@bivelio/savings-layer 0.4.26 → 0.4.28

package/README.md

@@ -256,6 +256,19 @@ exists, as for Anthropic models), and `"unmeasured"` when the provider returned
 usage. Only `measured` and `reference` figures reach your invoice; estimated ones
 never do.
 
+A repeat is valued by reference only when the call that ran reached the provider's
+own API (`https://api.openai.com`, `https://api.anthropic.com` or
+`https://generativelanguage.googleapis.com`), because the list price it is valued at
+is the invoice only there. When `OpenAICompatibleProvider` or `AnthropicProvider`
+points anywhere else — a proxy such as LiteLLM, a gateway, Azure, Bedrock, Vertex —
+or when the Anthropic SDK or LangChain adapter reads that its client does, that call
+leaves no reference: its repeats stay `estimated`, and the `reference` trace of the
+call that ran says `unofficial-origin`. When the layer cannot tell where the call
+went (a provider of your own, the AI SDK and Mastra adapters, a LangChain model that
+does not expose its base URL), the rule applies as it always has. The layer reads the
+configured base URL, not the wire: a custom `fetch` that sends the request elsewhere
+is not seen.
+
 **Streaming.** `generateStream()` takes the same request and hands you the answer as
 the provider writes it, then the same `SavingsReport`:
 
@@ -679,6 +692,77 @@ const placedLayer = createSavingsLayer({
 
 </details>
 
+<details>
+<summary><b>Serverless: keep what the layer learns between instances</b></summary>
+
+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:
+
+```ts
+import { InMemorySemanticCache } from "@bivelio/savings-layer";
+
+// Any key-value client your instances share, such as Redis.
+type SharedKeyValue = {
+  get(key: string): Promise<unknown>;
+  set(key: string, value: string, options?: { px: number }): Promise<unknown>;
+};
+
+async function serverlessLayer(kv: SharedKeyValue) {
+  const semanticCache = new InMemorySemanticCache({ calibration: { maxErrorRate: 0.02 } });
+  const saved = await kv.get("bvsala:semantic-calibration");
+  if (saved) semanticCache.importCalibration(saved); // the object or its JSON text
+
+  const layer = createSavingsLayer({
+    provider,
+    licenseKey: process.env.BIVELIO_LICENSE_KEY,
+    semanticCache,
+    cacheBreakpoints: "auto",
+    cacheBreakpointStore: {
+      get: (key) => kv.get(key),
+      set: (key, value, ttlMs) => kv.set(key, value, { px: ttlMs }),
+    },
+  });
+
+  // After handling requests with `layer`, save the calibration:
+  const save = async () => {
+    await semanticCache.settled();
+    await kv.set("bvsala:semantic-calibration", JSON.stringify(semanticCache.exportCalibration()));
+  };
+  return { layer, save };
+}
+```
+
+- **Breakpoint placement.** The store holds prompt-prefix fingerprints, timestamps and
+  counts, one key per traffic class, never prompt text. The layer reads it before
+  placing breakpoints and reads and rewrites it after the provider answers, so every
+  instance extends the same measurement. If the store fails, that request is placed as
+  a fresh instance would place it and nothing is written over what others learned.
+  Without a store nothing changes.
+- **Calibration.** The saved state is the whole experiment: both halves of the labels,
+  how many tests have already drawn on the confidence budget, and the floor in force.
+  Restoring it continues the same calibration, so the bound keeps holding across
+  restarts; it never starts the budget over. A state recorded under another embedder,
+  verifier, judge or configuration is refused with `CalibrationStateError`. If you pass
+  your own `embed`, `verify` or `judge`, name them (`embedderId`, `verifierId`,
+  `calibration.judgeId`) so a state can say what it was recorded with.
+- **One line of history.** The bound holds along one line: restore the latest state and
+  save from where it went. Instances that continue the same state at the same time
+  each spend its remaining confidence again, and the last one to save overwrites the
+  others' labels. For one bound across many concurrent instances, calibrate in a single
+  long-lived process, or offline with `calibrateSemanticFloor()`, and give every
+  instance the certified floor as `acceptThreshold`. The cached answers themselves
+  still live in each instance's memory.
+- **Why a predicted cache read did not happen** (Anthropic's own API). With
+  `cacheMissDiagnostics: true`, the layer asks Anthropic to compare each marked request
+  with the one whose cache entry it expected to read, and puts the provider's
+  `cache_miss_reason` in `report.traces` when that read does not arrive. Anthropic
+  keeps a short-lived fingerprint (hashes and token-count estimates) of each request
+  that opts in; check that against your data-retention terms first.
+
+</details>
+
 <details>
 <summary><b>Structured output — validated <code>data</code>, not just text</b></summary>
 
@@ -710,10 +794,11 @@ console.log(data.answer); // 8
 
 When you pass `tools` and the model asks to invoke one, the reply is not the final
 answer: `text` is empty by design and the request is in `toolCalls`. Execute it,
-append the result to `messages`, and call `generate()` again.
+append the assistant turn and one `role: "tool"` message per call to `messages`, and
+call `generate()` again.
 
 ```ts
-import type { ToolDescriptor } from "@bivelio/savings-layer";
+import type { ChatMessage, ToolDescriptor } from "@bivelio/savings-layer";
 
 const tools: ToolDescriptor[] = [
   {
@@ -730,23 +815,33 @@ const tools: ToolDescriptor[] = [
   },
 ];
 
-const res = await layer.generate({
-  model: "openai/gpt-4o-mini",
-  messages: [{ role: "user", content: "What's the weather in Madrid?" }],
-  tools,
-});
+// ← your tool
+const getWeather = async (city: string): Promise<string> => `${city}: 21 °C, clear`;
 
-if (res.toolCalls?.length) {
-  // res.finishReason === "tool_calls"
+const messages: ChatMessage[] = [{ role: "user", content: "What's the weather in Madrid?" }];
+let res = await layer.generate({ model: "openai/gpt-4o-mini", messages, tools });
+
+while (res.toolCalls?.length) {
+  // res.finishReason === "tool_calls". Send the calls back as they came, ids included.
+  messages.push({ role: "assistant", content: res.text, toolCalls: res.toolCalls });
   for (const call of res.toolCalls) {
     const args = JSON.parse(call.arguments) as { city: string };
-    console.log(call.name, args); // ← run your tool here, then append its result
+    const output = await getWeather(args.city);
+    messages.push({ role: "tool", toolCallId: call.id, name: call.name, content: output });
   }
-} else {
-  console.log(res.text);
+  res = await layer.generate({ model: "openai/gpt-4o-mini", messages, tools });
 }
+console.log(res.text);
 ```
 
+- **Ids.** Each result carries the `toolCallId` of the call it answers, and every call
+  of a turn gets its result before the next user or assistant message. On Anthropic
+  that is what lets the turn travel as `tool_use`/`tool_result` blocks (with the tool
+  still in that request's `tools`); otherwise it is sent as plain text, as before, and
+  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.
+
 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.
 
@@ -767,6 +862,12 @@ cached prompt prefix stays the same from one question to the next.
   search on every request even with a warm prompt cache. `"always"` skips those checks
   (for example behind a proxy that forwards the search blocks). Requests with an output
   contract, or with the model cascade on, keep the selector.
+- **Across turns.** Append the assistant turn with its `toolCalls` (ids as returned)
+  before the tool results, as in any tool conversation. The Anthropic adapter then sends
+  the earlier search results back in that turn, so the model calls the tools it already
+  found without searching again. It keeps them in memory, per adapter instance and
+  bounded; a turn it no longer remembers (another process, say) goes out as before and
+  the model searches again.
 - **Whose saving.** The provider's. Your usage counter shows it; the layer credits
   nothing for it and charges no commission on it.
 

package/dist/adapters/ai-sdk.js

@@ -1,84 +1,84 @@
-import {createSavingsLayer,LicenseError,sha256,stableStringify}from'@bivelio/savings-layer';import {AsyncLocalStorage}from'async_hooks';var j=new Set;function $(t,e,n){t===true||j.has(e)||(j.add(e),console.warn(n));}var ae="[bvsala] internal: the layer changed a request this adapter must forward unchanged; no\
-thing was sent.",se="[bvsala] internal: the layer asked for a second model call within one framework call; nothing more was sent.",ie="[bvsala] internal: the layer called the model outside a framework\
- call; nothing was sent.";function C(t,e,n){return {renderizado:t,generar:e,...n?{transmitir:n}:{},invocada:false,completada:false}}async function x(t,e){try{let n=await e();return t.completada=!0,n}catch(n){
-throw t.errorDelMarco={error:n},n}}function de(t,e){if(t===e)return  true;try{return stableStringify(t)===stableStringify(e)}catch{return  false}}function N(t,e){let n=t??[],r=e??[];return n.length===r.length&&n.every((o,i)=>de(o,r[i]))}
-function ue(t,e){return t.model===e.model&&N(t.messages,e.messages)&&N(t.tools,e.tools)}function G(t,e){let n=r=>{let o=e.getStore();if(!o)throw new Error(ie);if(o.invocada)throw new Error(se);if(!ue(
-r,o.renderizado))throw new Error(ae);return o.invocada=true,o};return {id:`bivelio-adapter:${t}`,async generate(r,o){return n(r).generar(o?.signal)},async*generateStream(r,o){let i=n(r);if(i.transmitir){
-yield*i.transmitir(o?.signal);return}let a=await i.generar(o?.signal);a.text!==""&&(yield {type:"text",delta:a.text}),yield {type:"done",response:a};}}}function U(){return new AsyncLocalStorage}var fe=["licenseKey","licenseServerUrl","serviceAccountKey","licenseRefresh","licenseRevocation","licensePollIntervalSeconds","licenseFetch","metering","meteringUrl","meteringFetch","costModels","defa\
-ultCostModel","exactCache","exactCacheTtlMs","resolvers","exactTokenizer","silenceWarnings","failOpen"];function _(t,e){return e instanceof LicenseError||t.errorDelMarco&&t.errorDelMarco.error===e?false:t.invocada&&
-t.completada}var F="[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 W(t,e){
-let n=U(),r={};for(let a of fe)e[a]!==void 0&&(r[a]=e[a]);let o=createSavingsLayer({...r,provider:G(t,n)}),i=(a,s)=>$(e.silenceWarnings,`${t}:${a}`,s);return {ejecutar(a,s){return n.run(s,()=>o.generate(a))},async generar(a,s){
-try{let d=await n.run(s,()=>o.generate(a));return s.invocada?{tipo:"vivo",result:d}:{tipo:"atajo",result:d}}catch(d){if(_(s,d))return i("sin-informe",F),{tipo:"sin-informe"};throw d}},async transmitir(a,s,d){
-let u=n.run(s,()=>o.generateStream(a,d?{signal:d}:{})).result;s.resultado=u,u.catch(()=>{});let c;try{c=await Promise.race([s.abierta.then(()=>"abierta"),u.then(()=>"resuelta")]);}catch(l){if(_(s,l))return i(
-"sin-informe",F),{tipo:"sin-informe"};throw l}if(c==="abierta")return {tipo:"abierto"};let g=await u;return s.invocada?{tipo:"entero",result:g}:{tipo:"atajo",result:g}},informe(a){let s=JSON.parse(JSON.
-stringify(a));if(e.onReport)try{e.onReport(s);}catch{i("onreport","[bvsala] `onReport` threw; the error was caught and the model call was not affected.");}return s},avisar:i,close(){o.close();}}}function B(){
+import {createSavingsLayer,LicenseError,sha256,stableStringify}from'@bivelio/savings-layer';import {AsyncLocalStorage}from'async_hooks';var j=new Set;function $(t,e,n){t===true||j.has(e)||(j.add(e),console.warn(n));}var N=Symbol.for("@bivelio/savings-layer:provider-origin");var ie="[bvsala] internal: the layer changed a request this adapter must forward unchanged; nothing was sent.",se="[bvsala] internal: the layer asked for a second model call within one framework call;\
+ nothing more was sent.",de="[bvsala] internal: the layer called the model outside a framework call; nothing was sent.";function C(t,e,n,r){return {renderizado:t,generar:e,...n?{transmitir:n}:{},invocada:false,
+completada:false,...{}}}async function x(t,e){try{let n=await e();return t.completada=!0,n}catch(n){throw t.errorDelMarco={error:n},n}}function ue(t,e){if(t===e)return  true;try{return stableStringify(
+t)===stableStringify(e)}catch{return  false}}function G(t,e){let n=t??[],r=e??[];return n.length===r.length&&n.every((a,i)=>ue(a,r[i]))}function le(t,e){return t.model===e.model&&G(t.messages,e.messages)&&G(t.tools,e.tools)}
+function _(t,e){let n=a=>{let i=e.getStore();if(!i)throw new Error(de);if(i.invocada)throw new Error(se);if(!le(a,i.renderizado))throw new Error(ie);return i.invocada=true,i},r={id:`bivelio-adapter:${t}`,
+async generate(a,i){return n(a).generar(i?.signal)},async*generateStream(a,i){let o=n(a);if(o.transmitir){yield*o.transmitir(i?.signal);return}let s=await o.generar(i?.signal);s.text!==""&&(yield {type:"\
+text",delta:s.text}),yield {type:"done",response:s};}};return Object.defineProperty(r,N,{value:()=>e.getStore()?.origen}),r}function U(){return new AsyncLocalStorage}var fe=["licenseKey","licenseServerUrl","serviceAccountKey","licenseRefresh","licenseRevocation","licensePollIntervalSeconds","licenseFetch","metering","meteringUrl","meteringFetch","costModels","defa\
+ultCostModel","exactCache","exactCacheTtlMs","resolvers","exactTokenizer","silenceWarnings","failOpen"];function F(t,e){return e instanceof LicenseError||t.errorDelMarco&&t.errorDelMarco.error===e?false:t.invocada&&
+t.completada}var V="[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 W(t,e){
+let n=U(),r={};for(let o of fe)e[o]!==void 0&&(r[o]=e[o]);let a=createSavingsLayer({...r,provider:_(t,n)}),i=(o,s)=>$(e.silenceWarnings,`${t}:${o}`,s);return {ejecutar(o,s){return n.run(s,()=>a.generate(o))},async generar(o,s){
+try{let d=await n.run(s,()=>a.generate(o));return s.invocada?{tipo:"vivo",result:d}:{tipo:"atajo",result:d}}catch(d){if(F(s,d))return i("sin-informe",V),{tipo:"sin-informe"};throw d}},async transmitir(o,s,d){
+let u=n.run(s,()=>a.generateStream(o,d?{signal:d}:{})).result;s.resultado=u,u.catch(()=>{});let c;try{c=await Promise.race([s.abierta.then(()=>"abierta"),u.then(()=>"resuelta")]);}catch(l){if(F(s,l))return i(
+"sin-informe",V),{tipo:"sin-informe"};throw l}if(c==="abierta")return {tipo:"abierto"};let g=await u;return s.invocada?{tipo:"entero",result:g}:{tipo:"atajo",result:g}},informe(o){let s=JSON.parse(JSON.
+stringify(o));if(e.onReport)try{e.onReport(s);}catch{i("onreport","[bvsala] `onReport` threw; the error was caught and the model call was not affected.");}return s},avisar:i,close(){a.close();}}}function B(){
 let t;return {abierta:new Promise(n=>{t=n;}),abrir:t}}var b=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}))}}};function p(t){return typeof t=="object"&&t!==null&&!Array.isArray(t)}function J(t){return typeof t=="string"?[{type:"text",text:t}]:Array.isArray(t)?t.filter(p):[]}function D(t){return t.filter(e=>e.type===
+(e,n)=>{this.esperando={resolver:e,rechazar:n};}),return:()=>(this.terminada=true,this.pendientes.length=0,Promise.resolve({value:void 0,done:true}))}}};function f(t){return typeof t=="object"&&t!==null&&!Array.isArray(t)}function J(t){return typeof t=="string"?[{type:"text",text:t}]:Array.isArray(t)?t.filter(f):[]}function A(t){return t.filter(e=>e.type===
 "text"&&typeof e.text=="string").map(e=>e.text).join(`
-`)}function v(t){if(typeof t=="string")return t;try{return JSON.stringify(t??{})}catch{return "{}"}}function pe(t){if(!p(t))return v(t);switch(t.type){case "text":case "error-text":return typeof t.value==
-"string"?t.value:v(t.value);case "json":case "error-json":return v(t.value);case "execution-denied":return `execution denied${typeof t.reason=="string"?`: ${t.reason}`:""}`;case "content":return D(e(t.value));default:
-return v(t)}function e(n){return Array.isArray(n)?n.filter(p):[]}}function V(t){let e=[];for(let n of Array.isArray(t)?t:[]){if(!p(n))continue;let r=J(n.content);switch(n.role){case "system":e.push({role:"\
-system",content:typeof n.content=="string"?n.content:D(r)});break;case "user":e.push({role:"user",content:D(r)});break;case "assistant":{let o=r.filter(i=>i.type==="tool-call"&&i.providerExecuted!==true).
-map(i=>({...typeof i.toolCallId=="string"?{id:i.toolCallId}:{},name:String(i.toolName??""),arguments:v(i.input??i.args)}));e.push({role:"assistant",content:D(r),...o.length>0?{toolCalls:o}:{}});break}case "\
-tool":for(let o of r)o.type==="tool-result"&&e.push({role:"tool",content:pe(o.output??o.result),...typeof o.toolCallId=="string"?{toolCallId:o.toolCallId}:{},...typeof o.toolName=="string"?{name:o.toolName}:
-{}});break}}return e}var ge=new Set(["bivelio","mastra"]);function k(t){if(!p(t))return t;let e={};for(let[n,r]of Object.entries(t))ge.has(n)||(e[n]=r);return e}function me(t){let e={};for(let[n,r]of Object.
-entries(t))if(!(n==="toolCallId"||n==="id")&&!(n==="text"&&t.type==="text")){if(n==="providerOptions"||n==="providerMetadata"){e[n]=k(r);continue}e[n]=r;}return e}function ve(t){return (Array.isArray(t)?
-t:[]).map(e=>p(e)?{role:e.role,content:typeof e.content=="string"?"$text":J(e.content).map(me),providerOptions:k(e.providerOptions)}:e)}var ye=new Set(["anthropic-beta","openai-beta"]);function he(t){
-if(!p(t))return;let e={};for(let[n,r]of Object.entries(t))ye.has(n.toLowerCase())&&r!==void 0&&(e[n.toLowerCase()]=r);return Object.keys(e).length>0?e:void 0}function K(t){let e={};for(let[n,r]of Object.
-entries(t))n==="prompt"||n==="abortSignal"||n==="headers"||n==="providerOptions"||(e[n]=r);return e.prompt=ve(t.prompt),e.headers=he(t.headers),e.providerOptions=k(t.providerOptions),e}function H(t){let e=t.
-responseFormat;if(p(e)&&e.type==="json")return  true;let n=t.toolChoice;return p(n)&&(n.type==="required"||n.type==="tool")?true:n==="required"}function m(t){return typeof t=="number"&&Number.isFinite(t)?t:
-void 0}function Re(t,e){let n=p(t)?t:{},r=p(n.inputTokens)?n.inputTokens:void 0,o=p(n.outputTokens)?n.outputTokens:void 0;if(!r||!o){let c=m(n.inputTokens),g=m(n.outputTokens);return {inputTokens:c??0,
-outputTokens:g??0,complete:false}}let i=m(r.total),a=m(r.cacheRead),s=m(r.cacheWrite),d=m(o.total),f=m(o.reasoning);return {inputTokens:i??0,outputTokens:d??0,...a!==void 0?{cachedInputTokens:a}:{},...s!==
-void 0?{cacheWriteInputTokens:s}:{},...f!==void 0?{reasoningTokens:f}:{},...i!==void 0&&d!==void 0&&!e?{}:{complete:false}}}function we(t){if(typeof t=="string")return t;if(p(t)&&typeof t.unified=="strin\
-g")return t.unified}function Se(t){switch(t){case void 0:return;case "tool-calls":return "tool_calls";case "content-filter":return "content_filter";default:return t}}function xe(t){return p(t)&&Object.keys(
-k(t)).length>0}var R=class{texto="";toolCalls=[];reproducible=true;finish;error;anadir(e){switch(e.type){case "text-delta":{let n=typeof e.delta=="string"?e.delta:typeof e.textDelta=="string"?e.textDelta:
+`)}function v(t){if(typeof t=="string")return t;try{return JSON.stringify(t??{})}catch{return "{}"}}function ge(t){if(!f(t))return v(t);switch(t.type){case "text":case "error-text":return typeof t.value==
+"string"?t.value:v(t.value);case "json":case "error-json":return v(t.value);case "execution-denied":return `execution denied${typeof t.reason=="string"?`: ${t.reason}`:""}`;case "content":return A(e(t.value));default:
+return v(t)}function e(n){return Array.isArray(n)?n.filter(f):[]}}function H(t){let e=[];for(let n of Array.isArray(t)?t:[]){if(!f(n))continue;let r=J(n.content);switch(n.role){case "system":e.push({role:"\
+system",content:typeof n.content=="string"?n.content:A(r)});break;case "user":e.push({role:"user",content:A(r)});break;case "assistant":{let a=r.filter(i=>i.type==="tool-call"&&i.providerExecuted!==true).
+map(i=>({...typeof i.toolCallId=="string"?{id:i.toolCallId}:{},name:String(i.toolName??""),arguments:v(i.input??i.args)}));e.push({role:"assistant",content:A(r),...a.length>0?{toolCalls:a}:{}});break}case "\
+tool":for(let a of r)a.type==="tool-result"&&e.push({role:"tool",content:ge(a.output??a.result),...typeof a.toolCallId=="string"?{toolCallId:a.toolCallId}:{},...typeof a.toolName=="string"?{name:a.toolName}:
+{}});break}}return e}var me=new Set(["bivelio","mastra"]);function D(t){if(!f(t))return t;let e={};for(let[n,r]of Object.entries(t))me.has(n)||(e[n]=r);return e}function ve(t){let e={};for(let[n,r]of Object.
+entries(t))if(!(n==="toolCallId"||n==="id")&&!(n==="text"&&t.type==="text")){if(n==="providerOptions"||n==="providerMetadata"){e[n]=D(r);continue}e[n]=r;}return e}function ye(t){return (Array.isArray(t)?
+t:[]).map(e=>f(e)?{role:e.role,content:typeof e.content=="string"?"$text":J(e.content).map(ve),providerOptions:D(e.providerOptions)}:e)}var he=new Set(["anthropic-beta","openai-beta"]);function Re(t){
+if(!f(t))return;let e={};for(let[n,r]of Object.entries(t))he.has(n.toLowerCase())&&r!==void 0&&(e[n.toLowerCase()]=r);return Object.keys(e).length>0?e:void 0}function K(t){let e={};for(let[n,r]of Object.
+entries(t))n==="prompt"||n==="abortSignal"||n==="headers"||n==="providerOptions"||(e[n]=r);return e.prompt=ye(t.prompt),e.headers=Re(t.headers),e.providerOptions=D(t.providerOptions),e}function Z(t){let e=t.
+responseFormat;if(f(e)&&e.type==="json")return  true;let n=t.toolChoice;return f(n)&&(n.type==="required"||n.type==="tool")?true:n==="required"}function m(t){return typeof t=="number"&&Number.isFinite(t)?t:
+void 0}function we(t,e){let n=f(t)?t:{},r=f(n.inputTokens)?n.inputTokens:void 0,a=f(n.outputTokens)?n.outputTokens:void 0;if(!r||!a){let c=m(n.inputTokens),g=m(n.outputTokens);return {inputTokens:c??0,
+outputTokens:g??0,complete:false}}let i=m(r.total),o=m(r.cacheRead),s=m(r.cacheWrite),d=m(a.total),p=m(a.reasoning);return {inputTokens:i??0,outputTokens:d??0,...o!==void 0?{cachedInputTokens:o}:{},...s!==
+void 0?{cacheWriteInputTokens:s}:{},...p!==void 0?{reasoningTokens:p}:{},...i!==void 0&&d!==void 0&&!e?{}:{complete:false}}}function Se(t){if(typeof t=="string")return t;if(f(t)&&typeof t.unified=="strin\
+g")return t.unified}function xe(t){switch(t){case void 0:return;case "tool-calls":return "tool_calls";case "content-filter":return "content_filter";default:return t}}function be(t){return f(t)&&Object.keys(
+D(t)).length>0}var R=class{texto="";toolCalls=[];reproducible=true;finish;error;anadir(e){switch(e.type){case "text-delta":{let n=typeof e.delta=="string"?e.delta:typeof e.textDelta=="string"?e.textDelta:
 "";return this.texto+=n,n===""?void 0:n}case "text":{let n=typeof e.text=="string"?e.text:"";return this.texto+=n,n===""?void 0:n}case "tool-call":e.providerExecuted===true?this.reproducible=false:this.toolCalls.
 push({...typeof e.toolCallId=="string"?{id:e.toolCallId}:{},name:String(e.toolName??""),arguments:v(e.input??e.args)});return;case "reasoning":case "reasoning-start":case "reasoning-delta":case "reasoning\
--end":xe(e.providerMetadata)&&(this.reproducible=false);return;case "text-start":case "text-end":case "tool-input-start":case "tool-input-delta":case "tool-input-end":case "stream-start":case "response-metadata":case "\
-raw":return;case "finish":this.finish=e;return;case "error":this.error={error:e.error};return;default:this.reproducible=false;return}}respuesta(e,n,r){let o=we(n),i=Se(o),a=this.reproducible&&o==="stop";return {
-text:this.texto,...this.toolCalls.length>0?{toolCalls:this.toolCalls}:{},...i?{finishReason:i}:{},usage:Re(e,r),...a?{}:{replayable:false}}}};function P(t,e){let n=new R;for(let r of Array.isArray(t.content)?
-t.content:[])p(r)&&n.anadir(r);return n.respuesta(t.usage,t.finishReason,e)}var T={inputTokens:{total:0,noCache:0,cacheRead:0,cacheWrite:0},outputTokens:{total:0,text:0,reasoning:0}};function I(t){let e=[];
-return t.text!==""&&e.push({type:"text",text:t.text}),(t.toolCalls??[]).forEach((n,r)=>{e.push({type:"tool-call",toolCallId:n.id??`bvsala-${t.requestId}-${r}`,toolName:n.name,input:n.arguments});}),e}function L(t){
-return {unified:(t.toolCalls?.length??0)>0?"tool-calls":"stop",raw:void 0}}function O(t,e,n){let r=[{type:"stream-start",warnings:Array.isArray(n.warnings)?n.warnings:[]}];return n.response&&r.push({type:"\
-response-metadata",...n.response}),t.forEach((o,i)=>{if(!p(o))return;let a=`bvsala-${i}`;if(o.type==="text"){let s=o.providerMetadata!==void 0?{providerMetadata:o.providerMetadata}:{};r.push({type:"te\
-xt-start",id:a,...s}),typeof o.text=="string"&&o.text!==""&&r.push({type:"text-delta",id:a,delta:o.text}),r.push({type:"text-end",id:a,...s});}else if(o.type==="reasoning"){let s=o.providerMetadata!==void 0?
-{providerMetadata:o.providerMetadata}:{};r.push({type:"reasoning-start",id:a,...s}),typeof o.text=="string"&&o.text!==""&&r.push({type:"reasoning-delta",id:a,delta:o.text}),r.push({type:"reasoning-end",
-id:a,...s});}else r.push(o);}),r.push({type:"finish",usage:e.usage,finishReason:e.finishReason,...e.providerMetadata!==void 0?{providerMetadata:e.providerMetadata}:{}}),r}function E(t){let e=0;return new ReadableStream(
-{pull(n){e<t.length?n.enqueue(t[e++]):n.close();}})}function Z(t){let e=String(t.modelId??"");if(e.includes("/"))return e;let n=String(t.provider??"").split(".")[0]??"";return n!==""?`${n}/${e}`:e}function Q(t){
-return typeof t.provider=="string"&&t.provider.toLowerCase().startsWith("mock")}function y(t,e){return {...p(t)?t:{},bivelio:{savingsReport:e}}}var De={router:false,compression:false,elision:false,serialization:false,toolDiscovery:false,modelCascade:"off",outputCodebook:false,deferrable:false,semanticCache:false,verification:false};
-function ke(t,e,n={}){let r=o=>t?.[o]!==false&&e?.[o]!==false;return {...De,exactCache:r("exactCache")&&n.sinReutilizar!==true,singleflight:r("singleflight")&&n.sinReutilizar!==true,deterministicResolvers:r("det\
-erministicResolvers")&&n.sinResolvers!==true}}function Ae(t,e){let n=t.readOnlyTools;if(typeof n=="function")try{return n(e)===!0}catch{return  false}return Array.isArray(n)?n.includes(e):false}function Y(t,e){
-let n=e.delProveedor!==true&&Ae(t,e.nombre);return {id:e.nombre,intents:[],risk:n?"read":"write",...e.parametros!==void 0&&e.parametros!==null&&typeof e.parametros=="object"?{parameters:e.parametros}:{},
+-end":be(e.providerMetadata)&&(this.reproducible=false);return;case "text-start":case "text-end":case "tool-input-start":case "tool-input-delta":case "tool-input-end":case "stream-start":case "response-metadata":case "\
+raw":return;case "finish":this.finish=e;return;case "error":this.error={error:e.error};return;default:this.reproducible=false;return}}respuesta(e,n,r){let a=Se(n),i=xe(a),o=this.reproducible&&a==="stop";return {
+text:this.texto,...this.toolCalls.length>0?{toolCalls:this.toolCalls}:{},...i?{finishReason:i}:{},usage:we(e,r),...o?{}:{replayable:false}}}};function P(t,e){let n=new R;for(let r of Array.isArray(t.content)?
+t.content:[])f(r)&&n.anadir(r);return n.respuesta(t.usage,t.finishReason,e)}var L={inputTokens:{total:0,noCache:0,cacheRead:0,cacheWrite:0},outputTokens:{total:0,text:0,reasoning:0}};function O(t){let e=[];
+return t.text!==""&&e.push({type:"text",text:t.text}),(t.toolCalls??[]).forEach((n,r)=>{e.push({type:"tool-call",toolCallId:n.id??`bvsala-${t.requestId}-${r}`,toolName:n.name,input:n.arguments});}),e}function I(t){
+return {unified:(t.toolCalls?.length??0)>0?"tool-calls":"stop",raw:void 0}}function T(t,e,n){let r=[{type:"stream-start",warnings:Array.isArray(n.warnings)?n.warnings:[]}];return n.response&&r.push({type:"\
+response-metadata",...n.response}),t.forEach((a,i)=>{if(!f(a))return;let o=`bvsala-${i}`;if(a.type==="text"){let s=a.providerMetadata!==void 0?{providerMetadata:a.providerMetadata}:{};r.push({type:"te\
+xt-start",id:o,...s}),typeof a.text=="string"&&a.text!==""&&r.push({type:"text-delta",id:o,delta:a.text}),r.push({type:"text-end",id:o,...s});}else if(a.type==="reasoning"){let s=a.providerMetadata!==void 0?
+{providerMetadata:a.providerMetadata}:{};r.push({type:"reasoning-start",id:o,...s}),typeof a.text=="string"&&a.text!==""&&r.push({type:"reasoning-delta",id:o,delta:a.text}),r.push({type:"reasoning-end",
+id:o,...s});}else r.push(a);}),r.push({type:"finish",usage:e.usage,finishReason:e.finishReason,...e.providerMetadata!==void 0?{providerMetadata:e.providerMetadata}:{}}),r}function E(t){let e=0;return new ReadableStream(
+{pull(n){e<t.length?n.enqueue(t[e++]):n.close();}})}function Q(t){let e=String(t.modelId??"");if(e.includes("/"))return e;let n=String(t.provider??"").split(".")[0]??"";return n!==""?`${n}/${e}`:e}function X(t){
+return typeof t.provider=="string"&&t.provider.toLowerCase().startsWith("mock")}function y(t,e){return {...f(t)?t:{},bivelio:{savingsReport:e}}}var De={router:false,compression:false,elision:false,serialization:false,toolDiscovery:false,modelCascade:"off",outputCodebook:false,deferrable:false,semanticCache:false,verification:false};
+function ke(t,e,n={}){let r=a=>t?.[a]!==false&&e?.[a]!==false;return {...De,exactCache:r("exactCache")&&n.sinReutilizar!==true,singleflight:r("singleflight")&&n.sinReutilizar!==true,deterministicResolvers:r("det\
+erministicResolvers")&&n.sinResolvers!==true}}function Me(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 ee(t,e){
+let n=e.delProveedor!==true&&Me(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 w(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 "$cycle";n.add(r);try{if(Array.isArray(r))return r.map(i=>{let a=w(i,e,n);return a===void 0?null:
-a});if(r instanceof Map)return {$map:[...r.entries()].map(i=>w(i,e,n))};if(r instanceof Set)return {$set:[...r.values()].map(i=>w(i,e,n))};let o={};for(let i of Object.keys(r).sort()){let a=w(r[i],e,n);
-a!==void 0&&(o[i]=a);}return o}finally{n.delete(r);}}async function X(t){let e=globalThis.crypto?.subtle;if(e){let r=new Uint8Array(await e.digest("SHA-256",t));return Array.from(r,o=>o.toString(16).padStart(
-2,"0")).join("")}let n="";for(let r=0;r<t.length;r+=32768)n+=String.fromCharCode(...t.subarray(r,r+32768));return sha256(n)}async function Me(t){try{let e=[],n=JSON.stringify(w(t,e,new WeakSet))??"null",r=await Promise.
-all(e.map(X)),o=new TextEncoder().encode(`${n}|${r.join(",")}`);return await X(o)}catch{return null}}async function ee(t,e){let n=await Me(e.proyeccion),r=`${e.context?.authorizationScope??"default"}|\
+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(i=>{let o=w(i,e,n);return o===void 0?null:
+o});if(r instanceof Map)return {$map:[...r.entries()].map(i=>w(i,e,n))};if(r instanceof Set)return {$set:[...r.values()].map(i=>w(i,e,n))};let a={};for(let i of Object.keys(r).sort()){let o=w(r[i],e,n);
+o!==void 0&&(a[i]=o);}return a}finally{n.delete(r);}}async function Y(t){let e=globalThis.crypto?.subtle;if(e){let r=new Uint8Array(await e.digest("SHA-256",t));return Array.from(r,a=>a.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 Ce(t){try{let e=[],n=JSON.stringify(w(t,e,new WeakSet))??"null",r=await Promise.
+all(e.map(Y)),a=new TextEncoder().encode(`${n}|${r.join(",")}`);return await Y(a)}catch{return null}}async function te(t,e){let n=await Ce(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:ke(t.savings,e.savings,{...e.restricciones,...n===null?{sinReutilizar:true}:{}})}}function te(t,e){if(!(!t&&!e))return {...t??{},...e??
-{}}}function ne(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 Ce="[bvsala] a streamed call finished but its savings report could not be produced; the model answer was delivered unchanged.";function He(t){let e=W("ai-sdk",t),n=t.reportInMetadata!==false;async function r(a,s){
-let d=ne(a.providerOptions?.bivelio),f;t.context&&(f=t.context({provider:s.provider,modelId:s.modelId,params:a}));let u=t.modelId?t.modelId({provider:s.provider,modelId:s.modelId}):Z(s),c=(Array.isArray(
-a.tools)?a.tools:[]).map(l=>Y(t,{nombre:String(l.name??l.id??""),descripcion:typeof l.description=="string"?l.description:void 0,parametros:l.inputSchema??l.parameters,delProveedor:l.type==="provider"||
-l.type==="provider-defined"}));return {req:await ee(t,{model:u,messages:V(a.prompt),tools:c,temperature:typeof a.temperature=="number"?a.temperature:void 0,seed:typeof a.seed=="number"?a.seed:void 0,context:te(
-f,d.context),savings:d.savings,proyeccion:{modelo:u,parametros:K(a)},restricciones:{sinResolvers:H(a)}}),simulado:Q(s)}}function o(a){return {model:a.model,messages:a.messages,...a.tools?{tools:a.tools}:
-{}}}return {specificationVersion:"v3",async wrapGenerate(a){let{req:s,simulado:d}=await r(a.params,a.model),f,u=C(o(s),async()=>(f=await x(u,()=>a.doGenerate()),P(f,d))),c=await e.generar(s,u);if(c.tipo===
-"sin-informe")return f;let g=e.informe(c.result.report);if(c.tipo==="vivo"){let l=f;return n?{...l,providerMetadata:y(l.providerMetadata,g)}:l}return Te(c.result,n?g:void 0,a.model)},async wrapStream(a){
-let{req:s,simulado:d}=await r(a.params,a.model),{abierta:f,abrir:u}=B(),c,g,l=Object.assign(C(o(s),async()=>(c=await x(l,()=>a.doGenerate()),P(c,d)),()=>{let M=new b;return x(l,()=>a.doStream()).then(
-S=>{g={...S,stream:Pe(S.stream,M,e,l,d,n)},u();},S=>M.fail(S)),M}),{abierta:f}),A=a.params.abortSignal instanceof AbortSignal?a.params.abortSignal:void 0,h=await e.transmitir(s,l,A);if(h.tipo==="abiert\
-o")return g;if(h.tipo==="sin-informe")return re(c,void 0);let q=e.informe(h.result.report);return h.tipo==="entero"?re(c,n?q:void 0):Ie(h.result,n?q:void 0,a.model)},close(){e.close();}}}function Pe(t,e,n,r,o,i){
-let a=t.getReader(),s=new R;return new ReadableStream({async pull(d){let f;try{f=await a.read();}catch(l){e.fail(l),d.error(l);return}if(f.done){e.fail(new Error("[bvsala] the model stream ended withou\
-t a finish part; nothing was measured.")),d.close();return}let u=f.value,c=s.anadir(u);if(u.type==="error"){e.fail(u.error??new Error("[bvsala] the model stream reported an error")),d.enqueue(u);return}
-if(c!==void 0&&e.push({type:"text",delta:c}),u.type!=="finish"||e.cerrada){d.enqueue(u);return}e.push({type:"done",response:s.respuesta(u.usage,u.finishReason,o)}),e.end();let g=u;try{let l=await r.resultado,
-A=n.informe(l.report);i&&(g={...u,providerMetadata:y(u.providerMetadata,A)});}catch{n.avisar("informe-del-stream",Ce);}d.enqueue(g);},cancel(d){return e.fail(new Error("[bvsala] the stream was cancelled \
-by its reader; nothing was measured.")),a.cancel(d)}})}function Te(t,e,n){return {content:I(t),finishReason:L(t),usage:T,...e?{providerMetadata:y(void 0,e)}:{},response:{id:`bvsala-${t.requestId}`,modelId:n.
-modelId,timestamp:new Date},warnings:[]}}function Ie(t,e,n){let r=O(I(t),{usage:T,finishReason:L(t),...e?{providerMetadata:y(void 0,e)}:{}},{warnings:[],response:{id:`bvsala-${t.requestId}`,modelId:n.
-modelId,timestamp:new Date}});return {stream:E(r)}}function re(t,e){let n=t.response,r=O(Array.isArray(t.content)?t.content:[],{usage:t.usage,finishReason:t.finishReason,...e?{providerMetadata:y(t.providerMetadata,
-e)}:t.providerMetadata!==void 0?{providerMetadata:t.providerMetadata}:{}},{warnings:t.warnings,response:n?Object.fromEntries(Object.entries(n).filter(([o])=>o!=="headers"&&o!=="body")):void 0});return {
-stream:E(r),...t.request!==void 0?{request:t.request}:{},...n?.headers!==void 0?{response:{headers:n.headers}}:{}}}function Ze(t){let n=t?.providerMetadata?.bivelio?.savingsReport;if(!(!n||typeof n!="\
-object"||typeof n.requestId!="string"))return n}export{Ze as getSavingsReport,He as savingsMiddleware};
+{seed:e.seed}:{},context:{...e.context??{},authorizationScope:r},savings:ke(t.savings,e.savings,{...e.restricciones,...n===null?{sinReutilizar:true}:{}})}}function ne(t,e){if(!(!t&&!e))return {...t??{},...e??
+{}}}function re(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 Pe="[bvsala] a streamed call finished but its savings report could not be produced; the model answer was delivered unchanged.";function Xe(t){let e=W("ai-sdk",t),n=t.reportInMetadata!==false;async function r(o,s){
+let d=re(o.providerOptions?.bivelio),p;t.context&&(p=t.context({provider:s.provider,modelId:s.modelId,params:o}));let u=t.modelId?t.modelId({provider:s.provider,modelId:s.modelId}):Q(s),c=(Array.isArray(
+o.tools)?o.tools:[]).map(l=>ee(t,{nombre:String(l.name??l.id??""),descripcion:typeof l.description=="string"?l.description:void 0,parametros:l.inputSchema??l.parameters,delProveedor:l.type==="provider"||
+l.type==="provider-defined"}));return {req:await te(t,{model:u,messages:H(o.prompt),tools:c,temperature:typeof o.temperature=="number"?o.temperature:void 0,seed:typeof o.seed=="number"?o.seed:void 0,context:ne(
+p,d.context),savings:d.savings,proyeccion:{modelo:u,parametros:K(o)},restricciones:{sinResolvers:Z(o)}}),simulado:X(s)}}function a(o){return {model:o.model,messages:o.messages,...o.tools?{tools:o.tools}:
+{}}}return {specificationVersion:"v3",async wrapGenerate(o){let{req:s,simulado:d}=await r(o.params,o.model),p,u=C(a(s),async()=>(p=await x(u,()=>o.doGenerate()),P(p,d))),c=await e.generar(s,u);if(c.tipo===
+"sin-informe")return p;let g=e.informe(c.result.report);if(c.tipo==="vivo"){let l=p;return n?{...l,providerMetadata:y(l.providerMetadata,g)}:l}return Oe(c.result,n?g:void 0,o.model)},async wrapStream(o){
+let{req:s,simulado:d}=await r(o.params,o.model),{abierta:p,abrir:u}=B(),c,g,l=Object.assign(C(a(s),async()=>(c=await x(l,()=>o.doGenerate()),P(c,d)),()=>{let M=new b;return x(l,()=>o.doStream()).then(
+S=>{g={...S,stream:Le(S.stream,M,e,l,d,n)},u();},S=>M.fail(S)),M}),{abierta:p}),k=o.params.abortSignal instanceof AbortSignal?o.params.abortSignal:void 0,h=await e.transmitir(s,l,k);if(h.tipo==="abiert\
+o")return g;if(h.tipo==="sin-informe")return oe(c,void 0);let q=e.informe(h.result.report);return h.tipo==="entero"?oe(c,n?q:void 0):Ie(h.result,n?q:void 0,o.model)},close(){e.close();}}}function Le(t,e,n,r,a,i){
+let o=t.getReader(),s=new R;return new ReadableStream({async pull(d){let p;try{p=await o.read();}catch(l){e.fail(l),d.error(l);return}if(p.done){e.fail(new Error("[bvsala] the model stream ended withou\
+t a finish part; nothing was measured.")),d.close();return}let u=p.value,c=s.anadir(u);if(u.type==="error"){e.fail(u.error??new Error("[bvsala] the model stream reported an error")),d.enqueue(u);return}
+if(c!==void 0&&e.push({type:"text",delta:c}),u.type!=="finish"||e.cerrada){d.enqueue(u);return}e.push({type:"done",response:s.respuesta(u.usage,u.finishReason,a)}),e.end();let g=u;try{let l=await r.resultado,
+k=n.informe(l.report);i&&(g={...u,providerMetadata:y(u.providerMetadata,k)});}catch{n.avisar("informe-del-stream",Pe);}d.enqueue(g);},cancel(d){return e.fail(new Error("[bvsala] the stream was cancelled \
+by its reader; nothing was measured.")),o.cancel(d)}})}function Oe(t,e,n){return {content:O(t),finishReason:I(t),usage:L,...e?{providerMetadata:y(void 0,e)}:{},response:{id:`bvsala-${t.requestId}`,modelId:n.
+modelId,timestamp:new Date},warnings:[]}}function Ie(t,e,n){let r=T(O(t),{usage:L,finishReason:I(t),...e?{providerMetadata:y(void 0,e)}:{}},{warnings:[],response:{id:`bvsala-${t.requestId}`,modelId:n.
+modelId,timestamp:new Date}});return {stream:E(r)}}function oe(t,e){let n=t.response,r=T(Array.isArray(t.content)?t.content:[],{usage:t.usage,finishReason:t.finishReason,...e?{providerMetadata:y(t.providerMetadata,
+e)}:t.providerMetadata!==void 0?{providerMetadata:t.providerMetadata}:{}},{warnings:t.warnings,response:n?Object.fromEntries(Object.entries(n).filter(([a])=>a!=="headers"&&a!=="body")):void 0});return {
+stream:E(r),...t.request!==void 0?{request:t.request}:{},...n?.headers!==void 0?{response:{headers:n.headers}}:{}}}function Ye(t){let n=t?.providerMetadata?.bivelio?.savingsReport;if(!(!n||typeof n!="\
+object"||typeof n.requestId!="string"))return n}export{Ye as getSavingsReport,Xe as savingsMiddleware};