@bivelio/savings-layer 0.4.30 → 0.4.32

package/README.md

@@ -211,7 +211,8 @@ import { createSavingsLayer, OpenAICompatibleProvider } from "@bivelio/savings-l
 // Your provider. LLM traffic goes straight to it — the SDK never proxies it.
 // Any OpenAI-compatible endpoint works: OpenAI, Azure, vLLM, SGLang, a gateway.
 // Model ids are qualified (`openai/gpt-4o-mini`); the prefix is stripped on the
-// wire for OpenAI and Azure and kept for gateways — see `stripProviderPrefix`.
+// wire for OpenAI, Azure and Google's Gemini API and kept for gateways — see
+// `stripProviderPrefix`.
 const provider = new OpenAICompatibleProvider({
   baseUrl: "https://api.openai.com/v1",
   apiKey: process.env.OPENAI_API_KEY,
@@ -282,6 +283,46 @@ nothing either. The same rule applies in the gateway. Any other refusal is price
 its `usage` as usual. That covers `bio`, `frontier_llm` and `reasoning_extraction`, a
 refusal after partial output, and a category the table does not list.
 
+Most refusals carry no signal from the API: the model writes "I can't help with that."
+as ordinary text and ends with `end_turn` or `stop`. The layer also looks for these,
+with a conservative text heuristic. It checks only the opening of the answer, or the
+whole answer when it is short, and it needs a real refusal clause, such as "I'm sorry,
+but I can't help with that", "I cannot assist", "I can't determine…" or "As an AI, I
+cannot provide". After an apology, any verb counts ("I'm sorry, but I can't
+determine…"), and so does a stated reason with no "I can't" ("I'm sorry, but
+accessing … is illegal", "… would be a violation of…"). When the answer opens with
+the refusal itself ("I can't", "I won't", "I'm not going to", "I'm unable to", and
+"No puedo", "Je ne peux pas", "Ich kann … nicht"…), any verb counts too ("I can't
+argue that…"), except for a tested list of idioms that are not refusals ("I can't
+wait", "I won't lie", "I can't find any issues", "I can't be sure, but…").
+A single word is not enough.
+An answer that opens with empathy and, in its first two sentences, sends the user to
+a mental health professional, a crisis line, emergency services or someone they
+trust, instead of answering, counts too (rule `redirection`). It covers English,
+Spanish, Catalan, French, German, Italian and Portuguese. A match is measured and
+served as usual. It is not written to the exact or semantic cache, and it is not
+replayed by reference. A concurrent identical request served from it is `estimated`.
+The trace says why (`skipReason: "text_refusal"`, with the `rule`). A refusal
+already stored by an earlier version is not served either. In the gateway, the
+coalesced twin of such a response is `estimated` (`primary-text-refusal`). A false
+positive costs one cache hit. A normal answer that starts with "Sorry for the delay,
+here is…", praises with "I can't recommend this enough", has "I can't" later on, or
+offers empathy with practical help and no referral ("I'm sorry to hear about your
+flight; here are three options…") is cached as before.
+
+**Sensitive topics.** A request whose user turns touch self-harm or suicide, a mental
+health crisis or a medical emergency ("How do I commit suicide?", "I'm having a panic
+attack", "my friend is not breathing") is never read from or written to the exact or
+semantic cache, even when the answer is useful, and leaves no reference record. It is
+measured and served as usual; a concurrent identical request served from it is
+`estimated`. The `policy` and `cache-write` traces say why
+(`skipReason: "sensitive_topic"`, with the `category`). The check is lexical and
+conservative: phrases, not single words ("kill a Python process", "Suicide Squad" and
+"cut my hair" do not count). It covers the same twelve languages as the freshness
+and personalization checks, with shorter lists outside English, Spanish, Catalan,
+French, German, Italian and Portuguese. In the gateway, the coalesced twin of such a
+request is `estimated` (`sensitive-topic`). A false positive costs one cache hit.
+
 **Streaming.** `generateStream()` takes the same request and hands you the answer as
 the provider writes it, then the same `SavingsReport`:
 
@@ -438,6 +479,45 @@ It runs once per HTTP attempt, inside the SDK's retries: an attempt the provider
 rejects reaches the SDK unchanged and is not recorded. Only message creation goes
 through the layer; `countTokens`, batches, models and files pass untouched.
 
+The middleware forwards your request body unchanged, history included, so it never
+edits a turn before a thinking block. It also does not remove parameters the model
+rejects. On Claude Sonnet 5.5, Opus 5.5, Fable 5.1 and Mythos 5.1 a forced
+`tool_choice` (`any` or `tool`) returns a 400. On those models, and on Opus 4.7/4.8,
+Opus 5, Sonnet 5 and Fable 5, so does a non-default `temperature`, `top_p` or `top_k`.
+The middleware warns once per model and cause (`silenceWarnings` turns it off) and
+the API's 400 reaches you unchanged.
+
+<details>
+<summary><b>Claude Sonnet 5.5, Opus 5.5, Fable 5.1: what <code>AnthropicProvider</code> sends instead</b></summary>
+
+These models reject some request parameters with a 400 ([forcing tool
+use](https://platform.claude.com/docs/en/agents-and-tools/tool-use/define-tools#forcing-tool-use),
+[Sonnet 5.5](https://platform.claude.com/docs/en/models/sonnet-5-5/overview)).
+`AnthropicProvider` sends a request they accept instead of failing:
+
+- **Output contract.** A forced `tool_choice` returns a 400 on these models, with or
+  without thinking. When your schema meets `strict` mode, the contract travels as
+  structured output (`output_config.format`). Without your tools, the provider still
+  enforces the contract. With your tools, they stay callable with `tool_choice: auto`
+  and the final text answer comes back in the contract's shape. When your schema does
+  not meet `strict`, the contract travels as a tool the model is not forced to call,
+  with one sentence in the system prompt that tells it to answer through that tool.
+  Models that support forced tool use (Opus 5, Sonnet 5 and the 4.x models) get the
+  same request as before.
+- **`temperature`.** A non-default value is dropped, with one warning per model,
+  because it can only return a 400. The answer is then sampled, not deterministic.
+  `temperature: 1`, the API default, is sent as is.
+- **A model the table does not list** (a new model or a gateway alias): when the API
+  rejects either parameter, the provider repeats the request once without it and
+  remembers the model for later requests. A 400 is not billed, so the retry is not
+  counted in `retries`.
+- **Thinking blocks.** `ChatMessage` does not carry thinking blocks, so this adapter
+  never sends one. The preserved-thinking check, which returns a 400 when a turn
+  before a thinking block has been edited, has nothing to check on its requests, even
+  when the levers that rewrite history are on.
+
+</details>
+
 ## What you get on every request
 
 A `SavingsReport` with the numbers behind every claim, kept in **four separate
@@ -666,7 +746,16 @@ const agentLayer = createSavingsLayer({
 
 By default the layer writes the provider's prompt cache at two fixed places: the end
 of the system header and the end of the history before your last user message, and
-stops marking a prefix whose measured hit rate does not pay the write premium.
+stops marking a prefix whose measured hit rate does not pay the write premium. On
+Anthropic, where a write costs 1.25x, the history and a tool catalogue sent without a
+system header are marked only when the request itself shows they will be sent again:
+a tool-loop step after your last user message, or a conversation already past its
+second turn. The first follow-up of a conversation (`[question, answer, question]`)
+marks only the system header, so two-turn chats pay no premium that nothing reads.
+In an agent loop with no system header, the end of the tool list is marked once the
+catalogue reaches the model's minimum cacheable size (512 tokens on Claude Sonnet 5.5
+and Opus 5.5, 1,024 on Sonnet 5, 4,096 on Haiku 4.5). That discount is the provider's
+cache on your bytes: it is never credited as the layer's saving.
 
 `cacheBreakpoints: "auto"` places them from what your traffic actually does. The layer
 compares each request's prefix with the ones it sent before, per model, tool set and
@@ -711,7 +800,11 @@ const placedLayer = createSavingsLayer({
 Calibrated reuse and `cacheBreakpoints: "auto"` learn from your traffic, and by default
 what they learn lives in the process. On serverless platforms (Vercel Functions, AWS
 Lambda) an instance may serve one request or live a few minutes, so neither would ever
-get past its starting point. Give each one somewhere to keep it:
+get past its starting point. Give each one somewhere to keep it. Giving the layer a
+`semanticCache` only says where reworded questions are kept; `semanticCacheDefault: true`
+is what switches reuse on for every request (a request's own `savings.semanticCache`
+still wins), and without it, or `savings.semanticCache` on each request, the cache
+stays empty and learns nothing:
 
 ```ts
 import { InMemorySemanticCache } from "@bivelio/savings-layer";
@@ -731,6 +824,8 @@ async function serverlessLayer(kv: SharedKeyValue) {
     provider,
     licenseKey: process.env.BIVELIO_LICENSE_KEY,
     semanticCache,
+    // Passing the cache does not switch it on; this asks for it on every request.
+    semanticCacheDefault: true,
     cacheBreakpoints: "auto",
     cacheBreakpointStore: {
       get: (key) => kv.get(key),
@@ -802,6 +897,54 @@ console.log(data.answer); // 8
 
 </details>
 
+<details>
+<summary><b>Output cap — with or without a contract</b></summary>
+
+`maxOutputTokens` on the request caps the answer of any request, free text
+included. With `response.maxOutputTokens` as well, the smaller of the two is sent.
+
+```ts
+const { text, finishReason } = await layer.generate({
+  model: "openai/gpt-5-mini",
+  messages: [{ role: "user", content: "Summarize this in two sentences: …" }],
+  maxOutputTokens: 300,
+});
+```
+
+On the wire it becomes Anthropic's `max_tokens`; on Chat Completions,
+`max_completion_tokens` when `baseUrl` is OpenAI's own API (`max_tokens` is
+deprecated there and its reasoning models, o-series and gpt-5.x, reject it) and
+`max_tokens` on any other OpenAI-compatible endpoint, which is the field they all
+understand. If an endpoint answers 400 naming the field it got as unsupported, the
+request is repeated once with the other one (nothing was generated, so nothing is
+billed twice) and the model is remembered. To choose the field yourself:
+`new OpenAICompatibleProvider({ baseUrl, maxTokensParam: "max_completion_tokens" })`.
+
+**OpenAI-compatible endpoints that are not OpenAI** (Gemini's
+`generativelanguage.googleapis.com/v1beta/openai`, LiteLLM, vLLM, Ollama, a router).
+`prompt_cache_key`, OpenAI's prefix-cache routing hint, goes only to OpenAI's own API:
+Gemini answers 400 to a field it does not know, and up to 0.4.31 every request with a
+cached prefix failed there. A proxy in front of OpenAI that forwards it can opt in with
+`sendPromptCacheKey: true`. Any optional field an endpoint still rejects by name
+(`prompt_cache_key`, `service_tier`, `seed`, `reasoning_effort`) is dropped, the request
+repeated once and the model remembered; `reasoning_effort` is never dropped on OpenAI's
+own API, where its 400 is the answer. Gemini leaves its thinking tokens out of
+`completion_tokens` and bills them as output, so the layer counts output as
+`total_tokens − prompt_tokens` whenever the total is larger (and reports the difference
+as reasoning); on OpenAI's API `completion_tokens` already include reasoning and nothing
+is added. Gemini stops a filtered answer with `finish_reason: "content_filter: OTHER"`
+(or another reason): it is `finishReason: "content_filter"` (the adapter keeps the
+reason in `finishReasonDetail` and the report's cache trace names it), and like every
+provider refusal it is measured, never
+cached and never credited. Google does not document that such refusals are free, so
+they are costed from their `usage` like any call.
+
+The cap is yours, not a saving: it is never credited as output avoided. An answer
+it cut (`finishReason: "length"`) is not cached, and
+`report.qualityLedger.truncation` says the limit was yours.
+
+</details>
+
 <details>
 <summary><b>Tool calls</b></summary>
 
@@ -854,6 +997,18 @@ console.log(res.text);
   the SDK warns once.
 - **Errors.** When a tool fails, put the error in `content` and set `isError: true`.
   Anthropic receives it as `is_error`; providers without such a flag get the text.
+- **Gemini 3's thought signature.** Through its OpenAI-compatible endpoint Gemini
+  returns each call with `extra_content.google.thought_signature` and rejects the next
+  turn with a 400 if the call comes back without it. It arrives in `call.extraContent`;
+  pushing `res.toolCalls` back as they came (as above) sends it back untouched. It is
+  opaque and not part of any cache key.
+- **gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna with tools need `reasoningEffort: "none"`.**
+  On Chat Completions OpenAI rejects function tools for these models unless
+  `reasoning_effort` is `"none"` («Function tools with reasoning_effort are not supported for gpt-5.6-sol in
+  /v1/chat/completions»). Declare it on the request:
+  `layer.generate({ model: "openai/gpt-5.6-sol", messages, tools, reasoningEffort: "none" })`.
+  The layer never picks an effort itself: without the field nothing is sent. It is part
+  of the cache key, like `temperature` and `seed`.
 
 A tool‑call turn is never served from cache and never counted as an avoided call:
 an action that has not run yet cannot be a saving. The answer that follows a tool loop
@@ -862,13 +1017,21 @@ each result answers and its `isError`. Two conversations that differ only there
 share a cached answer; the fresh call ids of a new run of the same loop are not a
 difference.
 
+**Which tools reach the model.** The layer never cuts a catalogue it cannot rank. If
+your tools declare no `intents` and no `risk` (the usual shape of MCP and OpenAI tool
+lists), or you pass `savings: { router: false }`, every tool goes to the provider
+exactly as you sent it, and the layer credits nothing for the tool block. Only when
+your tools declare `intents`/`risk` does the selector keep the relevant ones, up to
+`new ToolSelector({ maxTools })` (8 by default), and hold back write tools from a
+request that is not an action. Whatever it leaves out, the model cannot call in that
+request: the report's `tool-selector` trace lists it, the tokens count as eliminated
+input, and the SDK warns once per process (`silenceWarnings` turns the warning off).
+
 **Large catalogues: let the provider search them (opt‑in).** With hundreds of tools
 (several MCP servers, say), `savings: { toolSearch: "auto" }` sends every tool deferred
-behind the provider's own tool search instead of letting the layer's selector keep a
-few. A catalogue that declares no `intents` or `risk` gives the selector nothing to go
-on, so it keeps the first ones and the model cannot call the rest; with the provider's
-search the model looks up what it needs, no tool you declared is out of reach, and the
-cached prompt prefix stays the same from one question to the next.
+behind the provider's own tool search instead of putting the whole catalogue in every
+request's prompt. The model looks up what it needs, no tool you declared is out of
+reach, and the cached prompt prefix stays the same from one question to the next.
 
 - **Where.** The Anthropic adapter, on models in Anthropic's tool search compatibility
   table. OpenAI documents its tool search for the Responses and Agents APIs, not for

package/dist/adapters/anthropic-sdk.js

@@ -1,67 +1,73 @@
 import {createSavingsLayer,LicenseError,sha256,stableStringify}from'@bivelio/savings-layer';import {AsyncLocalStorage}from'async_hooks';var x=t=>typeof t=="number"&&Number.isFinite(t)&&t>0?t:0;function J(t){if(!Array.isArray(t))return null;let e={input:0,output:0,cacheRead:0,cacheWrite:0,cacheWrite1h:0,iteraciones:0};for(let n of t){if(n===
 null||typeof n!="object")continue;let r=n;if(r.type!=="compaction")continue;e.iteraciones++,e.input+=x(r.input_tokens),e.output+=x(r.output_tokens),e.cacheRead+=x(r.cache_read_input_tokens);let o=r.cache_creation,
 s=x(o?.ephemeral_5m_input_tokens),a=x(o?.ephemeral_1h_input_tokens),i=typeof r.cache_creation_input_tokens=="number"?x(r.cache_creation_input_tokens):s+a;e.cacheWrite+=i,e.cacheWrite1h+=Math.min(a,i);}
-return e.iteraciones>0?e:null}function B(t,e,n){let r=t??{},o=J(r.iterations),s=W(r.cache_read_input_tokens,o?.cacheRead),a=W(r.cache_creation_input_tokens,o?.cacheWrite),i=r.cache_creation?.ephemeral_1h_input_tokens,u=typeof i=="\
+return e.iteraciones>0?e:null}function W(t,e,n){let r=t??{},o=J(r.iterations),s=F(r.cache_read_input_tokens,o?.cacheRead),a=F(r.cache_creation_input_tokens,o?.cacheWrite),i=r.cache_creation?.ephemeral_1h_input_tokens,l=typeof i=="\
 number"||(o?.cacheWrite1h??0)>0?Math.min(Math.max(0,(typeof i=="number"?i:0)+(o?.cacheWrite1h??0)),a??0):void 0,d=(r.input_tokens??0)+(o?.input??0)+(a??0)+(s??0),g=(e??0)+(o?.output??0),m=typeof r.input_tokens!=
-"number",h=typeof e!="number";return {usage:{inputTokens:d,outputTokens:g,...s!=null?{cachedInputTokens:s}:{},...a!=null?{cacheWriteInputTokens:a}:{},...u!==void 0?{cacheWrite1hInputTokens:u}:{},...n||
-h||o?{}:{reasoningTokens:0},...m||h?{complete:false}:{}},entradaAusente:m,salidaAusente:h}}function W(t,e){return e?(t??0)+e:t}function F(t){let e=t?.service_tier;return typeof e=="string"&&e!==""?e:void 0}
+"number",h=typeof e!="number";return {usage:{inputTokens:d,outputTokens:g,...s!=null?{cachedInputTokens:s}:{},...a!=null?{cacheWriteInputTokens:a}:{},...l!==void 0?{cacheWrite1hInputTokens:l}:{},...n||
+h||o?{}:{reasoningTokens:0},...m||h?{complete:false}:{}},entradaAusente:m,salidaAusente:h}}function F(t,e){return e?(t??0)+e:t}function B(t){let e=t?.service_tier;return typeof e=="string"&&e!==""?e:void 0}
 function q(t,e){let n=Object.fromEntries(Object.entries(e??{}).filter(([,o])=>o!=null)),r=t?.cache_creation||e?.cache_creation?{...t?.cache_creation??{},...e?.cache_creation??{}}:void 0;return {...t??{},
 ...n,...r?{cache_creation:r}:{}}}function V(t){return t==="thinking"||t==="redacted_thinking"}function G(t){switch(t){case "end_turn":return "stop";case "max_tokens":return "length";case "tool_use":return "\
-tool_calls";case "refusal":return "content_filter";default:return t}}var ye=new Set([null,"cyber","general_harms"]);function H(t){if(t.stopReason!=="refusal")return  false;let e=t.stopDetails;if(e===null||typeof e!=
-"object"||Array.isArray(e))return  false;let n=e;if(n.type!=="refusal"||!("category"in n))return  false;let r=n.category;return !(r!==null&&typeof r!="string"||!ye.has(r)||t.bloques!==0||t.outputTokens!==0||Array.
-isArray(t.iterations)&&t.iterations.length>1)}var ve={"api.anthropic.com":"anthropic","api.openai.com":"openai","generativelanguage.googleapis.com":"google"};function be(t){if(typeof t!="string")return null;try{let e=new URL(t);return e.protocol!==
-"https:"||e.port!==""?null:ve[e.hostname.toLowerCase()]??null}catch{return null}}function K(t){return be(t)!==null?"oficial":"no-oficial"}var Z=Symbol.for("@bivelio/savings-layer:provider-origin");var L=class{decoder;resto="";crPendiente=false;alPrincipio=true;datos="";evento="";ultimoId;push(e){let n=typeof e=="string"?e:(this.decoder??=new TextDecoder("utf-8")).decode(e,{stream:true});return this.procesar(
+tool_calls";case "refusal":return "content_filter";default:return t}}var be=new Set([null,"cyber","general_harms"]);function H(t){if(t.stopReason!=="refusal")return  false;let e=t.stopDetails;if(e===null||typeof e!=
+"object"||Array.isArray(e))return  false;let n=e;if(n.type!=="refusal"||!("category"in n))return  false;let r=n.category;return !(r!==null&&typeof r!="string"||!be.has(r)||t.bloques!==0||t.outputTokens!==0||Array.
+isArray(t.iterations)&&t.iterations.length>1)}var Re=["claude-opus-5-5","claude-sonnet-5-5","claude-fable-5-1","claude-mythos-5-1"],we=["claude-opus-4-7","claude-opus-4-8","claude-opus-5","claude-sonnet-5","claude-fable-5","claude-mythos-5"];function Se(t){return /claude-[a-z]+-\d[a-z0-9-]*/.exec(t.toLowerCase())?.[0]}function K(t,e){let n=Se(t);return n===void 0?false:e.some(r=>n===r||n.startsWith(`${r}-`))}function Z(t){return K(t,Re)}function Q(t){
+return K(t,we)}var xe={"api.anthropic.com":"anthropic","api.openai.com":"openai","generativelanguage.googleapis.com":"google"};function Ae(t){if(typeof t!="string")return null;try{let e=new URL(t);return e.protocol!==
+"https:"||e.port!==""?null:xe[e.hostname.toLowerCase()]??null}catch{return null}}function X(t){return Ae(t)!==null?"oficial":"no-oficial"}var Y=Symbol.for("@bivelio/savings-layer:provider-origin");var O=class{decoder;resto="";crPendiente=false;alPrincipio=true;datos="";evento="";ultimoId;push(e){let n=typeof e=="string"?e:(this.decoder??=new TextDecoder("utf-8")).decode(e,{stream:true});return this.procesar(
 n)}end(){let e=[];if(this.decoder){let n=this.decoder.decode();n!==""&&e.push(...this.procesar(n));}return this.resto!==""&&(this.linea(this.resto,e),this.resto=""),this.despachar(e),this.crPendiente=false,
 e}procesar(e){let n=[];if(e==="")return n;let r=e;if(this.alPrincipio&&(this.alPrincipio=false,r.charCodeAt(0)===65279&&(r=r.slice(1)),r===""))return n;let o=0;this.crPendiente&&(this.crPendiente=false,r.charCodeAt(
 0)===10&&(o=1));let s=o;for(;o<r.length;){let a=r.charCodeAt(o);if(a!==10&&a!==13){o++;continue}this.linea(this.resto+r.slice(s,o),n),this.resto="",a===13&&(o+1<r.length?r.charCodeAt(o+1)===10&&o++:this.
 crPendiente=true),o++,s=o;}return this.resto+=r.slice(s),n}linea(e,n){if(e===""){this.despachar(n);return}if(e.charCodeAt(0)===58)return;let r=e.indexOf(":"),o,s;switch(r===-1?(o=e,s=""):(o=e.slice(0,r),
 s=e.slice(r+1),s.charCodeAt(0)===32&&(s=s.slice(1))),o){case "data":this.datos+=s+`
 `;break;case "event":this.evento=s;break;case "id":s.includes("\0")||(this.ultimoId=s);break;}}despachar(e){if(this.datos===""){this.evento="";return}e.push({event:this.evento===""?"message":
-this.evento,data:this.datos.slice(0,-1),...this.ultimoId!==void 0?{id:this.ultimoId}:{}}),this.datos="",this.evento="";}};var Q=new Set;function X(t,e,n){t===true||Q.has(e)||(Q.add(e),console.warn(n));}var Re="[bvsala] internal: the layer changed a request this adapter must forward unchanged; nothing was sent.",Se="[bvsala] internal: the layer asked for a second model call within one framework call;\
- nothing more was sent.",we="[bvsala] internal: the layer called the model outside a framework call; nothing was sent.";function U(t,e,n,r){return {renderizado:t,generar:e,...n?{transmitir:n}:{},invocada:false,
-completada:false,...r!==void 0?{origen:r}:{}}}function xe(t,e){if(t===e)return  true;try{return stableStringify(t)===stableStringify(e)}catch{return  false}}function ee(t,e){let n=t??[],r=e??[];return n.length===r.length&&n.every((o,s)=>xe(o,r[s]))}function Ae(t,e){return t.model===
-e.model&&ee(t.messages,e.messages)&&ee(t.tools,e.tools)}function te(t,e){let n=o=>{let s=e.getStore();if(!s)throw new Error(we);if(s.invocada)throw new Error(Se);if(!Ae(o,s.renderizado))throw new Error(
-Re);return s.invocada=true,s},r={id:`bivelio-adapter:${t}`,async generate(o,s){return n(o).generar(s?.signal)},async*generateStream(o,s){let a=n(o);if(a.transmitir){yield*a.transmitir(s?.signal);return}
-let i=await a.generar(s?.signal);i.text!==""&&(yield {type:"text",delta:i.text}),yield {type:"done",response:i};}};return Object.defineProperty(r,Z,{value:()=>e.getStore()?.origen}),r}function ne(){return new AsyncLocalStorage}var Ce=["licenseKey","licenseServerUrl","serviceAccountKey","licenseRefresh","licenseRevocation","licensePollIntervalSeconds","licenseFetch","metering","meteringUrl","meteringFetch","costModels","defa\
-ultCostModel","exactCache","exactCacheTtlMs","resolvers","exactTokenizer","silenceWarnings","failOpen"];function re(t,e){return e instanceof LicenseError||t.errorDelMarco&&t.errorDelMarco.error===e?false:t.invocada&&
-t.completada}var oe="[bvsala] the model answered, but the savings layer failed while accounting for it; the answer is returned unchanged, without a savings report and without a usage record.";function se(t,e){
-let n=ne(),r={};for(let a of Ce)e[a]!==void 0&&(r[a]=e[a]);let o=createSavingsLayer({...r,provider:te(t,n)}),s=(a,i)=>X(e.silenceWarnings,`${t}:${a}`,i);return {ejecutar(a,i){return n.run(i,()=>o.generate(a))},async generar(a,i){
-try{let u=await n.run(i,()=>o.generate(a));return i.invocada?{tipo:"vivo",result:u}:{tipo:"atajo",result:u}}catch(u){if(re(i,u))return s("sin-informe",oe),{tipo:"sin-informe"};throw u}},async transmitir(a,i,u){
-let g=n.run(i,()=>o.generateStream(a,u?{signal:u}:{})).result;i.resultado=g,g.catch(()=>{});let m;try{m=await Promise.race([i.abierta.then(()=>"abierta"),g.then(()=>"resuelta")]);}catch(_){if(re(i,_))return s(
-"sin-informe",oe),{tipo:"sin-informe"};throw _}if(m==="abierta")return {tipo:"abierto"};let h=await g;return i.invocada?{tipo:"entero",result:h}:{tipo:"atajo",result:h}},informe(a){let i=JSON.parse(JSON.
-stringify(a));if(e.onReport)try{e.onReport(i);}catch{s("onreport","[bvsala] `onReport` threw; the error was caught and the model call was not affected.");}return i},avisar:s,close(){o.close();}}}function ae(){
+this.evento,data:this.datos.slice(0,-1),...this.ultimoId!==void 0?{id:this.ultimoId}:{}}),this.datos="",this.evento="";}};var ee=new Set;function te(t,e,n){t===true||ee.has(e)||(ee.add(e),console.warn(n));}var De="[bvsala] internal: the layer changed a request this adapter must forward unchanged; nothing was sent.",Ce="[bvsala] internal: the layer asked for a second model call within one framework call;\
+ nothing more was sent.",Pe="[bvsala] internal: the layer called the model outside a framework call; nothing was sent.";function I(t,e,n,r){return {renderizado:t,generar:e,...n?{transmitir:n}:{},invocada:false,
+completada:false,...r!==void 0?{origen:r}:{}}}function Oe(t,e){if(t===e)return  true;try{return stableStringify(t)===stableStringify(e)}catch{return  false}}function re(t,e){let n=t??[],r=e??[];return n.length===r.length&&n.every((o,s)=>Oe(o,r[s]))}function Te(t,e){return t.model===
+e.model&&re(t.messages,e.messages)&&re(t.tools,e.tools)}function oe(t,e){let n=o=>{let s=e.getStore();if(!s)throw new Error(Pe);if(s.invocada)throw new Error(Ce);if(!Te(o,s.renderizado))throw new Error(
+De);return s.invocada=true,s},r={id:`bivelio-adapter:${t}`,async generate(o,s){return n(o).generar(s?.signal)},async*generateStream(o,s){let a=n(o);if(a.transmitir){yield*a.transmitir(s?.signal);return}
+let i=await a.generar(s?.signal);i.text!==""&&(yield {type:"text",delta:i.text}),yield {type:"done",response:i};}};return Object.defineProperty(r,Y,{value:()=>e.getStore()?.origen}),r}function se(){return new AsyncLocalStorage}var Le=["licenseKey","licenseServerUrl","serviceAccountKey","licenseRefresh","licenseRevocation","licensePollIntervalSeconds","licenseFetch","metering","meteringUrl","meteringFetch","costModels","defa\
+ultCostModel","exactCache","exactCacheTtlMs","resolvers","exactTokenizer","silenceWarnings","failOpen"];function ae(t,e){return e instanceof LicenseError||t.errorDelMarco&&t.errorDelMarco.error===e?false:t.invocada&&
+t.completada}var ie="[bvsala] the model answered, but the savings layer failed while accounting for it; the answer is returned unchanged, without a savings report and without a usage record.";function ce(t,e){
+let n=se(),r={};for(let a of Le)e[a]!==void 0&&(r[a]=e[a]);let o=createSavingsLayer({...r,provider:oe(t,n)}),s=(a,i)=>te(e.silenceWarnings,`${t}:${a}`,i);return {ejecutar(a,i){return n.run(i,()=>o.generate(a))},async generar(a,i){
+try{let l=await n.run(i,()=>o.generate(a));return i.invocada?{tipo:"vivo",result:l}:{tipo:"atajo",result:l}}catch(l){if(ae(i,l))return s("sin-informe",ie),{tipo:"sin-informe"};throw l}},async transmitir(a,i,l){
+let g=n.run(i,()=>o.generateStream(a,l?{signal:l}:{})).result;i.resultado=g,g.catch(()=>{});let m;try{m=await Promise.race([i.abierta.then(()=>"abierta"),g.then(()=>"resuelta")]);}catch(b){if(ae(i,b))return s(
+"sin-informe",ie),{tipo:"sin-informe"};throw b}if(m==="abierta")return {tipo:"abierto"};let h=await g;return i.invocada?{tipo:"entero",result:h}:{tipo:"atajo",result:h}},informe(a){let i=JSON.parse(JSON.
+stringify(a));if(e.onReport)try{e.onReport(i);}catch{s("onreport","[bvsala] `onReport` threw; the error was caught and the model call was not affected.");}return i},avisar:s,close(){o.close();}}}function le(){
 let t;return {abierta:new Promise(n=>{t=n;}),abrir:t}}var C=class{pendientes=[];esperando;terminada=false;fallo;push(e){if(this.terminada||this.fallo)return;let n=this.esperando;if(n){this.esperando=void 0,n.resolver({value:e,done:false});return}this.pendientes.
 push(e);}end(){if(this.terminada||this.fallo)return;this.terminada=true;let e=this.esperando;e&&this.pendientes.length===0&&(this.esperando=void 0,e.resolver({value:void 0,done:true}));}fail(e){if(this.terminada||
 this.fallo)return;this.fallo={error:e};let n=this.esperando;n&&this.pendientes.length===0&&(this.esperando=void 0,n.rechazar(e));}get cerrada(){return this.terminada||this.fallo!==void 0}[Symbol.asyncIterator](){
 return {next:()=>this.pendientes.length>0?Promise.resolve({value:this.pendientes.shift(),done:false}):this.fallo?Promise.reject(this.fallo.error):this.terminada?Promise.resolve({value:void 0,done:true}):new Promise(
-(e,n)=>{this.esperando={resolver:e,rechazar:n};}),return:()=>(this.terminada=true,this.pendientes.length=0,Promise.resolve({value:void 0,done:true}))}}};var Le={router:false,compression:false,elision:false,serialization:false,toolDiscovery:false,modelCascade:"off",outputCodebook:false,deferrable:false,semanticCache:false,verification:false};
-function Me(t,e,n={}){let r=o=>t?.[o]!==false&&e?.[o]!==false;return {...Le,exactCache:r("exactCache")&&n.sinReutilizar!==true,singleflight:r("singleflight")&&n.sinReutilizar!==true,deterministicResolvers:r("det\
-erministicResolvers")&&n.sinResolvers!==true}}function Oe(t,e){let n=t.readOnlyTools;if(typeof n=="function")try{return n(e)===!0}catch{return  false}return Array.isArray(n)?n.includes(e):false}function ce(t,e){
-let n=e.delProveedor!==true&&Oe(t,e.nombre);return {id:e.nombre,intents:[],risk:n?"read":"write",...e.parametros!==void 0&&e.parametros!==null&&typeof e.parametros=="object"?{parameters:e.parametros}:{},
+(e,n)=>{this.esperando={resolver:e,rechazar:n};}),return:()=>(this.terminada=true,this.pendientes.length=0,Promise.resolve({value:void 0,done:true}))}}};var qe={router:false,compression:false,elision:false,serialization:false,toolDiscovery:false,modelCascade:"off",outputCodebook:false,deferrable:false,semanticCache:false,verification:false};
+function Ie(t,e,n={}){let r=o=>t?.[o]!==false&&e?.[o]!==false;return {...qe,exactCache:r("exactCache")&&n.sinReutilizar!==true,singleflight:r("singleflight")&&n.sinReutilizar!==true,deterministicResolvers:r("det\
+erministicResolvers")&&n.sinResolvers!==true}}function Ue(t,e){let n=t.readOnlyTools;if(typeof n=="function")try{return n(e)===!0}catch{return  false}return Array.isArray(n)?n.includes(e):false}function de(t,e){
+let n=e.delProveedor!==true&&Ue(t,e.nombre);return {id:e.nombre,intents:[],risk:n?"read":"write",...e.parametros!==void 0&&e.parametros!==null&&typeof e.parametros=="object"?{parameters:e.parametros}:{},
 ...typeof e.descripcion=="string"?{description:e.descripcion}:{}}}function P(t,e,n){if(t===null)return null;switch(typeof t){case "string":case "boolean":return t;case "number":return Number.isFinite(t)?
 t:String(t);case "bigint":return `${t.toString()}n`;case "undefined":case "function":case "symbol":return}let r=t;if(r instanceof Uint8Array)return e.push(r),{$bytes:e.length-1};if(r instanceof ArrayBuffer)
 return e.push(new Uint8Array(r)),{$bytes:e.length-1};if(ArrayBuffer.isView(r))return e.push(new Uint8Array(r.buffer,r.byteOffset,r.byteLength)),{$bytes:e.length-1};if(r instanceof URL)return {$url:r.href};
 if(r instanceof Date)return {$date:Number.isNaN(r.getTime())?"invalid":r.toISOString()};if(n.has(r))return "$cycle";n.add(r);try{if(Array.isArray(r))return r.map(s=>{let a=P(s,e,n);return a===void 0?null:
 a});if(r instanceof Map)return {$map:[...r.entries()].map(s=>P(s,e,n))};if(r instanceof Set)return {$set:[...r.values()].map(s=>P(s,e,n))};let o={};for(let s of Object.keys(r).sort()){let a=P(r[s],e,n);
-a!==void 0&&(o[s]=a);}return o}finally{n.delete(r);}}async function ie(t){let e=globalThis.crypto?.subtle;if(e){let r=new Uint8Array(await e.digest("SHA-256",t));return Array.from(r,o=>o.toString(16).padStart(
-2,"0")).join("")}let n="";for(let r=0;r<t.length;r+=32768)n+=String.fromCharCode(...t.subarray(r,r+32768));return sha256(n)}async function Te(t){try{let e=[],n=JSON.stringify(P(t,e,new WeakSet))??"null",r=await Promise.
-all(e.map(ie)),o=new TextEncoder().encode(`${n}|${r.join(",")}`);return await ie(o)}catch{return null}}async function ue(t,e){let n=await Te(e.proyeccion),r=`${e.context?.authorizationScope??"default"}\
+a!==void 0&&(o[s]=a);}return o}finally{n.delete(r);}}async function ue(t){let e=globalThis.crypto?.subtle;if(e){let r=new Uint8Array(await e.digest("SHA-256",t));return Array.from(r,o=>o.toString(16).padStart(
+2,"0")).join("")}let n="";for(let r=0;r<t.length;r+=32768)n+=String.fromCharCode(...t.subarray(r,r+32768));return sha256(n)}async function ze(t){try{let e=[],n=JSON.stringify(P(t,e,new WeakSet))??"null",r=await Promise.
+all(e.map(ue)),o=new TextEncoder().encode(`${n}|${r.join(",")}`);return await ue(o)}catch{return null}}async function pe(t,e){let n=await ze(e.proyeccion),r=`${e.context?.authorizationScope??"default"}\
 |bvsala-adapter:${n??"none"}`;return {model:e.model,messages:e.messages,...e.tools.length>0?{tools:e.tools}:{},...typeof e.temperature=="number"?{temperature:e.temperature}:{},...typeof e.seed=="number"?
-{seed:e.seed}:{},context:{...e.context??{},authorizationScope:r},savings:Me(t.savings,e.savings,{...e.restricciones,...n===null?{sinReutilizar:true}:{}})}}function le(t,e){if(!(!t&&!e))return {...t??{},...e??
-{}}}function de(t){if(!t||typeof t!="object")return {};let e=t,n={};return e.context&&typeof e.context=="object"&&(n.context=e.context),e.savings&&typeof e.savings=="object"&&(n.savings=e.savings),n}var I="[bvsala] could not read the request body; nothing was sent.";function p(t){return typeof t=="object"&&t!==null&&!Array.isArray(t)}function Ee(t){if((t.method??"GET").toUpperCase()!=="POST")return  false;
-try{return new URL(t.url).pathname.replace(/\/+$/,"").endsWith("/v1/messages")}catch{return  false}}function A(t){return typeof t=="string"?[{type:"text",text:t}]:Array.isArray(t)?t.filter(p):[]}function M(t){
+{seed:e.seed}:{},context:{...e.context??{},authorizationScope:r},savings:Ie(t.savings,e.savings,{...e.restricciones,...n===null?{sinReutilizar:true}:{}})}}function fe(t,e){if(!(!t&&!e))return {...t??{},...e??
+{}}}function ge(t){if(!t||typeof t!="object")return {};let e=t,n={};return e.context&&typeof e.context=="object"&&(n.context=e.context),e.savings&&typeof e.savings=="object"&&(n.savings=e.savings),n}var U="[bvsala] could not read the request body; nothing was sent.";function p(t){return typeof t=="object"&&t!==null&&!Array.isArray(t)}function Ne(t){if((t.method??"GET").toUpperCase()!=="POST")return  false;
+try{return new URL(t.url).pathname.replace(/\/+$/,"").endsWith("/v1/messages")}catch{return  false}}function A(t){return typeof t=="string"?[{type:"text",text:t}]:Array.isArray(t)?t.filter(p):[]}function T(t){
 return t.filter(e=>e.type==="text"&&typeof e.text=="string").map(e=>e.text).join(`
-`)}function me(t){if(typeof t=="string")return t;try{return JSON.stringify(t??{})}catch{return "{}"}}function je(t){let e=[],n=t.system;(typeof n=="string"?n!=="":Array.isArray(n))&&e.push({role:"syste\
-m",content:typeof n=="string"?n:M(A(n))});for(let r of Array.isArray(t.messages)?t.messages:[]){if(!p(r))continue;let o=A(r.content);if(r.role==="assistant"){let a=o.filter(i=>i.type==="tool_use").map(
-i=>({...typeof i.id=="string"?{id:i.id}:{},name:String(i.name??""),arguments:me(i.input)}));e.push({role:"assistant",content:M(o),...a.length>0?{toolCalls:a}:{}});continue}for(let a of o)a.type==="too\
-l_result"&&e.push({role:"tool",content:typeof a.content=="string"?a.content:M(A(a.content)),...typeof a.tool_use_id=="string"?{toolCallId:a.tool_use_id}:{}});let s=M(o);(s!==""||!o.some(a=>a.type==="t\
-ool_result"))&&e.push({role:"user",content:s});}return e}function pe(t){let e={};for(let[n,r]of Object.entries(t))n==="id"||n==="tool_use_id"||n==="text"&&t.type==="text"||(e[n]=r);return e}function qe(t,e){
-let n={};for(let[r,o]of Object.entries(t))r==="stream"||r==="metadata"||r==="messages"||r==="system"||(n[r]=o);return n.system=typeof t.system=="string"?"$text":A(t.system).map(pe),n.messages=(Array.isArray(
-t.messages)?t.messages:[]).filter(p).map(r=>({role:r.role,content:A(r.content).map(pe)})),e&&(n["anthropic-beta"]=e),n}function Ue(t){let e=t.tool_choice;return p(e)&&(e.type==="any"||e.type==="tool")||
-p(t.output_config)&&t.output_config.format!==void 0?true:t.output_format!==void 0}var O=class{id;texto="";razono=false;reproducible=true;parada;detallesDeParada;bloquesVistos=0;usoInicio;usoDelta;vioDelta=false;vioStop=false;error;herramientas=new Map;toolCallsJson=[];bloque(e){
+`)}function ve(t){if(typeof t=="string")return t;try{return JSON.stringify(t??{})}catch{return "{}"}}function Je(t){let e=[],n=t.system;(typeof n=="string"?n!=="":Array.isArray(n))&&e.push({role:"syste\
+m",content:typeof n=="string"?n:T(A(n))});for(let r of Array.isArray(t.messages)?t.messages:[]){if(!p(r))continue;let o=A(r.content);if(r.role==="assistant"){let a=o.filter(i=>i.type==="tool_use").map(
+i=>({...typeof i.id=="string"?{id:i.id}:{},name:String(i.name??""),arguments:ve(i.input)}));e.push({role:"assistant",content:T(o),...a.length>0?{toolCalls:a}:{}});continue}for(let a of o)a.type==="too\
+l_result"&&e.push({role:"tool",content:typeof a.content=="string"?a.content:T(A(a.content)),...typeof a.tool_use_id=="string"?{toolCallId:a.tool_use_id}:{}});let s=T(o);(s!==""||!o.some(a=>a.type==="t\
+ool_result"))&&e.push({role:"user",content:s});}return e}function me(t){let e={};for(let[n,r]of Object.entries(t))n==="id"||n==="tool_use_id"||n==="text"&&t.type==="text"||(e[n]=r);return e}function Fe(t,e){
+let n={};for(let[r,o]of Object.entries(t))r==="stream"||r==="metadata"||r==="messages"||r==="system"||(n[r]=o);return n.system=typeof t.system=="string"?"$text":A(t.system).map(me),n.messages=(Array.isArray(
+t.messages)?t.messages:[]).filter(p).map(r=>({role:r.role,content:A(r.content).map(me)})),e&&(n["anthropic-beta"]=e),n}function We(t){let e=t.tool_choice;return p(e)&&(e.type==="any"||e.type==="tool")||
+p(t.output_config)&&t.output_config.format!==void 0?true:t.output_format!==void 0}function Be(t){let e=typeof t.model=="string"?t.model:"";if(e==="")return [];let n=[],r=t.tool_choice;p(r)&&(r.type==="an\
+y"||r.type==="tool")&&Z(e)&&n.push({clave:`tool-choice-forzado:${e}`,mensaje:`[bvsala] Model "${e}" rejects a forced \`tool_choice\` (\`any\` or \`tool\`) with a 400, with or without thinking. This adapter \
+forwards your request unchanged, so the API will reject it: use \`tool_choice: {type: "auto"}\` with \`strict: true\` tools, or \`output_config.format\` for a fixed JSON shape. See https://platform.claude.c\
+om/docs/en/agents-and-tools/tool-use/define-tools#forcing-tool-use`});let o=["temperature","top_p","top_k"].filter(s=>typeof t[s]=="number"&&!(s==="temperature"&&t[s]===1));return o.length>0&&Q(e)&&n.
+push({clave:`muestreo:${e}`,mensaje:`[bvsala] Model "${e}" rejects a non-default ${o.map(s=>`\`${s}\``).join(", ")} with a 400. This adapter forwards your request unchanged, so the API will reject it:\
+ drop the parameter for this model. See https://platform.claude.com/docs/en/models/sonnet-5-5/overview`}),n}var M=class{id;texto="";razono=false;reproducible=true;parada;detallesDeParada;bloquesVistos=0;usoInicio;usoDelta;vioDelta=false;vioStop=false;error;herramientas=new Map;toolCallsJson=[];bloque(e){
 this.bloquesVistos+=1,V(e.type)?(this.razono=true,this.reproducible=false):e.type==="text"?Array.isArray(e.citations)&&e.citations.length>0&&(this.reproducible=false):e.type!=="tool_use"&&(this.reproducible=false);}json(e){
 this.id=typeof e.id=="string"?e.id:void 0;for(let n of A(e.content))this.bloque(n),n.type==="text"&&typeof n.text=="string"&&(this.texto+=n.text),n.type==="tool_use"&&this.toolCallsJson.push({...typeof n.
-id=="string"?{id:n.id}:{},name:String(n.name??""),arguments:me(n.input)});this.parada=typeof e.stop_reason=="string"?e.stop_reason:void 0,this.detallesDeParada=e.stop_details,this.usoInicio=p(e.usage)?
+id=="string"?{id:n.id}:{},name:String(n.name??""),arguments:ve(n.input)});this.parada=typeof e.stop_reason=="string"?e.stop_reason:void 0,this.detallesDeParada=e.stop_details,this.usoInicio=p(e.usage)?
 e.usage:void 0,this.usoDelta=this.usoInicio,this.vioDelta=true,this.vioStop=true;}evento(e){let n;try{n=JSON.parse(e.data);}catch{return}if(!p(n))return;switch(e.event!=="message"?e.event:String(n.type??e.event)){case "\
 message_start":{let o=p(n.message)?n.message:{};typeof o.id=="string"&&(this.id=o.id),p(o.usage)&&(this.usoInicio=o.usage);return}case "content_block_start":{let o=p(n.content_block)?n.content_block:{};
 this.bloque(o),o.type==="tool_use"&&typeof n.index=="number"&&this.herramientas.set(n.index,{...typeof o.id=="string"?{id:o.id}:{},name:String(o.name??""),json:""});return}case "content_block_delta":{let o=p(
@@ -69,36 +75,36 @@ n.delta)?n.delta:{};if(o.type==="text_delta"&&typeof o.text=="string")return thi
 "number"){let s=this.herramientas.get(n.index);s&&(s.json+=o.partial_json);}o.type==="citations_delta"&&(this.reproducible=false);return}case "message_delta":{this.vioDelta=true;let o=p(n.delta)?n.delta:{};typeof o.
 stop_reason=="string"&&(this.parada=o.stop_reason),"stop_details"in o&&(this.detallesDeParada=o.stop_details),p(n.usage)&&(this.usoDelta=q(this.usoDelta,n.usage));return}case "message_stop":this.vioStop=
 true;return;case "error":{let o=p(n.error)?n.error:{};this.error=`${String(o.type??"error")}: ${String(o.message??"")}`;return}default:return}}get toolCalls(){return this.toolCallsJson.length>0?this.toolCallsJson:
-[...this.herramientas.entries()].sort((e,n)=>e[0]-n[0]).map(([,e])=>({...e.id?{id:e.id}:{},name:e.name,arguments:Ie(e.json)})).filter(e=>e.name!=="")}respuesta(){let e=q(this.usoInicio,this.usoDelta),
-n=this.vioDelta?B(e,this.usoDelta?.output_tokens,this.razono).usage:{inputTokens:0,outputTokens:0,complete:false},r=e.iterations;Array.isArray(r)&&r.length>1&&(n.complete=false);let o=this.toolCalls,s=F(e);
+[...this.herramientas.entries()].sort((e,n)=>e[0]-n[0]).map(([,e])=>({...e.id?{id:e.id}:{},name:e.name,arguments:Ve(e.json)})).filter(e=>e.name!=="")}respuesta(){let e=q(this.usoInicio,this.usoDelta),
+n=this.vioDelta?W(e,this.usoDelta?.output_tokens,this.razono).usage:{inputTokens:0,outputTokens:0,complete:false},r=e.iterations;Array.isArray(r)&&r.length>1&&(n.complete=false);let o=this.toolCalls,s=B(e);
 return {text:this.texto,...o.length>0?{toolCalls:o}:{},...this.parada?{finishReason:G(this.parada)}:{},...this.vioDelta&&H({stopReason:this.parada,stopDetails:this.detallesDeParada,bloques:this.bloquesVistos,
-outputTokens:this.usoDelta?.output_tokens,iterations:r})?{unbilledRefusal:true}:{},...s?{servedServiceTier:s}:{},usage:n,...this.reproducible&&this.parada==="end_turn"?{}:{replayable:false}}}};function Ie(t){
-try{return JSON.stringify(JSON.parse(t===""?"{}":t))}catch{return t}}function ze(t,e){let n=[];return t.text!==""&&n.push({type:"text",text:t.text}),(t.toolCalls??[]).forEach((r,o)=>{let s={};try{s=JSON.
+outputTokens:this.usoDelta?.output_tokens,iterations:r})?{unbilledRefusal:true}:{},...s?{servedServiceTier:s}:{},usage:n,...this.reproducible&&this.parada==="end_turn"?{}:{replayable:false}}}};function Ve(t){
+try{return JSON.stringify(JSON.parse(t===""?"{}":t))}catch{return t}}function Ge(t,e){let n=[];return t.text!==""&&n.push({type:"text",text:t.text}),(t.toolCalls??[]).forEach((r,o)=>{let s={};try{s=JSON.
 parse(r.arguments||"{}");}catch{s={};}n.push({type:"tool_use",id:r.id??`toolu_bvsala_${t.requestId}_${o}`,name:r.name,input:s});}),{id:`msg_bvsala_${t.requestId}`,type:"message",role:"assistant",model:e,
-content:n,stop_reason:(t.toolCalls?.length??0)>0?"tool_use":"end_turn",stop_sequence:null,usage:{input_tokens:0,output_tokens:0}}}function fe(t,e,n){let r=ze(t,e),o={"request-id":`bvsala_${t.requestId}`};
-if(!n)return new Response(JSON.stringify(r),{status:200,headers:{...o,"content-type":"application/json"}});let s=(i,u)=>`event: ${i}
-data: ${JSON.stringify(u)}
+content:n,stop_reason:(t.toolCalls?.length??0)>0?"tool_use":"end_turn",stop_sequence:null,usage:{input_tokens:0,output_tokens:0}}}function he(t,e,n){let r=Ge(t,e),o={"request-id":`bvsala_${t.requestId}`};
+if(!n)return new Response(JSON.stringify(r),{status:200,headers:{...o,"content-type":"application/json"}});let s=(i,l)=>`event: ${i}
+data: ${JSON.stringify(l)}
 
-`,a=s("message_start",{type:"message_start",message:{...r,content:[],stop_reason:null,stop_sequence:null}});return r.content.forEach((i,u)=>{i.type==="text"?(a+=s("content_block_start",{type:"content_\
-block_start",index:u,content_block:{type:"text",text:""}}),a+=s("content_block_delta",{type:"content_block_delta",index:u,delta:{type:"text_delta",text:i.text}})):(a+=s("content_block_start",{type:"co\
-ntent_block_start",index:u,content_block:{type:"tool_use",id:i.id,name:i.name,input:{}}}),a+=s("content_block_delta",{type:"content_block_delta",index:u,delta:{type:"input_json_delta",partial_json:JSON.
-stringify(i.input)}})),a+=s("content_block_stop",{type:"content_block_stop",index:u});}),a+=s("message_delta",{type:"message_delta",delta:{stop_reason:r.stop_reason,stop_sequence:null},usage:{output_tokens:0}}),
-a+=s("message_stop",{type:"message_stop"}),new Response(a,{status:200,headers:{...o,"content-type":"text/event-stream"}})}var T=class extends Error{constructor(n){super(`[bvsala] provider responded ${n.
-status}`);this.respuesta=n;}respuesta};function lt(t){let e=se("anthropic-sdk",t),n=Math.max(1,Math.floor(t.reportCacheSize??1e3)),r=new Map,o=(a,i)=>{if(a)for(r.delete(a),r.set(a,i);r.size>n;)r.delete(
-r.keys().next().value);};return {middleware:async(a,i,u)=>{if(!Ee(a))return i(a);if(typeof a.body!="string")throw new Error(I);let d;try{let c=JSON.parse(a.body);if(!p(c))throw new Error(I);d=c;}catch{throw new Error(
-I)}let g=u.options,m=de(g?.bivelio),h=t.context?.({body:d,requestOptions:g}),_=(Array.isArray(d.tools)?d.tools:[]).filter(p).map(c=>ce(t,{nombre:String(c.name??""),descripcion:typeof c.description=="s\
-tring"?c.description:void 0,parametros:c.input_schema,delProveedor:c.input_schema===void 0&&typeof c.type=="string"&&c.type!=="custom"})),l=await ue(t,{model:`anthropic/${String(d.model??"")}`,messages:je(
-d),tools:_,temperature:typeof d.temperature=="number"?d.temperature:void 0,context:le(h,m.context),savings:m.savings,proyeccion:qe(d,a.headers.get("anthropic-beta")),restricciones:{sinResolvers:Ue(d)}}),
-v={model:l.model,messages:l.messages,...l.tools?{tools:l.tools}:{}},b=K(a.url),R=async c=>{let f;try{f=await i(a);}catch(y){throw c.errorDelMarco={error:y},y}if(!f.ok){let y=new T(f);throw c.errorDelMarco=
-{error:y},y}return c.completada=true,f},z=c=>{if(c instanceof T)return c.respuesta;throw c};if(d.stream!==true){let c,f,y=U(v,async()=>{c=await R(y);let D=new O;return D.json(await c.clone().json()),f=D.id,
-D.respuesta()},void 0,b),w;try{w=await e.generar(l,y);}catch(D){return z(D)}if(w.tipo==="sin-informe")return c;let N=e.informe(w.result.report);return w.tipo==="atajo"?(o(`msg_bvsala_${w.result.requestId}`,
-N),fe(w.result,d.model,false)):(o(f,N),c)}let{abierta:he,abrir:$}=ae(),E,j,k=Object.assign(U(v,async()=>{let c=await R(k),f=new C;j=ge(c,f,e,k,o),$();for await(let y of f)if(y.type==="done")return y.response;
-throw new Error("[bvsala] the model stream ended before completion; nothing was measured.")},()=>{let c=new C;return R(k).then(f=>{E=ge(f,c,e,k,o),$();},f=>c.fail(f)),c},b),{abierta:he}),S;try{S=await e.
-transmitir(l,k,a.signal??void 0);}catch(c){return z(c)}if(S.tipo==="abierto")return E??j;if(S.tipo==="atajo"){let c=e.informe(S.result.report);return o(`msg_bvsala_${S.result.requestId}`,c),fe(S.result,
-d.model,true)}return E??j},reportFor(a){let i=typeof a=="string"?a:a?.id;return typeof i=="string"?r.get(i):void 0},close(){e.close();}}}function ge(t,e,n,r,o){let s=new O,a,i=async()=>{try{let l=await r.
-resultado;o(s.id,n.informe(l.report));}catch{}},u=()=>{e.cerrada||(s.error?e.fail(new Error(`[bvsala] provider stream error: ${s.error}`)):s.vioDelta||s.vioStop?(e.push({type:"done",response:s.respuesta()}),
-e.end(),a=i()):e.fail(new Error("[bvsala] the model stream ended before completion; nothing was measured.")));};if((t.headers.get("content-type")??"").includes("application/json"))return t.clone().json().
-then(l=>{s.json(p(l)?l:{}),s.texto!==""&&e.push({type:"text",delta:s.texto}),u();},l=>e.fail(l)),t;let d=t.body;if(!d)return e.fail(new Error("[bvsala] the model stream had no body; nothing was measure\
-d.")),t;let g=d.getReader(),m=new L,h=l=>{for(let v of l){let b=s.evento(v);b!==void 0&&e.push({type:"text",delta:b}),(s.vioStop||s.error)&&u();}},_=new ReadableStream({async pull(l){let v;try{v=await g.
-read();}catch(R){e.fail(R),l.error(R);return}if(v.done){h(m.end()),u(),await a,l.close();return}let b=v.value;l.enqueue(b),h(m.push(b));},cancel(l){return e.fail(new Error("[bvsala] the stream was cance\
-lled by its reader; nothing was measured.")),g.cancel(l)}});return new Response(_,{status:t.status,statusText:t.statusText,headers:t.headers})}export{lt as anthropicSavings};
+`,a=s("message_start",{type:"message_start",message:{...r,content:[],stop_reason:null,stop_sequence:null}});return r.content.forEach((i,l)=>{i.type==="text"?(a+=s("content_block_start",{type:"content_\
+block_start",index:l,content_block:{type:"text",text:""}}),a+=s("content_block_delta",{type:"content_block_delta",index:l,delta:{type:"text_delta",text:i.text}})):(a+=s("content_block_start",{type:"co\
+ntent_block_start",index:l,content_block:{type:"tool_use",id:i.id,name:i.name,input:{}}}),a+=s("content_block_delta",{type:"content_block_delta",index:l,delta:{type:"input_json_delta",partial_json:JSON.
+stringify(i.input)}})),a+=s("content_block_stop",{type:"content_block_stop",index:l});}),a+=s("message_delta",{type:"message_delta",delta:{stop_reason:r.stop_reason,stop_sequence:null},usage:{output_tokens:0}}),
+a+=s("message_stop",{type:"message_stop"}),new Response(a,{status:200,headers:{...o,"content-type":"text/event-stream"}})}var E=class extends Error{constructor(n){super(`[bvsala] provider responded ${n.
+status}`);this.respuesta=n;}respuesta};function bt(t){let e=ce("anthropic-sdk",t),n=Math.max(1,Math.floor(t.reportCacheSize??1e3)),r=new Map,o=(a,i)=>{if(a)for(r.delete(a),r.set(a,i);r.size>n;)r.delete(
+r.keys().next().value);};return {middleware:async(a,i,l)=>{if(!Ne(a))return i(a);if(typeof a.body!="string")throw new Error(U);let d;try{let c=JSON.parse(a.body);if(!p(c))throw new Error(U);d=c;}catch{throw new Error(
+U)}for(let c of Be(d))e.avisar(c.clave,c.mensaje);let g=l.options,m=ge(g?.bivelio),h=t.context?.({body:d,requestOptions:g}),b=(Array.isArray(d.tools)?d.tools:[]).filter(p).map(c=>de(t,{nombre:String(c.
+name??""),descripcion:typeof c.description=="string"?c.description:void 0,parametros:c.input_schema,delProveedor:c.input_schema===void 0&&typeof c.type=="string"&&c.type!=="custom"})),u=await pe(t,{model:`\
+anthropic/${String(d.model??"")}`,messages:Je(d),tools:b,temperature:typeof d.temperature=="number"?d.temperature:void 0,context:fe(h,m.context),savings:m.savings,proyeccion:Fe(d,a.headers.get("anthro\
+pic-beta")),restricciones:{sinResolvers:We(d)}}),v={model:u.model,messages:u.messages,...u.tools?{tools:u.tools}:{}},_=X(a.url),R=async c=>{let f;try{f=await i(a);}catch(y){throw c.errorDelMarco={error:y},
+y}if(!f.ok){let y=new E(f);throw c.errorDelMarco={error:y},y}return c.completada=true,f},z=c=>{if(c instanceof E)return c.respuesta;throw c};if(d.stream!==true){let c,f,y=I(v,async()=>{c=await R(y);let D=new M;
+return D.json(await c.clone().json()),f=D.id,D.respuesta()},void 0,_),S;try{S=await e.generar(u,y);}catch(D){return z(D)}if(S.tipo==="sin-informe")return c;let N=e.informe(S.result.report);return S.tipo===
+"atajo"?(o(`msg_bvsala_${S.result.requestId}`,N),he(S.result,d.model,false)):(o(f,N),c)}let{abierta:_e,abrir:$}=le(),L,j,k=Object.assign(I(v,async()=>{let c=await R(k),f=new C;j=ye(c,f,e,k,o),$();for await(let y of f)
+if(y.type==="done")return y.response;throw new Error("[bvsala] the model stream ended before completion; nothing was measured.")},()=>{let c=new C;return R(k).then(f=>{L=ye(f,c,e,k,o),$();},f=>c.fail(f)),
+c},_),{abierta:_e}),w;try{w=await e.transmitir(u,k,a.signal??void 0);}catch(c){return z(c)}if(w.tipo==="abierto")return L??j;if(w.tipo==="atajo"){let c=e.informe(w.result.report);return o(`msg_bvsala_${w.
+result.requestId}`,c),he(w.result,d.model,true)}return L??j},reportFor(a){let i=typeof a=="string"?a:a?.id;return typeof i=="string"?r.get(i):void 0},close(){e.close();}}}function ye(t,e,n,r,o){let s=new M,
+a,i=async()=>{try{let u=await r.resultado;o(s.id,n.informe(u.report));}catch{}},l=()=>{e.cerrada||(s.error?e.fail(new Error(`[bvsala] provider stream error: ${s.error}`)):s.vioDelta||s.vioStop?(e.push(
+{type:"done",response:s.respuesta()}),e.end(),a=i()):e.fail(new Error("[bvsala] the model stream ended before completion; nothing was measured.")));};if((t.headers.get("content-type")??"").includes("ap\
+plication/json"))return t.clone().json().then(u=>{s.json(p(u)?u:{}),s.texto!==""&&e.push({type:"text",delta:s.texto}),l();},u=>e.fail(u)),t;let d=t.body;if(!d)return e.fail(new Error("[bvsala] the mode\
+l stream had no body; nothing was measured.")),t;let g=d.getReader(),m=new O,h=u=>{for(let v of u){let _=s.evento(v);_!==void 0&&e.push({type:"text",delta:_}),(s.vioStop||s.error)&&l();}},b=new ReadableStream(
+{async pull(u){let v;try{v=await g.read();}catch(R){e.fail(R),u.error(R);return}if(v.done){h(m.end()),l(),await a,u.close();return}let _=v.value;u.enqueue(_),h(m.push(_));},cancel(u){return e.fail(new Error(
+"[bvsala] the stream was cancelled by its reader; nothing was measured.")),g.cancel(u)}});return new Response(b,{status:t.status,statusText:t.statusText,headers:t.headers})}export{bt as anthropicSavings};