@urun-sh/openai 0.2.53 → 0.2.55

package/dist/pi-extension/index.d.cts

@@ -1,6 +1,6 @@
 import { ExtensionAPI } from '@earendil-works/pi-coding-agent';
 import { Api, Model, Context, SimpleStreamOptions, AssistantMessageEventStream } from '@earendil-works/pi-ai';
-import { a as UrunSessionLike, U as UrunResponses } from '../ResponsesClient-CQ3l8Wft.cjs';
+import { a as UrunSessionLike, U as UrunResponses } from '../ResponsesClient-DrccxHTc.cjs';
 
 /**
  * Pi coding-agent extension: register the `urun` model provider, with one pi
@@ -40,22 +40,26 @@ import { a as UrunSessionLike, U as UrunResponses } from '../ResponsesClient-CQ3
  *      empty provider.
  *   2. `streamSimple` (pi's custom-API hook, keyed by `api: "urun-serve"`):
  *      opens ONE uRun session per app slug, POOLED and reused across turns,
- *      translates pi's Context into Responses `input` items, drives
- *      `UrunResponses.responses.create({stream: true})`, and maps
- *      `response.output_text.delta` events into pi `text_delta` events.
- *      Text-only, matching the precedent extension's fidelity (no tool-call
- *      streaming through the serve lane; pi's toolResult turns are folded
- *      into user-role input items — the Responses `function_call_output` item
- *      form needs call ids this lane never emits).
+ *      translates pi's Context into Responses `input` items (tool loop
+ *      included: assistant `toolCall` content → `function_call` items,
+ *      `toolResult` turns → `function_call_output` items), forwards
+ *      `context.tools` upstream, drives
+ *      `UrunResponses.responses.create({stream: true})`, and maps the lane's
+ *      decoded events onto pi-ai's REAL AssistantMessageEvent union:
+ *      output_text → text_*, reasoning_text → thinking_*, function-call
+ *      argument fragments → toolcall_* (stopReason "toolUse" ends a
+ *      tool-calling turn, which is what makes pi's agent loop execute the
+ *      tools and feed results back).
  *   3. ABORT/CANCEL (loud contract): the Responses lane has NO request-cancel
  *      primitive (`SdkTransport.sendResponses` writes an envelope and the
  *      decode loop just reads the reply lane), so Ctrl-C / the stall watchdog
- *      stop consuming AND evict+close the pooled session — tearing down the
- *      WebRTC transport is what stops server-side generation. A clean turn
- *      keeps its session for the next turn; a failed/aborted/stalled turn
- *      always reopens.
- *   4. cleanup: pooled sessions are released (`session.close()`) on pi's
- *      `session_shutdown` lifecycle event.
+ *      stop consuming AND evict+END the pooled session (core's own
+ *      `Session.end()`) — tearing down the WebRTC transport is what stops
+ *      server-side generation. A clean turn keeps its session for the next
+ *      turn; a failed/aborted/stalled turn always reopens — and a turn whose
+ *      session DIED re-homes once and replays (see REHOME_ATTEMPTS).
+ *   4. cleanup: pooled sessions are released (core's `Session.end()`) on
+ *      pi's `session_shutdown` lifecycle event.
  *
  * RUNTIME DEPENDENCY STORY (owner requirement — stated honestly): token
  * streams are named-DATA session streams, i.e. SCTP data channels over
@@ -115,7 +119,24 @@ declare function resolveSessionEnv(env: NodeJS.ProcessEnv): {
  * `@urun-sh/core`.
  */
 interface SessionFactory {
-    (env: NodeJS.ProcessEnv, appSlug: string, fnName: string): Promise<UrunSessionLike> | UrunSessionLike;
+    (env: NodeJS.ProcessEnv, appSlug: string, fnName: string): Promise<OwnedSession> | OwnedSession;
+}
+/**
+ * A session this lane OWNS: core's transport surface plus core's own terminal
+ * release, `Session.end()` (the same requirement the compat proxy asserts in
+ * `sessionOf`). A session without it could never be released — it would leak
+ * the backend session and its capacity lease until the platform's abandonment
+ * backstop reclaimed it — so the factory refuses to hand one back.
+ */
+type OwnedSession = UrunSessionLike & {
+    end: () => Promise<unknown>;
+};
+/** The structured diagnostic shape core's transports emit (typed locally). */
+interface CoreDiagnosticLike {
+    level: string;
+    kind: string;
+    message: string;
+    detail?: Record<string, unknown>;
 }
 /**
  * The subset of `@urun-sh/core` a session factory needs — typed locally so
@@ -128,6 +149,7 @@ interface CoreModuleLike {
         orgId: string;
         jwt?: string;
         getAccessToken?: () => Promise<string>;
+        diagnosticSink?: (diagnostic: CoreDiagnosticLike) => void;
     }) => Record<string, (args?: Record<string, unknown>) => UrunSessionLike>;
     createClientToken: (apiKey: string, opts: {
         baseUrl?: string;
@@ -149,10 +171,20 @@ interface CoreModuleLike {
  *     self-contained bundle INLINES core (and, through core's own dynamic
  *     werift import, the WebRTC backend) — that lane has no node_modules.
  */
+/**
+ * A diagnostic sink appending formatted lines to `path` — SYNCHRONOUS writes
+ * (a crashing pi must still leave the lines on disk), lazy directory
+ * creation, one open descriptor per sink. An open/write failure THROWS: core's
+ * `_emitDiagnostic` catches it, complains once on the console, and keeps the
+ * structured event flowing — loud once, never TUI spam, never silent.
+ */
+declare function createFileDiagnosticSink(path: string): (diagnostic: CoreDiagnosticLike) => void;
+/** Where the pi lane's quiet diagnostics land (the compat log convention). */
+declare function piDiagnosticLogPath(home?: string): string;
 declare function makeSessionFactory(loadCore: () => Promise<CoreModuleLike> | CoreModuleLike): SessionFactory;
 /** One pooled backhaul: the session plus its Responses client (cli.ts shape). */
 interface PoolEntry {
-    session: UrunSessionLike;
+    session: OwnedSession;
     responses: UrunResponses;
 }
 /**
@@ -174,22 +206,36 @@ declare class SessionPool {
     /** Release everything (pi session_shutdown). */
     closeAll(): void;
 }
-/** A Responses input item (message form; content stays a plain string). */
-interface ResponsesInputItem {
+/** A Responses input item: role message or function-call round-trip item. */
+type ResponsesInputItem = {
     role: 'system' | 'user' | 'assistant';
     content: string;
-}
+} | {
+    type: 'function_call';
+    call_id: string;
+    name: string;
+    arguments: string;
+} | {
+    type: 'function_call_output';
+    call_id: string;
+    output: string;
+};
 /**
- * Pi `Context` -> Responses `input` items. Pi models tool results as
- * `role:"toolResult"`; the Responses item form for those
- * (`{type:'function_call_output', call_id, …}`) needs call ids this text-only
- * lane never emits, so tool results are folded into `user`-role items, then
- * any consecutive same-role runs are coalesced into one item so a chat
- * template that assumes strict role alternation never sees two turns of the
- * same role back-to-back (same rationale as the precedent extension).
+ * Pi `Context` -> Responses `input` items, in ENCOUNTER ORDER (the tool
+ * loop's canonical Responses shapes — the same translation the compat proxy
+ * performs):
+ *   - assistant `toolCall` content → `{type:'function_call', call_id, name,
+ *     arguments}` (arguments re-serialized: pi stores the PARSED object);
+ *   - `role:"toolResult"` turns    → `{type:'function_call_output', call_id,
+ *     output}` keyed by pi's `toolCallId`;
+ *   - assistant `thinking` content is the producing model's internal state —
+ *     not re-submittable — and is skipped by design (same rule as the proxy);
+ *   - text runs become role messages where they sit; adjacent same-role
+ *     messages are coalesced so chat templates that assume strict role
+ *     alternation never see two turns of one role back-to-back.
  */
 declare function toResponsesInput(context: Context): ResponsesInputItem[];
-/** Fold consecutive same-role items into one, joining with a blank line. */
+/** Fold consecutive same-role message items into one, joining with a blank line. */
 declare function coalesceSameRole(items: ResponsesInputItem[]): ResponsesInputItem[];
 /** Tunable timeouts (overridable so tests can drive them deterministically). */
 interface StreamTiming {
@@ -219,4 +265,4 @@ declare function createUrunExtension(opts?: UrunExtensionOptions): (pi: Extensio
 /** The pi extension factory (default export — what pi's loader invokes). */
 declare const _default: (pi: ExtensionAPI) => Promise<void>;
 
-export { type CoreModuleLike, type ResponsesInputItem, type SessionFactory, SessionPool, type StreamTiming, URUN_API, type UrunExtensionOptions, coalesceSameRole, createUrunExtension, _default as default, makeSessionFactory, makeStreamSimple, resolveSessionEnv, toResponsesInput };
+export { type CoreDiagnosticLike, type CoreModuleLike, type OwnedSession, type ResponsesInputItem, type SessionFactory, SessionPool, type StreamTiming, URUN_API, type UrunExtensionOptions, coalesceSameRole, createFileDiagnosticSink, createUrunExtension, _default as default, makeSessionFactory, makeStreamSimple, piDiagnosticLogPath, resolveSessionEnv, toResponsesInput };

package/dist/pi-extension/index.d.ts

@@ -1,6 +1,6 @@
 import { ExtensionAPI } from '@earendil-works/pi-coding-agent';
 import { Api, Model, Context, SimpleStreamOptions, AssistantMessageEventStream } from '@earendil-works/pi-ai';
-import { a as UrunSessionLike, U as UrunResponses } from '../ResponsesClient-CQ3l8Wft.js';
+import { a as UrunSessionLike, U as UrunResponses } from '../ResponsesClient-DrccxHTc.js';
 
 declare const URUN_API: Api;
 
@@ -18,7 +18,18 @@ declare function resolveSessionEnv(env: NodeJS.ProcessEnv): {
 };
 
 interface SessionFactory {
-    (env: NodeJS.ProcessEnv, appSlug: string, fnName: string): Promise<UrunSessionLike> | UrunSessionLike;
+    (env: NodeJS.ProcessEnv, appSlug: string, fnName: string): Promise<OwnedSession> | OwnedSession;
+}
+
+type OwnedSession = UrunSessionLike & {
+    end: () => Promise<unknown>;
+};
+
+interface CoreDiagnosticLike {
+    level: string;
+    kind: string;
+    message: string;
+    detail?: Record<string, unknown>;
 }
 
 interface CoreModuleLike {
@@ -27,6 +38,7 @@ interface CoreModuleLike {
         orgId: string;
         jwt?: string;
         getAccessToken?: () => Promise<string>;
+        diagnosticSink?: (diagnostic: CoreDiagnosticLike) => void;
     }) => Record<string, (args?: Record<string, unknown>) => UrunSessionLike>;
     createClientToken: (apiKey: string, opts: {
         baseUrl?: string;
@@ -37,10 +49,13 @@ interface CoreModuleLike {
     }>;
 }
 
+declare function createFileDiagnosticSink(path: string): (diagnostic: CoreDiagnosticLike) => void;
+
+declare function piDiagnosticLogPath(home?: string): string;
 declare function makeSessionFactory(loadCore: () => Promise<CoreModuleLike> | CoreModuleLike): SessionFactory;
 
 interface PoolEntry {
-    session: UrunSessionLike;
+    session: OwnedSession;
     responses: UrunResponses;
 }
 
@@ -57,10 +72,19 @@ declare class SessionPool {
     closeAll(): void;
 }
 
-interface ResponsesInputItem {
+type ResponsesInputItem = {
     role: 'system' | 'user' | 'assistant';
     content: string;
-}
+} | {
+    type: 'function_call';
+    call_id: string;
+    name: string;
+    arguments: string;
+} | {
+    type: 'function_call_output';
+    call_id: string;
+    output: string;
+};
 
 declare function toResponsesInput(context: Context): ResponsesInputItem[];
 
@@ -85,4 +109,4 @@ declare function createUrunExtension(opts?: UrunExtensionOptions): (pi: Extensio
 
 declare const _default: (pi: ExtensionAPI) => Promise<void>;
 
-export { type CoreModuleLike, type ResponsesInputItem, type SessionFactory, SessionPool, type StreamTiming, URUN_API, type UrunExtensionOptions, coalesceSameRole, createUrunExtension, _default as default, makeSessionFactory, makeStreamSimple, resolveSessionEnv, toResponsesInput };
+export { type CoreDiagnosticLike, type CoreModuleLike, type OwnedSession, type ResponsesInputItem, type SessionFactory, SessionPool, type StreamTiming, URUN_API, type UrunExtensionOptions, coalesceSameRole, createFileDiagnosticSink, createUrunExtension, _default as default, makeSessionFactory, makeStreamSimple, piDiagnosticLogPath, resolveSessionEnv, toResponsesInput };

package/dist/pi-extension/index.js

@@ -1,3 +1,4 @@
-import{c as C}from"../chunk-FSS2A7IU.js";import{a as A,c as D}from"../chunk-PS4BLX7J.js";import{createAssistantMessageEventStream as J}from"@earendil-works/pi-ai";async function F(t){let e=await D(t),n=e.filter(o=>o.function_name===t.fnName&&o.deployment_status==="active").map(o=>o.app_slug);if(n.length===0)throw new Error(`urun pi extension: the org has no active deployed app exposing "${t.fnName}" (GET ${t.apiUrl}/apps returned ${e.length} app(s), none servable) \u2014 deploy one with \`urun serve <model>\` first`);return n.sort()}var O="urun-serve",G="serve",H=131072,Y=16384,$=3,B=42e4,X=6e4,z=15e3;function Q(t){let e=(t.URUN_API_KEY??"").trim();if(!e)throw new Error("urun pi extension: URUN_API_KEY is required \u2014 model discovery lists the org's deployed serve apps via the org API (URUN_JWT alone cannot list apps)");return{apiKey:e,apiUrl:(t.URUN_API_URL??"").trim()||A,fnName:(t.URUN_FUNCTION??"").trim()||G}}function V(t){let e=(t.URUN_BASE_URL??"").trim(),n=(t.URUN_ORG_ID??"").trim();if(!e)throw new Error("urun pi extension: URUN_BASE_URL is required to open a session");if(/\/v1\/*$/.test(new URL(e).pathname))throw new Error(`urun pi extension: URUN_BASE_URL must be the session-gateway base (e.g. https://api.urun.sh), got ${e} \u2014 a trailing /v1 is the org API form (URUN_API_URL); session allocation 404s against it. Drop the /v1.`);if(!n)throw new Error("urun pi extension: URUN_ORG_ID is required to open a session");let o=(t.URUN_JWT??"").trim();if(o)return{baseUrl:e,orgId:n,auth:{lane:"jwt",jwt:o}};let i=(t.URUN_API_KEY??"").trim();if(!i)throw new Error("urun pi extension: URUN_JWT or URUN_API_KEY is required to open a session");let a=(t.URUN_GATEWAY_URL??"").trim()||void 0;return{baseUrl:e,orgId:n,auth:{lane:"api-key",apiKey:i,gatewayUrl:a}}}function Z(t){return async(e,n,o)=>{let{baseUrl:i,orgId:a,auth:s}=V(e),{App:u,createClientToken:l}=await t(),r=(s.lane==="jwt"?u(n,{baseUrl:i,orgId:a,jwt:s.jwt}):u(n,{baseUrl:i,orgId:a,getAccessToken:async()=>(await l(s.apiKey,{baseUrl:s.gatewayUrl,expiresIn:300,allowedFunctions:[`${n}/${o}`]})).token}))[o];if(typeof r!="function")throw new Error(`urun pi extension: app "${n}" has no function "${o}"`);let d=r(),p=d.connect;return typeof p=="function"&&await p.call(d),d}}var ee=Z(async()=>{let{createRequire:t}=await import("module"),e=t(import.meta.url);if(typeof globalThis.RTCPeerConnection>"u")try{e.resolve("werift")}catch{throw new Error('urun pi extension: token streams require a WebRTC data channel, and this Node runtime has no RTCPeerConnection and no "werift" backend installed (@urun-sh/core\'s optional Node WebRTC backend \u2014 it was probably stripped by a --no-optional install). Install werift (`npm i werift`) and retry.')}return e("@urun-sh/core")});function te(t){try{Promise.resolve(t.close?.()).catch(()=>{})}catch{}}var x=class{constructor(e,n,o){this.open=e;this.env=n;this.fnName=o}open;env;fnName;pool=new Map;acquire(e){let n=this.pool.get(e);if(!n){let o=Promise.resolve(this.open(this.env,e,this.fnName)).then(i=>({session:i,responses:new C(i)}));n=o,this.pool.set(e,o),o.catch(()=>{this.pool.get(e)===o&&this.pool.delete(e)})}return n}evict(e){let n=this.pool.get(e);n&&(this.pool.delete(e),n.then(o=>te(o.session),()=>{}))}closeAll(){for(let e of[...this.pool.keys()])this.evict(e)}};function ne(t){let e=[];t.systemPrompt&&e.push({role:"system",content:t.systemPrompt});for(let n of t.messages){let o=n.role==="assistant"?"assistant":"user";e.push({role:o,content:re(n.content)})}return oe(e)}function oe(t){let e=[];for(let n of t){let o=e[e.length-1];o&&o.role===n.role?o.content=[o.content,n.content].filter(i=>i.length>0).join(`
+import{c as G}from"../chunk-HQVV5UN5.js";import{a as F,c as H}from"../chunk-5CEMLB6H.js";import{createAssistantMessageEventStream as oe}from"@earendil-works/pi-ai";import{mkdirSync as re,openSync as se,writeSync as ie}from"fs";import ae from"os";import{dirname as le,join as ce}from"path";async function Y(t){let e=await H(t),n=e.filter(o=>o.function_name===t.fnName&&o.deployment_status==="active").map(o=>o.app_slug);if(n.length===0)throw new Error(`urun pi extension: the org has no active deployed app exposing "${t.fnName}" (GET ${t.apiUrl}/apps returned ${e.length} app(s), none servable) \u2014 deploy one with \`urun serve <model>\` first`);return n.sort()}var V="urun-serve",ue="serve",pe=131072,de=16384,B=3,fe=42e4,ge=6e4,X=2,me=15e3;function ye(t){let e=(t.URUN_API_KEY??"").trim();if(!e)throw new Error("urun pi extension: URUN_API_KEY is required \u2014 model discovery lists the org's deployed serve apps via the org API (URUN_JWT alone cannot list apps)");return{apiKey:e,apiUrl:(t.URUN_API_URL??"").trim()||F,fnName:(t.URUN_FUNCTION??"").trim()||ue}}function he(t){let e=(t.URUN_BASE_URL??"").trim(),n=(t.URUN_ORG_ID??"").trim();if(!e)throw new Error("urun pi extension: URUN_BASE_URL is required to open a session");if(/\/v1\/*$/.test(new URL(e).pathname))throw new Error(`urun pi extension: URUN_BASE_URL must be the session-gateway base (e.g. https://api.urun.sh), got ${e} \u2014 a trailing /v1 is the org API form (URUN_API_URL); session allocation 404s against it. Drop the /v1.`);if(!n)throw new Error("urun pi extension: URUN_ORG_ID is required to open a session");let o=(t.URUN_JWT??"").trim();if(o)return{baseUrl:e,orgId:n,auth:{lane:"jwt",jwt:o}};let i=(t.URUN_API_KEY??"").trim();if(!i)throw new Error("urun pi extension: URUN_JWT or URUN_API_KEY is required to open a session");let u=(t.URUN_GATEWAY_URL??"").trim()||void 0;return{baseUrl:e,orgId:n,auth:{lane:"api-key",apiKey:i,gatewayUrl:u}}}function we(t){let e=null;return n=>{e===null&&(re(le(t),{recursive:!0}),e=se(t,"a"));let o=n.detail?` ${JSON.stringify(n.detail)}`:"";ie(e,`${new Date().toISOString()} [${n.level}] [urun] ${n.message}${o}
+`)}}function _e(t=ae.homedir()){return ce(t,".urun","logs","pi-extension.log")}function xe(t){let e=we(_e());return async(n,o,i)=>{let{baseUrl:u,orgId:s,auth:d}=he(n),{App:x,createClientToken:a}=await t(),w=(d.lane==="jwt"?x(o,{baseUrl:u,orgId:s,jwt:d.jwt,diagnosticSink:e}):x(o,{baseUrl:u,orgId:s,diagnosticSink:e,getAccessToken:async()=>(await a(d.apiKey,{baseUrl:d.gatewayUrl,expiresIn:300,allowedFunctions:[`${o}/${i}`]})).token}))[i];if(typeof w!="function")throw new Error(`urun pi extension: app "${o}" has no function "${i}"`);let m=w();if(typeof m.end!="function")throw new Error(`urun pi extension: app "${o}" function "${i}" returned a session without end()`);let E=m.connect;return typeof E=="function"&&await E.call(m),m}}var ve=xe(async()=>{let{createRequire:t}=await import("module"),e=t(import.meta.url);if(typeof globalThis.RTCPeerConnection>"u")try{e.resolve("werift")}catch{throw new Error('urun pi extension: token streams require a WebRTC data channel, and this Node runtime has no RTCPeerConnection and no "werift" backend installed (@urun-sh/core\'s optional Node WebRTC backend \u2014 it was probably stripped by a --no-optional install). Install werift (`npm i werift`) and retry.')}return e("@urun-sh/core")});function ke(t){try{Promise.resolve(t.end()).catch(()=>{})}catch{}}var K=class{constructor(e,n,o){this.open=e;this.env=n;this.fnName=o}open;env;fnName;pool=new Map;acquire(e){let n=this.pool.get(e);if(!n){let o=Promise.resolve(this.open(this.env,e,this.fnName)).then(i=>({session:i,responses:new G(i)}));n=o,this.pool.set(e,o),o.catch(()=>{this.pool.get(e)===o&&this.pool.delete(e)})}return n}evict(e){let n=this.pool.get(e);n&&(this.pool.delete(e),n.then(o=>ke(o.session),()=>{}))}closeAll(){for(let e of[...this.pool.keys()])this.evict(e)}};function Re(t){let e=[];t.systemPrompt&&e.push({role:"system",content:t.systemPrompt});for(let n of t.messages){if(n.role==="toolResult"){e.push({type:"function_call_output",call_id:n.toolCallId,output:Ae(n.content)});continue}if(n.role==="assistant"&&Array.isArray(n.content)){let i="",u=()=>{i&&(e.push({role:"assistant",content:i}),i="")};for(let s of n.content)if(s.type==="text")i+=String(s.text??"");else if(s.type==="toolCall")u(),e.push({type:"function_call",call_id:String(s.id??""),name:String(s.name??""),arguments:JSON.stringify(s.arguments??{})});else{if(s.type==="thinking")continue;throw new Error(`urun pi extension: unsupported assistant content part "${String(s.type)}"`)}u();continue}let o=n.role==="assistant"?"assistant":"user";e.push({role:o,content:be(n.content)})}return Te(e)}function Te(t){let e=[];for(let n of t){let o=e[e.length-1];o&&"role"in o&&"role"in n&&o.role===n.role?o.content=[o.content,n.content].filter(i=>i.length>0).join(`
 
-`):e.push({...n})}return e}function re(t){return typeof t=="string"?t:Array.isArray(t)?t.map(e=>e&&typeof e=="object"&&"text"in e?String(e.text):"").join(""):""}function se(t){let e=t instanceof Error?t.message:String(t),n=t?.code;return/\b529\b/.test(e)||/overloaded/i.test(e)||n===529||n==="529"}function ie(t){return new Promise(e=>setTimeout(e,t))}function ae(t,e,n,o){return new Promise((i,a)=>{let s=!1,u=r=>{s||(s=!0,clearTimeout(l),o?.removeEventListener("abort",c),r())},l=setTimeout(()=>u(()=>a(n())),e),c=()=>u(()=>a(new Error("aborted")));if(o?.aborted){c();return}o?.addEventListener("abort",c,{once:!0}),Promise.resolve(t).then(r=>u(()=>i(r)),r=>u(()=>a(r)))})}async function*ce(t,e){let n=t[Symbol.asyncIterator](),o,i=new Promise((a,s)=>{o=()=>s(new Error("aborted")),e.addEventListener("abort",o,{once:!0})});try{for(;;){if(e.aborted)throw new Error("aborted");let a=await Promise.race([n.next(),i]);if(a.done)return;yield a.value}}finally{o&&e.removeEventListener("abort",o),n.return?.(void 0)}}function ue(t,e){let n=t?.sendMessage;if(typeof n=="function")try{n.call(t,{customType:"urun-status",content:e,display:!0},{triggerTurn:!1})}catch{}}function pe(t,e={},n){let o=e.connectDeadlineMs??B,i=e.stallTimeoutMs??X,a=e.phaseHeartbeatMs??z;return(s,u,l)=>{let c=J(),r={role:"assistant",content:[{type:"text",text:""}],api:s.api,provider:s.provider,model:s.id,usage:{input:0,output:0,cacheRead:0,cacheWrite:0,totalTokens:0,cost:{input:0,output:0,cacheRead:0,cacheWrite:0,total:0}},stopReason:"stop",timestamp:Date.now()},d=r.content[0],p=new AbortController,m=l?.signal,P=!1,b=()=>{P=!0,p.abort()};m&&(m.aborted?b():m.addEventListener("abort",b,{once:!0}));let E=!1,y,I=()=>{y&&clearTimeout(y),y=setTimeout(()=>{E=!0,p.abort()},i)},_=()=>{y&&clearTimeout(y),y=void 0};return(async()=>{c.push({type:"start",partial:r}),c.push({type:"text_start",contentIndex:0,partial:r});let k=!1,v,S=()=>{v&&clearInterval(v),v=void 0},K=Date.now();v=setInterval(()=>{let f=Math.round((Date.now()-K)/1e3);ue(n,`uRun: still opening the "${s.id}" session\u2026 (${f}s \u2014 serve apps scale to zero, first turn can cold-start ~6min)`)},a);try{let{responses:f}=await ae(t.acquire(s.id),o,()=>new Error(`uRun session for "${s.id}" did not connect within ${Math.round(o/1e3)}s \u2014 likely cold-starting (scaled to zero) or queued behind capacity. Retry shortly.`),p.signal);S();let T=ne(u),N,U;for(let R=1;R<=$;R++)try{let h=await f.responses.create({model:s.id,input:T,stream:!0,max_output_tokens:s.maxTokens});I();for await(let W of ce(h,p.signal)){I();let g=W;if(g.type==="response.output_text.delta"){let w=typeof g.delta=="string"?g.delta:"";w&&(d.text+=w,c.push({type:"text_delta",contentIndex:0,delta:w,partial:r}))}else if(g.type==="response.completed"){N=g.response;break}else if(g.type==="error"){let w=g.error,L=new Error(`uRun serve error: ${w?.message??"unknown error"}`);throw L.code=w?.code??null,L}}U=void 0;break}catch(h){if(U=h,p.signal.aborted)throw h;if(R<$&&se(h)&&d.text.length===0){await ie(R*500);continue}throw h}if(U)throw U;_(),c.push({type:"text_end",contentIndex:0,content:d.text,partial:r});let M=N?.status==="incomplete"?"length":"stop";r.stopReason=M,c.push({type:"done",reason:M,message:r}),c.end(r)}catch(f){if(k=!0,_(),P||m?.aborted&&!E)r.stopReason="aborted",r.errorMessage="aborted by user",c.push({type:"error",reason:"aborted",error:r});else{let T=E?`uRun stream stalled \u2014 no output for ${Math.round(i/1e3)}s; the model may be hung. Retry, or Ctrl-C to abort.`:f instanceof Error?f.message:String(f);r.stopReason="error",r.errorMessage=T,c.push({type:"error",reason:"error",error:r})}c.end(r)}finally{_(),S(),m&&m.removeEventListener("abort",b),k&&t.evict(s.id)}})(),c}}function le(t){return{id:t,name:`uRun ${t}`,api:O,reasoning:!1,input:["text"],cost:{input:0,output:0,cacheRead:0,cacheWrite:0},contextWindow:H,maxTokens:Y}}function de(t={}){return async e=>{let n=t.env??process.env,{apiKey:o,apiUrl:i,fnName:a}=Q(n),s=await F({apiUrl:i,apiKey:o,fnName:a,fetchImpl:t.fetchImpl}),u=new x(t.openSession??ee,n,a);e.on("session_shutdown",()=>u.closeAll());let l={name:"uRun",api:O,baseUrl:"urun://session",apiKey:"$URUN_API_KEY",models:s.map(le),streamSimple:pe(u,t.timing??{},e)};e.registerProvider("urun",l)}}var Ue=de();export{x as SessionPool,O as URUN_API,oe as coalesceSameRole,de as createUrunExtension,Ue as default,Z as makeSessionFactory,pe as makeStreamSimple,V as resolveSessionEnv,ne as toResponsesInput};
+`):e.push({...n})}return e}function be(t){return typeof t=="string"?t:Array.isArray(t)?t.map(e=>e&&typeof e=="object"&&"text"in e?String(e.text):"").join(""):""}function Ae(t){if(typeof t=="string")return t;if(!Array.isArray(t))return"";let e="";for(let n of t){let o=String(n.type??"");if(o!=="text")throw new Error(`urun pi extension: unsupported toolResult content part "${o}" \u2014 the serve lane carries text tool results only`);e+=String(n.text??"")}return e}function Ee(t){if(!(!Array.isArray(t)||t.length===0))return t.map(e=>({type:"function",name:e.name,description:e.description,parameters:e.parameters}))}function z(t,e){if(!e.trim())return{};try{return JSON.parse(e)}catch{throw new Error(`uRun serve error: tool call "${t}" arguments are not valid JSON: ${e}`)}}function Ue(t){let e=t?.output;return Array.isArray(e)?e.filter(n=>n.type==="function_call"):[]}function Se(t){let e=t?.output;if(!Array.isArray(e))return"";let n="";for(let o of e)if(!(o.type!=="reasoning"||!Array.isArray(o.content)))for(let i of o.content)i.type==="reasoning_text"&&(n+=String(i.text??""));return n}var Ie=["<tool_call>","<function="];function Pe(t,e){let n=Ie.find(o=>e.includes(o));if(n)return new Error(`serving row emitted a prose tool call (literal ${JSON.stringify(n)} in the assistant text) for a request that carried tools, and no structured tool_calls arrived \u2014 the tool-call parser is not configured on the model row "${t}" (see catalog parser_defaults)`)}var L=class extends Error{};function Ce(t){let e=t instanceof Error?t.message:String(t),n=t?.code;return/\b529\b/.test(e)||/overloaded/i.test(e)||n===529||n==="529"}function Ne(t){return new Promise(e=>setTimeout(e,t))}function Me(t,e,n,o){return new Promise((i,u)=>{let s=!1,d=r=>{s||(s=!0,clearTimeout(x),o?.removeEventListener("abort",a),r())},x=setTimeout(()=>d(()=>u(n())),e),a=()=>d(()=>u(new Error("aborted")));if(o?.aborted){a();return}o?.addEventListener("abort",a,{once:!0}),Promise.resolve(t).then(r=>d(()=>i(r)),r=>d(()=>u(r)))})}async function*Oe(t,e){let n=t[Symbol.asyncIterator](),o,i=new Promise((u,s)=>{o=()=>s(new Error("aborted")),e.addEventListener("abort",o,{once:!0})});try{for(;;){if(e.aborted)throw new Error("aborted");let u=await Promise.race([n.next(),i]);if(u.done)return;yield u.value}}finally{o&&e.removeEventListener("abort",o),n.return?.(void 0)}}function Q(t,e){let n=t?.sendMessage;if(typeof n=="function")try{n.call(t,{customType:"urun-status",content:e,display:!0},{triggerTurn:!1})}catch{}}function Le(t,e={},n){let o=e.connectDeadlineMs??fe,i=e.stallTimeoutMs??ge,u=e.phaseHeartbeatMs??me;return(s,d,x)=>{let a=oe(),r={role:"assistant",content:[],api:s.api,provider:s.provider,model:s.id,usage:{input:0,output:0,cacheRead:0,cacheWrite:0,totalTokens:0,cost:{input:0,output:0,cacheRead:0,cacheWrite:0,total:0}},stopReason:"stop",timestamp:Date.now()},w=new AbortController,m=x?.signal,E=!1,M=()=>{E=!0,w.abort()};m&&(m.aborted?M():m.addEventListener("abort",M,{once:!0}));let P=!1,U,J=()=>{U&&clearTimeout(U),U=setTimeout(()=>{P=!0,w.abort()},i)},O=()=>{U&&clearTimeout(U),U=void 0},v=null,k=null,y=null,W="",C=!1,$=()=>{if(!k)return;let p=k;k=null,a.push({type:"thinking_end",contentIndex:r.content.indexOf(p),content:p.thinking,partial:r})},q=()=>{if(!v)return;let p=v;v=null,a.push({type:"text_end",contentIndex:r.content.indexOf(p),content:p.text,partial:r})},D=()=>{if(!y)return;let{call:p,args:h,contentIndex:R}=y;y=null,p.arguments=z(p.name,h),a.push({type:"toolcall_end",contentIndex:R,toolCall:p,partial:r})},Z=p=>{let h={type:"toolCall",id:String(p.call_id??p.id??`call_${r.content.length}`),name:String(p.name??""),arguments:{}};r.content.push(h);let R=r.content.length-1;a.push({type:"toolcall_start",contentIndex:R,partial:r});let N=typeof p.arguments=="string"?p.arguments:"";N&&a.push({type:"toolcall_delta",contentIndex:R,delta:N,partial:r}),h.arguments=z(h.name,N),a.push({type:"toolcall_end",contentIndex:R,toolCall:h,partial:r})};return(async()=>{a.push({type:"start",partial:r});let p=!1,h,R=()=>{h&&clearInterval(h),h=void 0},N=Date.now();h=setInterval(()=>{let _=Math.round((Date.now()-N)/1e3);Q(n,`uRun: still opening the "${s.id}" session\u2026 (${_}s \u2014 serve apps scale to zero, first turn can cold-start ~6min)`)},u);let ee=async _=>{let{responses:T}=await Me(t.acquire(s.id),o,()=>new Error(`uRun session for "${s.id}" did not connect within ${Math.round(o/1e3)}s \u2014 likely cold-starting (scaled to zero) or queued behind capacity. Retry shortly.`),w.signal);R();let S,b;for(let l=1;l<=B;l++)try{let g=await T.responses.create({model:s.id,input:_,stream:!0,tools:Ee(d.tools),max_output_tokens:s.maxTokens});J();for await(let A of Oe(g,w.signal)){J();let f=A;if(f.type==="response.output_text.delta"){let c=typeof f.delta=="string"?f.delta:"";c&&(C=!0,$(),D(),v||(v={type:"text",text:""},r.content.push(v),a.push({type:"text_start",contentIndex:r.content.length-1,partial:r})),v.text+=c,W+=c,a.push({type:"text_delta",contentIndex:r.content.indexOf(v),delta:c,partial:r}))}else if(f.type==="response.reasoning_text.delta"){let c=typeof f.delta=="string"?f.delta:"";c&&(C=!0,k||(k={type:"thinking",thinking:""},r.content.push(k),a.push({type:"thinking_start",contentIndex:r.content.length-1,partial:r})),k.thinking+=c,a.push({type:"thinking_delta",contentIndex:r.content.indexOf(k),delta:c,partial:r}))}else if(f.type==="response.function_call_arguments.delta"&&typeof f.tool_index=="number"){let c=f;if(C=!0,$(),q(),!y||y.toolIndex!==c.tool_index){D();let j={type:"toolCall",id:c.call_id??`call_${c.tool_index}`,name:c.name??"",arguments:{}};r.content.push(j),y={call:j,args:"",contentIndex:r.content.length-1,toolIndex:c.tool_index},a.push({type:"toolcall_start",contentIndex:y.contentIndex,partial:r})}c.call_id&&(y.call.id=c.call_id),c.name&&(y.call.name=c.name);let I=typeof c.delta=="string"?c.delta:"";I&&(y.args+=I,a.push({type:"toolcall_delta",contentIndex:y.contentIndex,delta:I,partial:r}))}else if(f.type==="response.completed"){S=f.response;break}else if(f.type==="error"){let c=f.error,I=new Error(`uRun serve error: ${c?.message??"unknown error"}`);throw I.code=c?.code??null,I}}if(!S)throw new L(`uRun backhaul for "${s.id}" closed before the turn completed \u2014 the serving session ended mid-turn (its pod was restarted, drained or deleted).`);b=void 0;break}catch(g){if(b=g,w.signal.aborted)throw g;if(l<B&&Ce(g)&&!C){await Ne(l*500);continue}throw g}if(b)throw b;return S};try{let _=Re(d),T;for(let l=1;l<=X;l++)try{T=await ee(_);break}catch(g){let A=P||g instanceof L;if(l>=X||!A||C||E)throw g;O(),t.evict(s.id),P=!1,w=new AbortController,Q(n,`uRun: the "${s.id}" session ended mid-turn \u2014 re-homing to a fresh assignment and replaying the turn\u2026`)}if(O(),$(),q(),D(),!r.content.some(l=>l.type==="toolCall"))for(let l of Ue(T))Z(l);if(!r.content.some(l=>l.type==="thinking")){let l=Se(T);if(l){let g={type:"thinking",thinking:""};r.content.push(g);let A=r.content.length-1;a.push({type:"thinking_start",contentIndex:A,partial:r}),g.thinking=l,a.push({type:"thinking_delta",contentIndex:A,delta:l,partial:r}),a.push({type:"thinking_end",contentIndex:A,content:l,partial:r})}}let S=r.content.filter(l=>l.type==="toolCall").length;if((d.tools?.length??0)>0&&S===0){let l=Pe(s.id,W);if(l)throw l}let b=S>0?"toolUse":T?.status==="incomplete"?"length":"stop";r.stopReason=b,a.push({type:"done",reason:b,message:r}),a.end(r)}catch(_){if(p=!0,O(),E||m?.aborted&&!P)r.stopReason="aborted",r.errorMessage="aborted by user",a.push({type:"error",reason:"aborted",error:r});else{let T=P?`uRun stream stalled \u2014 no output for ${Math.round(i/1e3)}s; the model may be hung. Retry, or Ctrl-C to abort.`:_ instanceof Error?_.message:String(_);r.stopReason="error",r.errorMessage=T,a.push({type:"error",reason:"error",error:r})}a.end(r)}finally{O(),R(),m&&m.removeEventListener("abort",M),p&&t.evict(s.id)}})(),a}}function $e(t){return{id:t,name:`uRun ${t}`,api:V,reasoning:!0,input:["text"],cost:{input:0,output:0,cacheRead:0,cacheWrite:0},contextWindow:pe,maxTokens:de}}function De(t={}){return async e=>{let n=t.env??process.env,{apiKey:o,apiUrl:i,fnName:u}=ye(n),s=await Y({apiUrl:i,apiKey:o,fnName:u,fetchImpl:t.fetchImpl}),d=new K(t.openSession??ve,n,u);e.on("session_shutdown",()=>d.closeAll());let x={name:"uRun",api:V,baseUrl:"urun://session",apiKey:"$URUN_API_KEY",models:s.map($e),streamSimple:Le(d,t.timing??{},e)};e.registerProvider("urun",x)}}var Xe=De();export{K as SessionPool,V as URUN_API,Te as coalesceSameRole,we as createFileDiagnosticSink,De as createUrunExtension,Xe as default,xe as makeSessionFactory,Le as makeStreamSimple,_e as piDiagnosticLogPath,he as resolveSessionEnv,Re as toResponsesInput};