(let e of l)if(`contentBlockStart`in e){let{contentBlockIndex:t,start:n}=e.contentBlockStart;if(n?.toolUse){let e=n.toolUse.name===c?`json-tool`:`tool_use`;u.set(t,e),e===`tool_use`&&!c&&(yield{type:`tool_call_delta`,index:t,id:n.toolUse.toolUseId,name:n.toolUse.name})}else u.set(t,`text`)}else if(`contentBlockDelta`in e){let{contentBlockIndex:t,delta:n}=e.contentBlockDelta;n&&`text`in n&&n.text!==void 0&&!c?yield{type:`text-delta`,delta:n.text}:n&&`toolUse`in n&&n.toolUse?.input!==void 0&&(u.get(t)===`json-tool`?yield{type:`text-delta`,delta:n.toolUse.input}:c||(yield{type:`tool_call_delta`,index:t,argumentsDelta:n.toolUse.input}))}else if(`metadata`in e&&e.metadata.usage)yield{type:`usage`,usage:{prompt_tokens:e.metadata.usage.inputTokens,completion_tokens:e.metadata.usage.outputTokens,total_tokens:e.metadata.usage.totalTokens}};else if(`throttlingException`in e)throw new p(e.throttlingException.message??`Bedrock throttled the request mid-stream`,`api`,{status:429,code:`provider_rate_limited`});else if(`validationException`in e)throw new p(e.validationException.message??`Bedrock rejected the request mid-stream`,`validation`);else if(`internalServerException`in e||`serviceUnavailableException`in e||`modelStreamErrorException`in e){let t=`internalServerException`in e&&e.internalServerException.message||`serviceUnavailableException`in e&&e.serviceUnavailableException.message||`modelStreamErrorException`in e&&e.modelStreamErrorException.message||`Bedrock reported a mid-stream error`,n=`modelStreamErrorException`in e&&e.modelStreamErrorException.originalStatusCode||`serviceUnavailableException`in e&&503||500;throw new p(t,`api`,{status:n,code:n>=500?`server_error`:void 0})}}}}}}function ar(e){if(!e||e===`auto`)return{auto:{}};if(e===`required`)return{any:{}};if(e===`none`)throw new p("'none' is not supported by fromBedrock: Bedrock Converse has no `tool_choice` equivalent to forbidding tool use while tools are still offered. Omit `tools` entirely for this call instead.",`invalid_params`,{code:`unsupported_capability`,issues:{capability:`toolChoice: 'none'`}});return{tool:{name:e.function.name}}}function or(e){if(e.role===`tool`)return{role:`user`,content:[{toolResult:{toolUseId:e.tool_call_id,content:[{text:e.content}],status:e.is_error?`error`:`success`}}]};if(e.role===`assistant`&&e.tool_calls?.length){let t=[];e.content&&t.push({text:e.content});for(let n of e.tool_calls){let e;if(!n.function.arguments.trim())e={};else try{e=JSON.parse(n.function.arguments)}catch(e){throw new p(`Assistant tool call "${n.function.name}" (${n.id}) has arguments that are not valid JSON.`,`validation`,{cause:e})}t.push({toolUse:{toolUseId:n.id,name:n.function.name,input:e}})}return{role:`assistant`,content:t}}return{role:e.role,content:Array.isArray(e.content)?Zn(e.content):[{text:e.content??``}]}}function sr(e){let t=e=>e.role===`user`&&e.content.length>0&&e.content.every(e=>`toolResult`in e),n=[];for(let r of e){let e=n.at(-1);t(r)&&e&&t(e)?e.content.push(...r.content):n.push(r)}return n}async function*cr(e){let t=e.getReader();try{for(;;){let{done:e,value:n}=await t.read();if(e)return;n&&(yield n)}}finally{try{await t.cancel()}catch{}t.releaseLock()}}async function lr(e,t){let n=await fetch(e,t);if(!n.ok){let e=await n.text().catch(()=>``),t=Error(`Fetch adapter stream request failed (${n.status}): ${e.slice(0,500)}`);throw t.status=n.status,t.headers=n.headers,t}if(!n.body)throw Error(`Fetch adapter stream request received a response with no body.`);return{headers:n.headers,body:cr(n.body)}}async function ur(e,t,n){let r=typeof e.url==`function`?e.url(t):e.url,i=typeof e.headers==`function`?await e.headers():e.headers,a=e.method??`POST`,o=![`GET`,`HEAD`].includes(a.toUpperCase());return{url:r,method:a,headers:o?{"Content-Type":`application/json`,...i}:{...i},...o?{body:JSON.stringify(n)}:{}}}function dr(e){return{chat:{completions:{async create(t,n){let{url:r,method:i,headers:a,body:o}=await ur(e,t,e.mapRequest(t)),s=await(e.request??fetch)(r,{method:i,headers:a,body:o,signal:n.signal});if(!s.ok){let e=await s.text().catch(()=>``),t=Error(`Fetch adapter request failed (${s.status}): ${e.slice(0,500)}`);throw t.status=s.status,t.headers=s.headers,t}let c=await s.json(),{content:l,usage:u,toolCalls:d}=e.mapResponse(c),f=d?.length?d.map(e=>({id:e.id,type:`function`,function:{name:e.name,arguments:e.arguments}})):void 0,p={choices:[{message:{content:l,...f?{tool_calls:f}:{}}}],usage:u?{prompt_tokens:u.promptTokens,completion_tokens:u.completionTokens,total_tokens:u.totalTokens}:void 0},m=e.parseRateLimitHint??B;return s.headers&&typeof s.headers.get==`function`&<(p,m(s.headers)),p},async*createStream(t,n){if(!e.mapStreamEvent)throw new p(`stream: true requires mapStreamEvent to be configured on fromFetch`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`mapStreamEvent`}});if(e.request&&!e.requestStream)throw new p("`stream: true` requires `requestStream` to be configured on fromFetch when a custom `request` transport is set. `requestStream` does not fall back to `request` (it needs an async-iterable byte stream, which `RequestLike`'s buffered `ResponseLike` has no way to provide), without it, `stream: true` would silently use plain native `fetch` instead of your configured transport. Add a `requestStream` that opens the same connection your `request` does, or omit `request` if native `fetch` is fine for both.",`invalid_params`,{code:`unsupported_capability`,issues:{capability:`requestStream`}});let{url:r,method:i,headers:a,body:o}=await ur(e,t,e.mapRequest(t)),s=e.parseStreamFrames??pn,c;if(e.requestStream)c=await e.requestStream(r,{method:i,headers:a,body:o,signal:n.signal});else{let t=await lr(r,{method:i,headers:a,body:o,signal:n.signal});if(c=t.body,t.headers&&typeof t.headers.get==`function`){let n=(e.parseRateLimitHint??B)(t.headers);(n.remainingRequests!==void 0||n.limitRequests!==void 0)&&(yield{type:`rate_limit_hint`,hint:n})}}for await(let t of s(c)){if(t===q){yield{type:`ping`};continue}let n=e.mapStreamEvent(t);n&&(Array.isArray(n)?yield*n:yield n)}}}}}}function fr(e,t){if(e.budget_tokens===void 0)return e;let{budget_tokens:n,...r}=e;return r.reasoning_effort===void 0?{...r,reasoning_effort:Q(n,t)}:r}function pr(e){return e.map(e=>e.type===`image`?{type:`image_url`,image_url:{url:`data:${Y(e.mimeType)};base64,${e.data}`}}:{type:`text`,text:e.text})}function mr(e){return e.messages.map(e=>{if(e.role===`user`&&Array.isArray(e.content))return{...e,content:pr(e.content)};if(e.role===`tool`){let{is_error:t,...n}=e;return n}return e})}function*hr(e){let t=e.choices?.[0]?.delta;if(t?.content&&(yield{type:`text-delta`,delta:t.content}),t?.tool_calls?.length)for(let e of t.tool_calls)yield{type:`tool_call_delta`,index:e.index,id:e.id,name:e.function?.name,argumentsDelta:e.function?.arguments};e.usage&&(yield{type:`usage`,usage:e.usage})}function $(e,t={}){let n=e,{supportsStreamUsage:r=!0,supportsWithResponse:i=!1}=t,a=Z(t.reasoningEffortTokens),o=n.chat.completions.create.bind(n.chat.completions);return{chat:{completions:{async create(e,t){let r=mr(e),o=fr({...e,messages:r},a);if(!i)return n.chat.completions.create(o,t);let{data:s,response:c}=await n.chat.completions.create(o,t).withResponse();return s&&typeof s==`object`&<(s,B(c.headers)),s},async*createStream(e,t){let s=mr(e),c=fr({...e,messages:s,stream:!0,...r?{stream_options:{include_usage:!0}}:{}},a),l;if(!i)l=await o(c,t);else{let{data:e,response:r}=await n.chat.completions.create(c,t).withResponse();l=e;let i=B(r.headers);(i.remainingRequests!==void 0||i.limitRequests!==void 0)&&(yield{type:`rate_limit_hint`,hint:i})}for await(let e of l)yield*hr(e)}}}}}const gr=$,_r=$,vr=$,yr=$,br=$,xr=$,Sr=$,Cr=$,wr=$,Tr=$,Er=$,Dr=$,Or=$,kr=$,Ar=$,jr=$,Mr=$,Nr=$,Pr=$,Fr=$,Ir=$,Lr=$,Rr=$,zr=$,Br=$,Vr=$,Hr=$,Ur=$,Wr=$,Gr=$,Kr=$,qr=$,Jr=$,Yr=$,Xr=$,Zr=$,Qr=$,$r=$;exports.CircuitBreaker=ze,exports.ConsecutiveTripping=Ne,exports.ConsoleLogger=ln,exports.FallbackExhaustedError=w,exports.InMemoryCacheAdapter=b,exports.LLMError=p,exports.NormalizedCacheAdapter=x,exports.RateLimiter=Qt,exports.RetryBudget=Be,exports.RollingTripping=Pe,exports.SSE_PING=q,exports.TieredCacheAdapter=S,exports.VernLLM=un,exports.createMiddleware=oe,exports.createMiddlewareStateBag=T,exports.createStateKey=ae,exports.defaultEstimateTokens=Jt,exports.defaultFallbackOn=ne,exports.defineCachedCallParams=fn,exports.defineCallParams=dn,exports.defineTool=C,exports.from01AI=$r,exports.fromAnthropic=Fn,exports.fromAnyscale=Jr,exports.fromAtlasCloud=Qr,exports.fromBaseten=zr,exports.fromBedrock=ir,exports.fromCerebras=br,exports.fromCloudflareWorkersAI=Ir,exports.fromDeepInfra=Er,exports.fromDeepSeek=yr,exports.fromFeatherless=Br,exports.fromFetch=dr,exports.fromFireworks=Sr,exports.fromFriendli=Vr,exports.fromGemini=qn,exports.fromGroq=_r,exports.fromHyperbolic=Or,exports.fromInferenceNet=Xr,exports.fromInfermatic=Zr,exports.fromLMStudio=jr,exports.fromLambdaLabs=Kr,exports.fromLepton=Yr,exports.fromMiniMax=Gr,exports.fromMistral=vr,exports.fromMoonshot=kr,exports.fromNebius=Lr,exports.fromNovita=Dr,exports.fromNvidiaNIM=Pr,exports.fromOllama=Cr,exports.fromOpenAI=gr,exports.fromOpenAICompatible=$,exports.fromOpenRouter=wr,exports.fromParasail=Ur,exports.fromPerplexity=Tr,exports.fromSambaNova=Rr,exports.fromSiliconFlow=Hr,exports.fromSnowflakeCortex=qr,exports.fromStepFun=Wr,exports.fromTogether=xr,exports.fromVLLM=Mr,exports.fromVercelAIGateway=Fr,exports.fromXAI=Nr,exports.fromZhipu=Ar,exports.hasIssues=h,exports.isFallbackExhaustedError=re,exports.isLLMError=m,exports.isToolCallResult=ee,exports.metaRef=ie,exports.parseSseStream=pn;
|
|
14
|
+
`);if(r===`[DONE]`)return G;try{return JSON.parse(r)}catch(e){throw new p(`Invalid JSON in SSE frame: ${r.slice(0,200)}`,`parse`,{cause:e,code:`stream_frame_invalid`})}}const _n=[`image/png`,`image/jpeg`,`image/gif`,`image/webp`];function Y(e){if(_n.includes(e))return e;throw new p(`Unsupported image mimeType "${e}": expected one of ${_n.join(`, `)}`,`invalid_params`)}function vn(e,t){return t?Array.isArray(t)?t.includes(e):t(e):!1}const X={minimal:1024,low:4096,medium:16e3,high:32e3};function Z(e){if(!e)return X;let t={...X,...e};if(!(t.minimal<t.low&&t.low<t.medium&&t.medium<t.high))throw new p(`reasoningEffortTokens must keep tiers in strictly ascending order (minimal < low < medium < high), got ${JSON.stringify(t)}. An out-of-order override doesn't just misrank tiers, it can make some of them unreachable.`,`invalid_params`);return t}function yn(e,t=X){return t[e]}function Q(e,t=X){return e<=t.minimal?`minimal`:e<=t.low?`low`:e<=t.medium?`medium`:`high`}function bn(e){let t=/opus-(\d+)(?:-(\d+))?/.exec(e);if(!t)return null;let n=t[2],r=n===void 0||n.length>=8?0:Number(n);return[Number(t[1]),r]}function xn(e){let t=bn(e);if(t){let[e,n]=t;return e>4||e===4&&n>=7}return[`sonnet-5`,`fable-5`,`mythos`].some(t=>e.includes(t))}function Sn(e,t){return xn(e)?!0:t?Array.isArray(t)?t.includes(e):t(e):!1}function Cn(e,t){return!Sn(e,t)}function wn(e,t){if(e<1024)throw new p(`budgetTokens (${e}) is below Anthropic's minimum of 1024. Raise budgetTokens, or use a reasoningEffort tier of 'low' or above with the default conversion table.`,`invalid_params`);if(e>=t)throw new p(`budgetTokens (${e}) must be less than maxTokens (${t}); the thinking budget and the reply share the same max_tokens ceiling on Anthropic. Raise maxTokens, or lower budgetTokens/reasoningEffort.`,`invalid_params`)}function Tn(e){if(e)throw new p(`budgetTokens/reasoningEffort was set alongside ${e}. Anthropic rejects thinking combined with a tool_choice that forces tool use, the model has to be able to reply with plain text for thinking to run. Use toolChoice: 'auto' (or omit toolChoice) for this call, or drop budgetTokens/reasoningEffort for it.`,`invalid_params`)}function En(e){return e===`minimal`?`low`:e}function Dn(e,t){return kn(t,e.toUpperCase())}function On(e){let t=/gemini-\d+\.(\d+)/.exec(e);return t?Number(t[1]):0}function kn(e,t){if(!e.includes(`pro`))return t;let n=An(e);return n===null||n<3?t:On(e)===0?t===`HIGH`?`HIGH`:`LOW`:t===`MINIMAL`?`LOW`:t}function An(e){let t=/gemini-(\d+)/.exec(e);return t?Number(t[1]):null}function jn(e){let t=An(e);return t!==null&&t>=3}function Mn(e,t){return jn(e)?!0:t?Array.isArray(t)?t.includes(e):t(e):!1}function Nn(e){return e.map(e=>e.type===`image`?{type:`image`,source:{type:`base64`,media_type:Y(e.mimeType),data:e.data}}:{type:`text`,text:e.text})}function Pn(e,t){if(e.type!==`object`)throw new p(`Tool "${t}"'s schema must have "type": "object" (Anthropic requires object-shaped tool parameters).`,`validation`);return e}function Fn(e){return!e||e===`auto`?{type:`auto`}:e===`none`?{type:`none`}:e===`required`?{type:`any`}:{type:`tool`,name:e.function.name}}function In(e,t){return{tools:e.map(e=>({name:e.function.name,description:e.function.description,input_schema:Pn(e.function.parameters,e.function.name)})),toolChoice:Fn(t)}}function Ln(e,t,n,r){let i=e.messages.find(e=>e.role===`system`),a=e.messages.filter(e=>e.role===`user`||e.role===`assistant`||e.role===`tool`),o=e.response_format?.type===`json_schema`?e.response_format.json_schema:void 0,s=o?.name.trim();if(o&&!s)throw new p(`json_schema.name must not be empty.`,`validation`);let c=!!o&&vn(e.model,t);if(o&&e.tools?.length&&!c)throw new p(`Anthropic model "${e.model}" is not covered by nativeStructuredOutputModels, so \`jsonSchema\` is emulated as a forced single tool call there, which collides with the \`tools\` you also provided. Either drop \`tools\` or \`jsonSchema\` for this call, or pass this model in fromAnthropic's \`nativeStructuredOutputModels\` option once you've confirmed it supports Anthropic's \`output_config.format\`.`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`tools_with_json_schema`}});if(e.response_format?.type===`json_object`)throw new p('response_format: "json_object" is not supported on Anthropic. Unlike OpenAI, Anthropic has no API-level field that mechanically guarantees valid JSON output for this mode, so it used to be emulated by injecting a "respond with JSON only" instruction into the system prompt, a guarantee this adapter can no longer make. Use `jsonSchema` instead, which maps to a real API-level constraint (Anthropic\'s native output_config.format on covered models, or a forced single tool call otherwise).',`validation`);let l,u,d,f;if(o&&c)u={type:`json_schema`,schema:o.schema},e.tools?.length&&({tools:d,toolChoice:f}=In(e.tools,e.tool_choice));else if(o&&s){let{schema:e,description:t,strict:n}=o;l=s,d=[{name:l,description:t,input_schema:Pn(e,l),strict:n}],f={type:`tool`,name:l}}!o&&e.tools?.length&&({tools:d,toolChoice:f}=In(e.tools,e.tool_choice));let m,h;if(e.budget_tokens!==void 0||e.reasoning_effort!==void 0){if(Tn(f?.type===`tool`?`toolChoice forcing the "${f.name}" tool`:f?.type===`any`?`toolChoice: 'required' (Anthropic's "any" tool_choice)`:void 0),Cn(e.model,r)){let t=e.budget_tokens??yn(e.reasoning_effort,n);wn(t,e.max_tokens),m={type:`enabled`,budget_tokens:t}}else{let t=e.reasoning_effort??Q(e.budget_tokens,n);m={type:`adaptive`},h=En(t)}}let g=i?.content,_=m?void 0:e.temperature;return{body:{model:e.model,max_tokens:e.max_tokens,..._===void 0?{}:{temperature:_},system:g||void 0,messages:zn(a.map(e=>Bn(e))),...d?{tools:d,tool_choice:f}:{},...u||h?{output_config:{...u?{format:u}:{},...h?{effort:h}:{}}}:{},...m?{thinking:m}:{}},toolName:l}}function Rn(e,t){let n=t?.nativeStructuredOutputModels,r=Z(t?.reasoningEffortTokens),i=t?.adaptiveOnlyModels,a=t?.supportsWithResponse??!1,o=e.messages.create.bind(e.messages);return{supportsJsonObjectMode:!1,chat:{completions:{async create(t,o){let{body:s,toolName:c}=Ln(t,n,r,i),l,u=a?await(async()=>{let{data:t,response:n}=await e.messages.create(s,o).withResponse();return l=n.headers,t})():await e.messages.create(s,o),d,f;if(c){let e=u.content.find(e=>e.type===`tool_use`&&e.name===c);if(!e)throw new p(`Anthropic did not return the required structured output tool "${c}".`,`validation`);if(!e.input||typeof e.input!=`object`||Array.isArray(e.input))throw new p(`Anthropic returned invalid structured output for tool "${c}". Expected an object.`,`validation`);d=JSON.stringify(e.input)}else{d=u.content.filter(e=>e.type===`text`).map(e=>e.text??``).join(``);let e=u.content.filter(e=>e.type===`tool_use`);e.length&&(f=e.map(e=>({id:e.id,type:`function`,function:{name:e.name,arguments:JSON.stringify(e.input??{})}})))}let m={choices:[{message:{content:d,...f?{tool_calls:f}:{}}}],usage:{prompt_tokens:u.usage?.input_tokens,completion_tokens:u.usage?.output_tokens,total_tokens:(u.usage?.input_tokens??0)+(u.usage?.output_tokens??0),...u.usage?.output_tokens_details?.thinking_tokens===void 0?{}:{completion_tokens_details:{reasoning_tokens:u.usage.output_tokens_details.thinking_tokens}}}};return l&&ft(m,ut(l)),m},async*createStream(t,s){let{body:c,toolName:l}=Ln(t,n,r,i),u;if(!a)u=await o({...c,stream:!0},s);else{let{data:t,response:n}=await e.messages.create({...c,stream:!0},s).withResponse();u=t;let r=ut(n.headers);(r.remainingRequests!==void 0||r.limitRequests!==void 0)&&(yield{type:`rate_limit_hint`,hint:r})}let d=new Map,f=0,m=!1;for await(let e of u)if(e.type===`message_start`)f=e.message.usage?.input_tokens??0;else if(e.type===`content_block_start`){if(e.content_block.type===`tool_use`){let t=e.content_block.name===l?`json-tool`:`tool_use`;d.set(e.index,t),t===`json-tool`?m=!0:l||(yield{type:`tool_call_delta`,index:e.index,id:e.content_block.id,name:e.content_block.name})}else d.set(e.index,`text`)}else if(e.type===`content_block_delta`)e.delta.type===`text_delta`?l||(yield{type:`text-delta`,delta:e.delta.text}):e.delta.type===`input_json_delta`&&(d.get(e.index)===`json-tool`?yield{type:`text-delta`,delta:e.delta.partial_json}:l||(yield{type:`tool_call_delta`,index:e.index,argumentsDelta:e.delta.partial_json}));else if(e.type===`message_delta`){let t=e.usage?.output_tokens??0,n=e.usage?.output_tokens_details?.thinking_tokens;yield{type:`usage`,usage:{prompt_tokens:f,completion_tokens:t,total_tokens:f+t,...n===void 0?{}:{completion_tokens_details:{reasoning_tokens:n}}}}}else e.type===`ping`&&(yield{type:`ping`});if(l&&!m)throw new p(`Anthropic did not return the required structured output tool "${l}".`,`validation`)}}}}}function zn(e){let t=e=>e.role===`user`&&Array.isArray(e.content)&&e.content.length>0&&e.content.every(e=>e.type===`tool_result`),n=[];for(let r of e){let e=n.at(-1);t(r)&&e&&t(e)?e.content.push(...r.content):n.push(r)}return n}function Bn(e){if(e.role===`tool`)return{role:`user`,content:[{type:`tool_result`,tool_use_id:e.tool_call_id,content:e.content,...e.is_error?{is_error:!0}:{}}]};if(e.role===`assistant`&&e.tool_calls?.length){let t=[];e.content&&t.push({type:`text`,text:e.content});for(let n of e.tool_calls){let e;try{e=n.function.arguments.trim()?JSON.parse(n.function.arguments):{}}catch(e){throw new p(`Assistant tool call "${n.function.name}" (${n.id}) has arguments that are not valid JSON.`,`validation`,{cause:e})}if(e===null||Array.isArray(e)||typeof e!=`object`)throw new p(`Assistant tool call "${n.function.name}" (${n.id}) arguments must be a JSON object.`,`validation`);t.push({type:`tool_use`,id:n.id,name:n.function.name,input:e})}return{role:`assistant`,content:t}}return{role:e.role,content:Array.isArray(e.content)?Nn(e.content):e.content??``}}function Vn(e){return e.map(e=>e.type===`image`?{inlineData:{mimeType:Y(e.mimeType),data:e.data}}:{text:e.text})}function Hn(e,t){return`${e}#${t}`}function Un(e){return!e||e===`auto`?{functionCallingConfig:{mode:`AUTO`}}:e===`none`?{functionCallingConfig:{mode:`NONE`}}:e===`required`?{functionCallingConfig:{mode:`ANY`}}:{functionCallingConfig:{mode:`ANY`,allowedFunctionNames:[e.function.name]}}}function Wn(e){let t=new Map;for(let n of e)if(n.role===`assistant`)for(let e of n.tool_calls??[])t.set(e.id,e.function.name);return t}function Gn(e,t){if(e.role===`tool`)return{role:`user`,parts:[{functionResponse:{id:e.tool_call_id,name:t.get(e.tool_call_id)??e.tool_call_id,response:qn(e.content)}}]};if(e.role===`assistant`&&e.tool_calls?.length){let t=[];return typeof e.content==`string`&&e.content&&t.push({text:e.content}),t.push(...e.tool_calls.map(e=>({functionCall:{id:e.id,name:e.function.name,args:Kn(e.function.arguments,e.function.name)}}))),{role:`model`,parts:t}}return{role:e.role===`assistant`?`model`:`user`,parts:Array.isArray(e.content)?Vn(e.content):[{text:e.content??``}]}}function Kn(e,t){let n;try{n=e.trim()?JSON.parse(e):{}}catch(e){throw new p(`Tool call "${t}" arguments are not valid JSON.`,`parse`,{cause:e,code:`tool_arguments_parse_failed`})}if(!n||Array.isArray(n)||typeof n!=`object`)throw new p(`Tool call "${t}" arguments must be a JSON object.`,`validation`);return n}function qn(e){let t;try{t=e.trim()?JSON.parse(e):``}catch{t=e}return t&&!Array.isArray(t)&&typeof t==`object`?t:{output:t}}function Jn(e){let t=e=>e.role===`user`&&e.parts.length>0&&e.parts.every(e=>`functionResponse`in e),n=[];for(let r of e){let e=n.at(-1);t(r)&&e&&t(e)?e.parts.push(...r.parts):n.push(r)}return n}function Yn(e,t,n){let r=e.messages.find(e=>e.role===`system`),i=e.messages.filter(e=>e.role===`user`||e.role===`assistant`||e.role===`tool`),a=!!e.response_format,o={...e.temperature===void 0?{}:{temperature:e.temperature},maxOutputTokens:e.max_tokens,...r?{systemInstruction:{parts:[{text:r.content}]}}:{}};if(a&&(o.responseMimeType=`application/json`),e.response_format?.type===`json_schema`){let{schema:t,description:n}=e.response_format.json_schema;o.responseSchema={...t,...n?{description:n}:{}}}if(e.tools?.length&&(o.tools=[{functionDeclarations:e.tools.map(e=>({name:e.function.name,description:e.function.description,parameters:e.function.parameters}))}],o.toolConfig=Un(e.tool_choice)),Mn(e.model,n)){let n=e.reasoning_effort??(e.budget_tokens===void 0?void 0:Q(e.budget_tokens,t));n!==void 0&&(o.thinkingConfig={thinkingLevel:Dn(n,e.model)})}else{let n=e.budget_tokens??(e.reasoning_effort?yn(e.reasoning_effort,t):void 0);n!==void 0&&(o.thinkingConfig={thinkingBudget:n})}let s=Wn(i);return{model:e.model,contents:Jn(i.map(e=>Gn(e,s))),config:o}}function Xn(e,t){let n=Z(t?.reasoningEffortTokens),r=t?.thinkingLevelModels,i=e.models??e;if(typeof i.generateContent!=`function`)throw new p(`fromGemini requires a client with generateContent: pass ai.models, or the whole ai client (fromGemini(ai)).`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`generateContent`}});let a=i.generateContent.bind(i),o=typeof i.generateContentStream==`function`?i.generateContentStream.bind(i):void 0;return{chat:{completions:{async create(e,t){let i=Yn(e,n,r);i.config={...i.config,abortSignal:t.signal};let o=await a(i),s=o.candidates?.[0]?.content?.parts??[],c=s.map(e=>e.text??``).join(``),l=s.filter(e=>e.functionCall),u;if(l.length){let e=new Map;u=l.map(t=>{let n=t.functionCall.name,r=t.functionCall.id,i=e.get(n)??0;return e.set(n,i+1),{id:r??Hn(n,i),type:`function`,function:{name:n,arguments:JSON.stringify(t.functionCall.args??{})}}})}return{choices:[{message:{content:c,...u?{tool_calls:u}:{}}}],usage:{prompt_tokens:o.usageMetadata?.promptTokenCount,completion_tokens:o.usageMetadata?.candidatesTokenCount,total_tokens:o.usageMetadata?.totalTokenCount,...o.usageMetadata?.thoughtsTokenCount===void 0?{}:{completion_tokens_details:{reasoning_tokens:o.usageMetadata.thoughtsTokenCount}}}}},async*createStream(e,t){if(!o)throw new p(`stream: true requires a Gemini client with generateContentStream`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`generateContentStream`}});let i=Yn(e,n,r);i.config={...i.config,abortSignal:t.signal};let a=await o(i),s=0,c=new Map,l;for await(let e of a){let t=e.candidates?.[0]?.content?.parts??[];for(let e of t)if(e.text&&(yield{type:`text-delta`,delta:e.text}),e.functionCall){let t=e.functionCall.name,n=e.functionCall.id,r=t?c.get(t)??0:0;t&&c.set(t,r+1),yield{type:`tool_call_delta`,index:s,id:n??(t?Hn(t,r):void 0),name:e.functionCall.name,argumentsDelta:JSON.stringify(e.functionCall.args??{}),complete:!0},s++}e.usageMetadata&&(l=e.usageMetadata)}l&&(yield{type:`usage`,usage:{prompt_tokens:l.promptTokenCount,completion_tokens:l.candidatesTokenCount,total_tokens:l.totalTokenCount,...l.thoughtsTokenCount===void 0?{}:{completion_tokens_details:{reasoning_tokens:l.thoughtsTokenCount}}}})}}}}}function Zn(e){return e.includes(`claude`)}function Qn(e){switch(Y(e)){case`image/png`:return`png`;case`image/jpeg`:return`jpeg`;case`image/gif`:return`gif`;case`image/webp`:return`webp`}}function $n(e){return new Uint8Array(Buffer.from(e,`base64`))}function er(e){return e.map(e=>e.type===`image`?{image:{format:Qn(e.mimeType),source:{bytes:$n(e.data)}}}:{text:e.text})}function tr(e,t){return{tools:e.map(e=>({toolSpec:{name:e.function.name,description:e.function.description,inputSchema:{json:e.function.parameters}}})),toolChoice:cr(t)}}function nr(e,t,n,r,i){let a=e.messages.find(e=>e.role===`system`),o=e.messages.filter(e=>e.role===`user`||e.role===`assistant`||e.role===`tool`),s=e.response_format?.type===`json_schema`?e.response_format.json_schema:void 0,c=s?.name.trim();if(s&&!c)throw new p(`json_schema.name must not be empty.`,`validation`);let l=!!s&&vn(e.model,n);if(s&&e.tools?.length&&!l)throw new p(`Bedrock model "${e.model}" is not covered by nativeStructuredOutputModels, so \`jsonSchema\` is emulated as a forced single tool call there (via \`toolConfig\`), which collides with the \`tools\` you also provided. Either drop \`tools\` or \`jsonSchema\` for this call, or pass this model in fromBedrock's \`nativeStructuredOutputModels\` option once you've confirmed it supports Converse's \`outputConfig.textFormat\`.`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`tools_with_json_schema`}});if(e.response_format?.type===`json_object`)throw new p('response_format: "json_object" is not supported on Bedrock. Converse has no field that mechanically guarantees valid JSON output for this mode, so it used to be emulated by injecting a "respond with JSON only" instruction into the system prompt, a guarantee this adapter can no longer make. Use `jsonSchema` instead, which maps to a real constraint (Converse\'s native outputConfig.textFormat on covered models, or a forced tool call otherwise).',`validation`);let u,d,f;if(s&&l){let{schema:e,description:t}=s;f={textFormat:{type:`json_schema`,structure:{jsonSchema:{schema:JSON.stringify(e),name:c,description:t}}}}}else if(s&&c){let{schema:e,description:t,strict:n}=s;u=c,d={tools:[{toolSpec:{name:u,description:t,inputSchema:{json:e},strict:n}}],toolChoice:{tool:{name:u}}}}if(e.tools?.length&&!u&&(d=tr(e.tools,e.tool_choice)),s&&d&&t&&!(Array.isArray(t)?t.includes(e.model):t(e.model)))throw new p(`Bedrock model "${e.model}" is not listed in toolUseSupportedModels, but this call requires Converse tool use (either jsonSchema emulated as a forced tool call, or real \`tools\` sent alongside native structured output).`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`toolUseSupportedModels`}});let m,h;if(Zn(e.model)&&(e.budget_tokens!==void 0||e.reasoning_effort!==void 0)){let t=d?.toolChoice;if(Tn(t&&`tool`in t?`toolChoice forcing the "${t.tool?.name}" tool`:t&&`any`in t?`toolChoice: 'required' (Converse's "any" tool_choice)`:void 0),Cn(e.model,i)){let t=e.budget_tokens??yn(e.reasoning_effort,r);wn(t,e.max_tokens),m={thinking:{type:`enabled`,budget_tokens:t}}}else{let t=e.reasoning_effort??Q(e.budget_tokens,r);m={thinking:{type:`adaptive`}},h=En(t)}}let g=m?void 0:e.temperature;return{request:{modelId:e.model,messages:ur(o.map(e=>lr(e))),system:a?.content?[{text:a.content}]:void 0,inferenceConfig:{...g===void 0?{}:{temperature:g},maxTokens:e.max_tokens},...d?{toolConfig:d}:{},...f||h?{outputConfig:{...f??{},...h?{effort:h}:{}}}:{},...m?{additionalModelRequestFields:m}:{}},toolName:u}}function rr(e){return typeof e.send==`function`}function ir(e){if(`messageStart`in e)return{messageStart:e.messageStart};if(`contentBlockStart`in e)return{contentBlockStart:e.contentBlockStart};if(`contentBlockDelta`in e)return{contentBlockDelta:e.contentBlockDelta};if(`contentBlockStop`in e)return{contentBlockStop:e.contentBlockStop};if(`messageStop`in e)return{messageStop:e.messageStop};if(`metadata`in e)return{metadata:e.metadata};if(`internalServerException`in e)return{internalServerException:e.internalServerException};if(`modelStreamErrorException`in e)return{modelStreamErrorException:e.modelStreamErrorException};if(`validationException`in e)return{validationException:e.validationException};if(`throttlingException`in e)return{throttlingException:e.throttlingException};if(`serviceUnavailableException`in e)return{serviceUnavailableException:e.serviceUnavailableException}}async function*ar(e){for await(let t of e){let e=ir(t);e&&(yield e)}}function or(e){let t;function n(){return t??=import(`@aws-sdk/client-bedrock-runtime`).then(e=>e,e=>{throw t=void 0,new p(`fromBedrock requires "@aws-sdk/client-bedrock-runtime" to be installed to use a raw AWS SDK client (it is not a dependency of vern-llm itself). Install it, or pass your own object with .converse()/.converseStream() methods instead.`,`validation`,{cause:e})}),t}return{converse:async(t,r)=>{let{ConverseCommand:i}=await n();return e.send(new i(t),{abortSignal:r.signal})},converseStream:async(t,r)=>{let{ConverseStreamCommand:i}=await n(),a=await e.send(new i(t),{abortSignal:r.signal});if(!a.stream)throw new p(`Bedrock ConverseStreamCommand response did not include a stream. This can happen if the request or the model doesn't actually support Converse streaming.`,`api`,{code:`server_error`});return{stream:ar(a.stream)}}}}function sr(e,t){let n=rr(e)?or(e):e,r=t?.toolUseSupportedModels,i=t?.nativeStructuredOutputModels,a=Z(t?.reasoningEffortTokens),o=t?.adaptiveOnlyModels;return{supportsJsonObjectMode:!1,chat:{completions:{async create(e,t){let{request:s,toolName:c}=nr(e,r,i,a,o),l=await n.converse(s,t),u,d;if(c){let e=l.output?.message?.content?.find(e=>e.toolUse?.name===c);u=e?.toolUse?JSON.stringify(e.toolUse.input):``}else{let e=l.output?.message?.content??[];u=e.map(e=>e.text??``).join(``);let t=e.filter(e=>!!e.toolUse);t.length&&(d=t.map((e,t)=>{let n=e.toolUse;if(!n.name)throw new p(`Bedrock returned a toolUse block without a name at index ${t}.`,`validation`);return{id:n.toolUseId??`${n.name}_${t}`,type:`function`,function:{name:n.name,arguments:JSON.stringify(n.input??{})}}}))}return{choices:[{message:{content:u,...d?{tool_calls:d}:{}}}],usage:{prompt_tokens:l.usage?.inputTokens,completion_tokens:l.usage?.outputTokens,total_tokens:l.usage?.totalTokens}}},async*createStream(e,t){if(!n.converseStream)throw new p(`stream: true requires a Bedrock client with converseStream`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`converseStream`}});let{request:s,toolName:c}=nr(e,r,i,a,o),{stream:l}=await n.converseStream(s,t),u=new Map;for await(let e of l)if(`contentBlockStart`in e){let{contentBlockIndex:t,start:n}=e.contentBlockStart;if(n?.toolUse){let e=n.toolUse.name===c?`json-tool`:`tool_use`;u.set(t,e),e===`tool_use`&&!c&&(yield{type:`tool_call_delta`,index:t,id:n.toolUse.toolUseId,name:n.toolUse.name})}else u.set(t,`text`)}else if(`contentBlockDelta`in e){let{contentBlockIndex:t,delta:n}=e.contentBlockDelta;n&&`text`in n&&n.text!==void 0&&!c?yield{type:`text-delta`,delta:n.text}:n&&`toolUse`in n&&n.toolUse?.input!==void 0&&(u.get(t)===`json-tool`?yield{type:`text-delta`,delta:n.toolUse.input}:c||(yield{type:`tool_call_delta`,index:t,argumentsDelta:n.toolUse.input}))}else if(`metadata`in e&&e.metadata.usage)yield{type:`usage`,usage:{prompt_tokens:e.metadata.usage.inputTokens,completion_tokens:e.metadata.usage.outputTokens,total_tokens:e.metadata.usage.totalTokens}};else if(`throttlingException`in e)throw new p(e.throttlingException.message??`Bedrock throttled the request mid-stream`,`api`,{status:429,code:`provider_rate_limited`});else if(`validationException`in e)throw new p(e.validationException.message??`Bedrock rejected the request mid-stream`,`validation`);else if(`internalServerException`in e||`serviceUnavailableException`in e||`modelStreamErrorException`in e){let t=`internalServerException`in e&&e.internalServerException.message||`serviceUnavailableException`in e&&e.serviceUnavailableException.message||`modelStreamErrorException`in e&&e.modelStreamErrorException.message||`Bedrock reported a mid-stream error`,n=`modelStreamErrorException`in e&&e.modelStreamErrorException.originalStatusCode||`serviceUnavailableException`in e&&503||500;throw new p(t,`api`,{status:n,code:n>=500?`server_error`:void 0})}}}}}}function cr(e){if(!e||e===`auto`)return{auto:{}};if(e===`required`)return{any:{}};if(e===`none`)throw new p("'none' is not supported by fromBedrock: Bedrock Converse has no `tool_choice` equivalent to forbidding tool use while tools are still offered. Omit `tools` entirely for this call instead.",`invalid_params`,{code:`unsupported_capability`,issues:{capability:`toolChoice: 'none'`}});return{tool:{name:e.function.name}}}function lr(e){if(e.role===`tool`)return{role:`user`,content:[{toolResult:{toolUseId:e.tool_call_id,content:[{text:e.content}],status:e.is_error?`error`:`success`}}]};if(e.role===`assistant`&&e.tool_calls?.length){let t=[];e.content&&t.push({text:e.content});for(let n of e.tool_calls){let e;if(!n.function.arguments.trim())e={};else try{e=JSON.parse(n.function.arguments)}catch(e){throw new p(`Assistant tool call "${n.function.name}" (${n.id}) has arguments that are not valid JSON.`,`validation`,{cause:e})}t.push({toolUse:{toolUseId:n.id,name:n.function.name,input:e}})}return{role:`assistant`,content:t}}return{role:e.role,content:Array.isArray(e.content)?er(e.content):[{text:e.content??``}]}}function ur(e){let t=e=>e.role===`user`&&e.content.length>0&&e.content.every(e=>`toolResult`in e),n=[];for(let r of e){let e=n.at(-1);t(r)&&e&&t(e)?e.content.push(...r.content):n.push(r)}return n}async function*dr(e){let t=e.getReader();try{for(;;){let{done:e,value:n}=await t.read();if(e)return;n&&(yield n)}}finally{try{await t.cancel()}catch{}t.releaseLock()}}async function fr(e,t){let n=await fetch(e,t);if(!n.ok){let e=await n.text().catch(()=>``),t=Error(`Fetch adapter stream request failed (${n.status}): ${e.slice(0,500)}`);throw t.status=n.status,t.headers=n.headers,t}if(!n.body)throw Error(`Fetch adapter stream request received a response with no body.`);return{headers:n.headers,body:dr(n.body)}}async function pr(e,t,n){let r=typeof e.url==`function`?e.url(t):e.url,i=typeof e.headers==`function`?await e.headers():e.headers,a=e.method??`POST`,o=![`GET`,`HEAD`].includes(a.toUpperCase());return{url:r,method:a,headers:o?{"Content-Type":`application/json`,...i}:{...i},...o?{body:JSON.stringify(n)}:{}}}function mr(e){return{chat:{completions:{async create(t,n){let{url:r,method:i,headers:a,body:o}=await pr(e,t,e.mapRequest(t)),s=await(e.request??fetch)(r,{method:i,headers:a,body:o,signal:n.signal});if(!s.ok){let e=await s.text().catch(()=>``),t=Error(`Fetch adapter request failed (${s.status}): ${e.slice(0,500)}`);throw t.status=s.status,t.headers=s.headers,t}let c=await s.json(),{content:l,usage:u,toolCalls:d}=e.mapResponse(c),f=d?.length?d.map(e=>({id:e.id,type:`function`,function:{name:e.name,arguments:e.arguments}})):void 0,p={choices:[{message:{content:l,...f?{tool_calls:f}:{}}}],usage:u?{prompt_tokens:u.promptTokens,completion_tokens:u.completionTokens,total_tokens:u.totalTokens}:void 0},m=e.parseRateLimitHint??B;return s.headers&&typeof s.headers.get==`function`&&ft(p,m(s.headers)),p},async*createStream(t,n){if(!e.mapStreamEvent)throw new p(`stream: true requires mapStreamEvent to be configured on fromFetch`,`invalid_params`,{code:`unsupported_capability`,issues:{capability:`mapStreamEvent`}});if(e.request&&!e.requestStream)throw new p("`stream: true` requires `requestStream` to be configured on fromFetch when a custom `request` transport is set. `requestStream` does not fall back to `request` (it needs an async-iterable byte stream, which `RequestLike`'s buffered `ResponseLike` has no way to provide), without it, `stream: true` would silently use plain native `fetch` instead of your configured transport. Add a `requestStream` that opens the same connection your `request` does, or omit `request` if native `fetch` is fine for both.",`invalid_params`,{code:`unsupported_capability`,issues:{capability:`requestStream`}});let{url:r,method:i,headers:a,body:o}=await pr(e,t,e.mapRequest(t)),s=e.parseStreamFrames??gn,c;if(e.requestStream)c=await e.requestStream(r,{method:i,headers:a,body:o,signal:n.signal});else{let t=await fr(r,{method:i,headers:a,body:o,signal:n.signal});if(c=t.body,t.headers&&typeof t.headers.get==`function`){let n=(e.parseRateLimitHint??B)(t.headers);(n.remainingRequests!==void 0||n.limitRequests!==void 0)&&(yield{type:`rate_limit_hint`,hint:n})}}for await(let t of s(c)){if(t===q){yield{type:`ping`};continue}let n=e.mapStreamEvent(t);n&&(Array.isArray(n)?yield*n:yield n)}}}}}}function hr(e,t){if(e.budget_tokens===void 0)return e;let{budget_tokens:n,...r}=e;return r.reasoning_effort===void 0?{...r,reasoning_effort:Q(n,t)}:r}function gr(e){return e.map(e=>e.type===`image`?{type:`image_url`,image_url:{url:`data:${Y(e.mimeType)};base64,${e.data}`}}:{type:`text`,text:e.text})}function _r(e){return e.messages.map(e=>{if(e.role===`user`&&Array.isArray(e.content))return{...e,content:gr(e.content)};if(e.role===`tool`){let{is_error:t,...n}=e;return n}return e})}function*vr(e){let t=e.choices?.[0]?.delta;if(t?.content&&(yield{type:`text-delta`,delta:t.content}),t?.tool_calls?.length)for(let e of t.tool_calls)yield{type:`tool_call_delta`,index:e.index,id:e.id,name:e.function?.name,argumentsDelta:e.function?.arguments};e.usage&&(yield{type:`usage`,usage:e.usage})}function $(e,t={}){let n=e,{supportsStreamUsage:r=!0,supportsWithResponse:i=!1}=t,a=Z(t.reasoningEffortTokens),o=n.chat.completions.create.bind(n.chat.completions);return{chat:{completions:{async create(e,t){let r=_r(e),o=hr({...e,messages:r},a);if(!i)return n.chat.completions.create(o,t);let{data:s,response:c}=await n.chat.completions.create(o,t).withResponse();return s&&typeof s==`object`&&ft(s,B(c.headers)),s},async*createStream(e,t){let s=_r(e),c=hr({...e,messages:s,stream:!0,...r?{stream_options:{include_usage:!0}}:{}},a),l;if(!i)l=await o(c,t);else{let{data:e,response:r}=await n.chat.completions.create(c,t).withResponse();l=e;let i=B(r.headers);(i.remainingRequests!==void 0||i.limitRequests!==void 0)&&(yield{type:`rate_limit_hint`,hint:i})}for await(let e of l)yield*vr(e)}}}}}const yr=$,br=$,xr=$,Sr=$,Cr=$,wr=$,Tr=$,Er=$,Dr=$,Or=$,kr=$,Ar=$,jr=$,Mr=$,Nr=$,Pr=$,Fr=$,Ir=$,Lr=$,Rr=$,zr=$,Br=$,Vr=$,Hr=$,Ur=$,Wr=$,Gr=$,Kr=$,qr=$,Jr=$,Yr=$,Xr=$,Zr=$,Qr=$,$r=$,ei=$,ti=$,ni=$;exports.CircuitBreaker=He,exports.ConsecutiveTripping=Ie,exports.ConsoleLogger=fn,exports.FallbackExhaustedError=w,exports.InMemoryCacheAdapter=b,exports.LLMError=p,exports.NormalizedCacheAdapter=x,exports.RateLimiter=tn,exports.RetryBudget=Ue,exports.RollingTripping=Le,exports.SSE_PING=q,exports.TieredCacheAdapter=S,exports.VernLLM=pn,exports.createMiddleware=le,exports.createMiddlewareStateBag=T,exports.createStateKey=ce,exports.defaultEstimateTokens=Zt,exports.defaultFallbackOn=ae,exports.defineCachedCallParams=hn,exports.defineCallParams=mn,exports.defineTool=C,exports.from01AI=ni,exports.fromAnthropic=Rn,exports.fromAnyscale=Zr,exports.fromAtlasCloud=ti,exports.fromBaseten=Hr,exports.fromBedrock=sr,exports.fromCerebras=Cr,exports.fromCloudflareWorkersAI=zr,exports.fromDeepInfra=kr,exports.fromDeepSeek=Sr,exports.fromFeatherless=Ur,exports.fromFetch=mr,exports.fromFireworks=Tr,exports.fromFriendli=Wr,exports.fromGemini=Xn,exports.fromGroq=br,exports.fromHyperbolic=jr,exports.fromInferenceNet=$r,exports.fromInfermatic=ei,exports.fromLMStudio=Pr,exports.fromLambdaLabs=Yr,exports.fromLepton=Qr,exports.fromMiniMax=Jr,exports.fromMistral=xr,exports.fromMoonshot=Mr,exports.fromNebius=Br,exports.fromNovita=Ar,exports.fromNvidiaNIM=Lr,exports.fromOllama=Er,exports.fromOpenAI=yr,exports.fromOpenAICompatible=$,exports.fromOpenRouter=Dr,exports.fromParasail=Kr,exports.fromPerplexity=Or,exports.fromSambaNova=Vr,exports.fromSiliconFlow=Gr,exports.fromSnowflakeCortex=Xr,exports.fromStepFun=qr,exports.fromTogether=wr,exports.fromVLLM=Fr,exports.fromVercelAIGateway=Rr,exports.fromXAI=Ir,exports.fromZhipu=Nr,exports.hasIssues=h,exports.isFallbackExhaustedError=oe,exports.isLLMError=m,exports.isStreamResult=re,exports.isToolCallResult=ee,exports.metaRef=se,exports.parseSseStream=gn;
|
|
15
15
|
//# sourceMappingURL=index.cjs.map
|