@urun-sh/openai 0.6.17 → 0.6.19

package/dist/hosted/index.d.cts

@@ -1,7 +1,7 @@
 import { Server } from 'node:http';
 import * as _prometheus_io_client from '@prometheus-io/client';
 import { Registry } from '@prometheus-io/client';
-import { c as LedgerOutcome, I as InferenceRecord, S as SessionGoneError, P as ProxyClients, M as ModelRouter, i as UsageRowReporter, b as InferenceUsageRecord, e as LedgerWrite, d as LedgerRecorded, L as LedgerFailure } from '../responses-turn-Dd-h46sQ.cjs';
+import { c as LedgerOutcome, I as InferenceRecord, S as SessionGoneError, P as ProxyClients, M as ModelRouter, i as UsageRowReporter, b as InferenceUsageRecord, e as LedgerWrite, d as LedgerRecorded, L as LedgerFailure } from '../responses-turn-B91w2YKn.cjs';
 import { createClientToken } from '@urun-sh/core';
 import { U as UrunResponses } from '../ResponsesClient-_OZERUjH.cjs';
 import { b as UrunSessionLike } from '../types-CHPtJx6d.cjs';

package/dist/hosted/index.d.ts

@@ -1,7 +1,7 @@
 import { Server } from 'node:http';
 import * as _prometheus_io_client from '@prometheus-io/client';
 import { Registry } from '@prometheus-io/client';
-import { c as LedgerOutcome, I as InferenceRecord, S as SessionGoneError, P as ProxyClients, M as ModelRouter, i as UsageRowReporter, b as InferenceUsageRecord, e as LedgerWrite, d as LedgerRecorded, L as LedgerFailure } from '../responses-turn-B0zlsQGn.js';
+import { c as LedgerOutcome, I as InferenceRecord, S as SessionGoneError, P as ProxyClients, M as ModelRouter, i as UsageRowReporter, b as InferenceUsageRecord, e as LedgerWrite, d as LedgerRecorded, L as LedgerFailure } from '../responses-turn-FcEDaymF.js';
 import { createClientToken } from '@urun-sh/core';
 import { U as UrunResponses } from '../ResponsesClient-CSSrOYD8.js';
 import { b as UrunSessionLike } from '../types-CHPtJx6d.js';

package/dist/hosted/index.js

@@ -1 +1 @@
-import{A as x,B as v,C as m,D as y,E as B,F as H,G as P,H as u,I as G,J as U,K as F,L as X,M as Y,N as b,O as h,P as k,Q as K,R as V,S as j,T as W,U as q,a as r,b as o,c as t,d as E,e as n,f as _,g as a,h as T,i as O,j as i,k as s,l as I,m as p,n as A,o as R,p as S,q as N,r as C,s as D,t as L,u as M,v as d,w as f,x as l,y as c,z as g}from"../chunk-6SN4EJ7Q.js";import"../chunk-VRCCLQ5P.js";import"../chunk-K23AZPI4.js";import"../chunk-U7EJSC2N.js";import"../chunk-76BZLCGT.js";import"../chunk-TMQJY56F.js";import"../chunk-3JWYIKHM.js";import"../chunk-6M5JK4YC.js";import"../chunk-7V2JKPIG.js";import{c as e}from"../chunk-YSFSRI3D.js";e();export{N as ANONYMOUS_KEY_ID,y as BIND_HOST,R as BUCKET_DOMAIN,O as BoundedLabel,f as ControlPlaneUnavailableError,X as ConversationBackhauls,F as ConversationCapacityError,v as DEFAULT_PORT,I as EDGE_IDENTITY_PORT,A as EDGE_KEY_FINGERPRINT_HEADER,p as EDGE_KEY_ID_HEADER,i as InferenceMetrics,Y as LEDGER_FUNCTION_PATH,b as LEDGER_WRITE_TIMEOUT_MS,V as LedgerClient,h as LedgerDuplicateError,k as LedgerSurfaceError,g as MAX_CACHED_KEYS,G as MAX_CONVERSATIONS,U as MAX_CONVERSATION_HANDLES,n as MAX_MODEL_LABELS,_ as MAX_ORG_LABELS,H as MAX_TENANTS,o as METRICS_PATH,r as METRICS_PORT,E as NO_MODEL_LABEL,c as ORG_BINDING_TTL_MS,t as OTHER_LABEL,x as OrgResolver,d as ProxyAuthError,W as READINESS_TIMEOUT_MS,j as READINESS_TTL_MS,m as SERVE_FUNCTION,S as TIER_DOMAIN,u as TenantRegistry,l as bearerFrom,L as bearerTokenOf,M as createEdgeIdentityServer,q as createHostedProxy,s as createMetricsServer,C as edgeBucketId,D as edgeTierFingerprint,a as laneOf,K as parseLedgerRecorded,B as resolveHostedConfig,T as statusClassOf,P as tenantSubject};
+import{A as x,B as v,C as m,D as y,E as B,F as H,G as P,H as u,I as G,J as U,K as F,L as X,M as Y,N as b,O as h,P as k,Q as K,R as V,S as j,T as W,U as q,a as r,b as o,c as t,d as E,e as n,f as _,g as a,h as T,i as O,j as i,k as s,l as I,m as p,n as A,o as R,p as S,q as N,r as C,s as D,t as L,u as M,v as d,w as f,x as l,y as c,z as g}from"../chunk-DR2ANRFC.js";import"../chunk-MLJ66LPI.js";import"../chunk-K23AZPI4.js";import"../chunk-7TOD3ORT.js";import"../chunk-76BZLCGT.js";import"../chunk-NYSZ3RGB.js";import"../chunk-3JWYIKHM.js";import"../chunk-6M5JK4YC.js";import"../chunk-7V2JKPIG.js";import{c as e}from"../chunk-YSFSRI3D.js";e();export{N as ANONYMOUS_KEY_ID,y as BIND_HOST,R as BUCKET_DOMAIN,O as BoundedLabel,f as ControlPlaneUnavailableError,X as ConversationBackhauls,F as ConversationCapacityError,v as DEFAULT_PORT,I as EDGE_IDENTITY_PORT,A as EDGE_KEY_FINGERPRINT_HEADER,p as EDGE_KEY_ID_HEADER,i as InferenceMetrics,Y as LEDGER_FUNCTION_PATH,b as LEDGER_WRITE_TIMEOUT_MS,V as LedgerClient,h as LedgerDuplicateError,k as LedgerSurfaceError,g as MAX_CACHED_KEYS,G as MAX_CONVERSATIONS,U as MAX_CONVERSATION_HANDLES,n as MAX_MODEL_LABELS,_ as MAX_ORG_LABELS,H as MAX_TENANTS,o as METRICS_PATH,r as METRICS_PORT,E as NO_MODEL_LABEL,c as ORG_BINDING_TTL_MS,t as OTHER_LABEL,x as OrgResolver,d as ProxyAuthError,W as READINESS_TIMEOUT_MS,j as READINESS_TTL_MS,m as SERVE_FUNCTION,S as TIER_DOMAIN,u as TenantRegistry,l as bearerFrom,L as bearerTokenOf,M as createEdgeIdentityServer,q as createHostedProxy,s as createMetricsServer,C as edgeBucketId,D as edgeTierFingerprint,a as laneOf,K as parseLedgerRecorded,B as resolveHostedConfig,T as statusClassOf,P as tenantSubject};

package/dist/pi-extension/index.cjs

@@ -1,8 +1,8 @@
-"use strict";var Oe=Object.create;var F=Object.defineProperty;var Ue=Object.getOwnPropertyDescriptor;var Ne=Object.getOwnPropertyNames;var Le=Object.getPrototypeOf,De=Object.prototype.hasOwnProperty;var Fe=(e,t)=>{for(var n in t)F(e,n,{get:t[n],enumerable:!0})},ie=(e,t,n,o)=>{if(t&&typeof t=="object"||typeof t=="function")for(let r of Ne(t))!De.call(e,r)&&r!==n&&F(e,r,{get:()=>t[r],enumerable:!(o=Ue(t,r))||o.enumerable});return e};var ae=(e,t,n)=>(n=e!=null?Oe(Le(e)):{},ie(t||!e||!e.__esModule?F(n,"default",{value:e,enumerable:!0}):n,e)),je=e=>ie(F({},"__esModule",{value:!0}),e);var ht={};Fe(ht,{SessionPool:()=>G,URUN_API:()=>te,applyServeUsage:()=>Pe,coalesceSameRole:()=>$e,createFileDiagnosticSink:()=>Se,createUrunExtension:()=>Ae,default:()=>mt,makeSessionFactory:()=>xe,makeStreamSimple:()=>Ce,notify:()=>ee,piDiagnosticLogPath:()=>H,resolveSessionEnv:()=>Re,toResponsesInput:()=>Ee});module.exports=je(ht);var Ke=()=>typeof document>"u"?new URL(`file:${__filename}`).href:document.currentScript&&document.currentScript.tagName.toUpperCase()==="SCRIPT"?document.currentScript.src:new URL("main.js",document.baseURI).href,m=Ke();var ke=require("@earendil-works/pi-ai"),M=require("fs"),be=require("path");function le(e,t,n,o){let r=e.response??{},l={request_id:n,consumer_id:o,stream:!0,kind:"chat",messages:t};return typeof r.instructions=="string"&&(l.instructions=r.instructions),Array.isArray(r.modalities)&&(l.modalities=r.modalities),typeof r.temperature=="number"&&(l.temperature=r.temperature),typeof r.max_output_tokens=="number"&&(l.max_output_tokens=r.max_output_tokens),r.tools!==void 0&&(l.tools=r.tools),l}function ue(e,t,n){let o={request_id:t,consumer_id:n,stream:!!e.stream,kind:"responses",input:e.input};return e.model&&(o.model=e.model),e.tools!==void 0&&(o.tools=e.tools),e.tool_choice!==void 0&&(o.tool_choice=e.tool_choice),typeof e.temperature=="number"&&(o.temperature=e.temperature),typeof e.max_output_tokens=="number"&&(o.max_output_tokens=e.max_output_tokens),typeof e.top_p=="number"&&(o.top_p=e.top_p),Array.isArray(e.stop)&&e.stop.length>0&&(o.stop=e.stop),typeof e.reasoning_effort=="string"&&(o.reasoning_effort=e.reasoning_effort),typeof e.instructions=="string"&&(o.instructions=e.instructions),e.chat_template_kwargs!==void 0&&(o.chat_template_kwargs=e.chat_template_kwargs),e.response_format!==void 0&&(o.response_format=e.response_format),typeof e.voice=="string"&&(o.voice=e.voice),typeof e.language=="string"&&(o.language=e.language),o}function ce(e){if(e instanceof Error)return{type:"error",error:{type:"urun_error",code:e.name||null,message:e.message}};if(e&&typeof e=="object"&&e.t==="error"){let t=e,n=t.body??{},o=n.error,r=o?.message??n.message??`unreadable serve error body: ${JSON.stringify(t.body??null)}`;return{type:"error",error:{type:o?.type??"urun_error",code:t.code??o?.code??null,message:r}}}return{type:"error",error:{type:"urun_error",code:null,message:String(e)}}}var Je="llm-resp";async function*V(e,t){let n=`${Je}:${t}`,o=`resp_${t}`;yield{type:"response.created",response:{id:o,status:"in_progress"}};let r=new Map;try{for await(let l of e.stream(n).messages()){let i=l;if(i.t==="delta"){if(typeof i.delta=="string"&&(yield{type:"response.output_text.delta",item_id:o,delta:i.delta}),typeof i.reasoning=="string"&&(yield{type:"response.reasoning_text.delta",item_id:o,delta:i.reasoning}),Array.isArray(i.tool_calls))for(let u of i.tool_calls){let f=typeof u.index=="number"?u.index:0,a=r.get(f)??{};u.id&&(a.id=u.id),u.function?.name&&(a.name=u.function.name),r.set(f,a),yield{type:"response.function_call_arguments.delta",item_id:`fc_${t}_${f}`,tool_index:f,call_id:a.id,name:a.name,delta:u.function?.arguments??""}}}else if(i.t==="response"){yield{type:"response.completed",response:{id:o,status:"completed",...i.body}};return}else if(i.t==="error"){yield ce(i);return}}}finally{try{e.doc("llm").set({requests:{[t]:null}})}catch{}}}var K=class e{constructor(t,n=e.DEFAULT_DOC_SYNC_GATE_TIMEOUT_MS){this.session=t;this.docSyncTimeoutMs=n;this.sessionTag=t.sessionId??globalThis.crypto.randomUUID()}session;docSyncTimeoutMs;sessionTag;static DEFAULT_DOC_SYNC_GATE_TIMEOUT_MS=15e3;get consumerId(){return this.session.consumerId}async ensureDocSynced(){let t=this.session.doc("llm");if(t.synced===!0||typeof t.onSynced!="function")return;let n;await new Promise((r,l)=>{let i=t.onSynced(()=>{clearTimeout(n),r()});n=setTimeout(()=>{i(),l(new Error(`[urun] llm doc plane never synced within ${this.docSyncTimeoutMs}ms \u2014 the request envelope write is withheld rather than raced into a possibly-lost pre-sync write (resumed-attach first-write-lost, prod-usw2 2026-09-23)`))},this.docSyncTimeoutMs)})}write(t){t.session_tag=this.sessionTag,this.session.doc("llm").set({requests:{[t.request_id]:{payload:t,consumer_id:t.consumer_id,stream:t.stream}}})}sendResponses(t,n){return this.writeGated(n,()=>{let o=ue(t,n,this.consumerId);return this.write(o),V(this.session,n)})}sendResponseCreate(t,n,o){return this.writeGated(o,()=>{let r=le(t,n,o,this.consumerId);return this.write(r),V(this.session,o)})}async writeGated(t,n){try{await this.ensureDocSynced()}catch(o){throw new Error(`[urun] responses dial ${t} aborted before its envelope write: ${o instanceof Error?o.message:String(o)}`)}return n()}};var de=0;function Be(){return de+=1,`req_${Date.now().toString(36)}_${de.toString(36)}`}var J=class{transport;constructor(t){this.transport=new K(t)}responses={create:async t=>{let n=Be(),o=await this.transport.sendResponses(t,n);return Object.assign((async function*(){yield*o})(),{requestId:n})}}};var B=require("fs"),ge=ae(require("os"),1),q=require("path"),fe=require("util");function H(e=ge.default.homedir()){return(0,q.join)(e,".urun","logs","pi-extension.log")}var pe=!1;function me(e,t){let n=H();try{(0,B.mkdirSync)((0,q.dirname)(n),{recursive:!0}),(0,B.appendFileSync)(n,`${new Date().toISOString()} [console.${e}] [urun] ${(0,fe.format)(...t)}
-`)}catch(o){if(!pe){pe=!0;try{process.stderr.write(`[urun] pi-extension quiet console: cannot write ${n} (${String(o)}); further urun-side console output will be dropped
-`)}catch{}}}}var g=e=>(...t)=>me(e,t),X={log:g("log"),info:g("info"),warn:g("warn"),error:g("error"),debug:g("debug"),trace:g("trace"),dir:g("dir"),table:g("table"),group:g("group"),groupCollapsed:g("groupCollapsed"),groupEnd:g("groupEnd"),count:g("count"),countReset:g("countReset"),time:g("time"),timeLog:g("timeLog"),timeEnd:g("timeEnd"),assert:(e,...t)=>{e||me("assert",t.length>0?t:["Assertion failed"])}};function Q(e){try{process.stderr.write(e)}catch{}}var Z="https://api.urun.sh/v1";async function he(e){let t=e.fetchImpl??fetch,n=`${e.apiUrl.replace(/\/+$/,"")}/apps`,o=await t(n,{headers:{Authorization:`Bearer ${e.apiKey}`,Accept:"application/json"}});if(!o.ok)throw new Error(`org apps listing failed: GET ${n} \u2192 ${o.status}`);let r=await o.json();if(!Array.isArray(r.apps))throw new Error(`org apps listing returned no "apps" array (GET ${n})`);if(r.truncated===!0)throw new Error(`org apps listing was truncated (GET ${n} returned ${r.apps.length} of more) \u2014 model discovery would silently omit deployed apps`);return r.apps}async function ye(e){let t=await he(e),n=t.filter(o=>o.function_name===e.fnName&&o.deployment_status==="active").map(o=>o.app_slug);if(n.length===0)throw new Error(`urun pi extension: the org has no active deployed app exposing "${e.fnName}" (GET ${e.apiUrl}/apps returned ${t.length} app(s), none servable) \u2014 deploy one with \`urun serve <model>\` first`);return n.sort()}var te="urun-serve",ze="serve",We=131072,Ye=16384,we=3,Ve=42e4,Xe=6e4,_e=2,Qe=15e3;function Ze(e){let t=(e.URUN_API_KEY??"").trim();if(!t)throw new Error("urun pi extension: URUN_API_KEY is required \u2014 model discovery lists the org's deployed serve apps via the org API (URUN_JWT alone cannot list apps)");return{apiKey:t,apiUrl:(e.URUN_API_URL??"").trim()||Z,fnName:(e.URUN_FUNCTION??"").trim()||ze}}function Re(e){let t=(e.URUN_BASE_URL??"").trim(),n=(e.URUN_ORG_ID??"").trim();if(!t)throw new Error("urun pi extension: URUN_BASE_URL is required to open a session");if(/\/v1\/*$/.test(new URL(t).pathname))throw new Error(`urun pi extension: URUN_BASE_URL must be the session-gateway base (e.g. https://api.urun.sh), got ${t} \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=(e.URUN_JWT??"").trim();if(o)return{baseUrl:t,orgId:n,auth:{lane:"jwt",jwt:o}};let r=(e.URUN_API_KEY??"").trim();if(!r)throw new Error("urun pi extension: URUN_JWT or URUN_API_KEY is required to open a session");let l=(e.URUN_GATEWAY_URL??"").trim()||void 0;return{baseUrl:t,orgId:n,auth:{lane:"api-key",apiKey:r,gatewayUrl:l}}}function Se(e){let t=null;return n=>{t===null&&((0,M.mkdirSync)((0,be.dirname)(e),{recursive:!0}),t=(0,M.openSync)(e,"a"));let o=n.detail?` ${JSON.stringify(n.detail)}`:"";(0,M.writeSync)(t,`${new Date().toISOString()} [${n.level}] [urun] ${n.message}${o}
-`)}}function xe(e){let t=Se(H());return async(n,o,r)=>{let{baseUrl:l,orgId:i,auth:u}=Re(n),{App:f,createClientToken:a}=await e(),y=(u.lane==="jwt"?f(o,{baseUrl:l,orgId:i,jwt:u.jwt,diagnosticSink:t}):f(o,{baseUrl:l,orgId:i,diagnosticSink:t,getAccessToken:async()=>(await a(u.apiKey,{baseUrl:u.gatewayUrl,expiresIn:300,allowedFunctions:[`${o}/${r}`]})).token}))[r];if(typeof y!="function")throw new Error(`urun pi extension: app "${o}" has no function "${r}"`);let _=y();if(typeof _.end!="function")throw new Error(`urun pi extension: app "${o}" function "${r}" returned a session without end()`);let P=_.connect;return typeof P=="function"&&await P.call(_),_}}var et=xe(async()=>{let{createRequire:e}=await import("module"),t=e(m);if(typeof globalThis.RTCPeerConnection>"u")try{t.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 t("@urun-sh/core")});function tt(e,t){try{return Promise.resolve(e.end()).then(()=>{},n=>t(n))}catch(n){return t(n),Promise.resolve()}}function nt(e){if(e?.hasUI===!1)return!0;if(e?.hasUI===!0)return!1;let t=process.argv.slice(2);if(t.includes("-p")||t.includes("--print"))return!0;let n=t.indexOf("--mode");return n!==-1&&(t[n+1]==="json"||t[n+1]==="rpc")?!0:!process.stdin.isTTY||!process.stdout.isTTY}function Te(e,t){return`uRun: releasing the pooled "${e}" session failed \u2014 the backend session may stay allocated until the platform reclaims it: ${t instanceof Error?t.message:String(t)}`}var G=class{constructor(t,n,o,r=(l,i)=>X.warn(Te(l,i))){this.open=t;this.env=n;this.fnName=o;this.onReleaseError=r}open;env;fnName;onReleaseError;pool=new Map;releasing=new Set;acquire(t){let n=this.pool.get(t);if(!n){let o=Promise.resolve(this.open(this.env,t,this.fnName)).then(r=>({session:r,responses:new J(r)}));n=o,this.pool.set(t,o),o.catch(()=>{this.pool.get(t)===o&&this.pool.delete(t)})}return n}evict(t){let n=this.pool.get(t);if(!n)return Promise.resolve();this.pool.delete(t);let o=n.then(r=>tt(r.session,l=>this.onReleaseError(t,l)),()=>{});return this.releasing.add(o),o.then(()=>this.releasing.delete(o)),o}closeAll(){let t=[...this.pool.keys()].map(n=>this.evict(n));return Promise.all([...t,...this.releasing]).then(()=>{})}};function Ee(e){let t=[];e.systemPrompt&&t.push({role:"system",content:e.systemPrompt});for(let n of e.messages){if(n.role==="toolResult"){t.push({type:"function_call_output",call_id:n.toolCallId,output:rt(n.content)});continue}if(n.role==="assistant"&&Array.isArray(n.content)){let r="",l=()=>{r&&(t.push({role:"assistant",content:r}),r="")};for(let i of n.content)if(i.type==="text")r+=String(i.text??"");else if(i.type==="toolCall")l(),t.push({type:"function_call",call_id:String(i.id??""),name:String(i.name??""),arguments:JSON.stringify(i.arguments??{})});else{if(i.type==="thinking")continue;throw new Error(`urun pi extension: unsupported assistant content part "${String(i.type)}"`)}l();continue}let o=n.role==="assistant"?"assistant":"user";t.push({role:o,content:ot(n.content)})}return $e(t)}function $e(e){let t=[];for(let n of e){let o=t[t.length-1];o&&"role"in o&&"role"in n&&o.role===n.role?o.content=[o.content,n.content].filter(r=>r.length>0).join(`
+"use strict";var Oe=Object.create;var F=Object.defineProperty;var Ue=Object.getOwnPropertyDescriptor;var Ne=Object.getOwnPropertyNames;var Le=Object.getPrototypeOf,De=Object.prototype.hasOwnProperty;var Fe=(e,t)=>{for(var n in t)F(e,n,{get:t[n],enumerable:!0})},ie=(e,t,n,r)=>{if(t&&typeof t=="object"||typeof t=="function")for(let o of Ne(t))!De.call(e,o)&&o!==n&&F(e,o,{get:()=>t[o],enumerable:!(r=Ue(t,o))||r.enumerable});return e};var ae=(e,t,n)=>(n=e!=null?Oe(Le(e)):{},ie(t||!e||!e.__esModule?F(n,"default",{value:e,enumerable:!0}):n,e)),je=e=>ie(F({},"__esModule",{value:!0}),e);var ht={};Fe(ht,{SessionPool:()=>G,URUN_API:()=>te,applyServeUsage:()=>Pe,coalesceSameRole:()=>Ae,createFileDiagnosticSink:()=>Se,createUrunExtension:()=>$e,default:()=>mt,makeSessionFactory:()=>xe,makeStreamSimple:()=>Ce,notify:()=>ee,piDiagnosticLogPath:()=>H,resolveSessionEnv:()=>ke,toResponsesInput:()=>Ee});module.exports=je(ht);var Ke=()=>typeof document>"u"?new URL(`file:${__filename}`).href:document.currentScript&&document.currentScript.tagName.toUpperCase()==="SCRIPT"?document.currentScript.src:new URL("main.js",document.baseURI).href,g=Ke();var be=require("@earendil-works/pi-ai"),M=require("fs"),Re=require("path");function le(e,t,n,r){let o=e.response??{},l={request_id:n,consumer_id:r,stream:!0,kind:"chat",messages:t};return typeof o.instructions=="string"&&(l.instructions=o.instructions),Array.isArray(o.modalities)&&(l.modalities=o.modalities),typeof o.temperature=="number"&&(l.temperature=o.temperature),typeof o.max_output_tokens=="number"&&(l.max_output_tokens=o.max_output_tokens),o.tools!==void 0&&(l.tools=o.tools),l}function ue(e,t,n){let r={request_id:t,consumer_id:n,stream:!!e.stream,kind:"responses",input:e.input};return e.model&&(r.model=e.model),e.tools!==void 0&&(r.tools=e.tools),e.tool_choice!==void 0&&(r.tool_choice=e.tool_choice),typeof e.temperature=="number"&&(r.temperature=e.temperature),typeof e.max_output_tokens=="number"&&(r.max_output_tokens=e.max_output_tokens),typeof e.top_p=="number"&&(r.top_p=e.top_p),Array.isArray(e.stop)&&e.stop.length>0&&(r.stop=e.stop),typeof e.reasoning_effort=="string"&&(r.reasoning_effort=e.reasoning_effort),typeof e.instructions=="string"&&(r.instructions=e.instructions),e.chat_template_kwargs!==void 0&&(r.chat_template_kwargs=e.chat_template_kwargs),e.response_format!==void 0&&(r.response_format=e.response_format),typeof e.voice=="string"&&(r.voice=e.voice),typeof e.language=="string"&&(r.language=e.language),r}function ce(e){if(e instanceof Error)return{type:"error",error:{type:"urun_error",code:e.name||null,message:e.message}};if(e&&typeof e=="object"&&e.t==="error"){let t=e,n=t.body??{},r=n.error,o=r?.message??n.message??`unreadable serve error body: ${JSON.stringify(t.body??null)}`;return{type:"error",error:{type:r?.type??"urun_error",code:t.code??r?.code??null,message:o}}}return{type:"error",error:{type:"urun_error",code:null,message:String(e)}}}var Je="llm-resp";async function*V(e,t){let n=`${Je}:${t}`,r=`resp_${t}`;yield{type:"response.created",response:{id:r,status:"in_progress"}};let o=new Map;try{for await(let l of e.stream(n).messages()){let i=l;if(i.t==="delta"){if(typeof i.delta=="string"&&(yield{type:"response.output_text.delta",item_id:r,delta:i.delta}),typeof i.reasoning=="string"&&(yield{type:"response.reasoning_text.delta",item_id:r,delta:i.reasoning}),Array.isArray(i.tool_calls))for(let u of i.tool_calls){let m=typeof u.index=="number"?u.index:0,a=o.get(m)??{};u.id&&(a.id=u.id),u.function?.name&&(a.name=u.function.name),o.set(m,a),yield{type:"response.function_call_arguments.delta",item_id:`fc_${t}_${m}`,tool_index:m,call_id:a.id,name:a.name,delta:u.function?.arguments??""}}}else if(i.t==="response"){yield{type:"response.completed",response:{id:r,status:"completed",...i.body}};return}else if(i.t==="error"){yield ce(i);return}}}finally{try{e.doc("llm").set({requests:{[t]:null}})}catch{}}}var K=class e{constructor(t,n=e.DEFAULT_DOC_SYNC_GATE_TIMEOUT_MS){this.session=t;this.docSyncTimeoutMs=n;this.sessionTag=t.sessionId??globalThis.crypto.randomUUID()}session;docSyncTimeoutMs;sessionTag;static DEFAULT_DOC_SYNC_GATE_TIMEOUT_MS=15e3;get consumerId(){return this.session.consumerId}async ensureDocSynced(){let t=this.session.doc("llm");if(t.synced===!0||typeof t.onSynced!="function")return;let n;await new Promise((o,l)=>{let i=t.onSynced(()=>{clearTimeout(n),o()});n=setTimeout(()=>{i(),l(new Error(`[urun] llm doc plane never synced within ${this.docSyncTimeoutMs}ms \u2014 the request envelope write is withheld rather than raced into a possibly-lost pre-sync write (resumed-attach first-write-lost, prod-usw2 2026-09-23)`))},this.docSyncTimeoutMs)})}write(t){t.session_tag=this.sessionTag,this.session.doc("llm").set({requests:{[t.request_id]:{payload:t,consumer_id:t.consumer_id,stream:t.stream}}})}sendResponses(t,n){return this.writeGated(n,()=>{let r=ue(t,n,this.consumerId);return this.write(r),V(this.session,n)})}sendResponseCreate(t,n,r){return this.writeGated(r,()=>{let o=le(t,n,r,this.consumerId);return this.write(o),V(this.session,r)})}async writeGated(t,n){try{await this.ensureDocSynced()}catch(r){throw new Error(`[urun] responses dial ${t} aborted before its envelope write: ${r instanceof Error?r.message:String(r)}`)}return n()}};var de=0;function qe(){return de+=1,`req_${Date.now().toString(36)}_${de.toString(36)}`}var J=class{transport;constructor(t){this.transport=new K(t)}responses={create:async t=>{let n=qe(),r=await this.transport.sendResponses(t,n);return Object.assign((async function*(){yield*r})(),{requestId:n})}}};var q=require("fs"),ge=ae(require("os"),1),B=require("path"),fe=require("util");function H(e=ge.default.homedir()){return(0,B.join)(e,".urun","logs","pi-extension.log")}var pe=!1;function me(e,t){let n=H();try{(0,q.mkdirSync)((0,B.dirname)(n),{recursive:!0}),(0,q.appendFileSync)(n,`${new Date().toISOString()} [console.${e}] [urun] ${(0,fe.format)(...t)}
+`)}catch(r){if(!pe){pe=!0;try{process.stderr.write(`[urun] pi-extension quiet console: cannot write ${n} (${String(r)}); further urun-side console output will be dropped
+`)}catch{}}}}var f=e=>(...t)=>me(e,t),X={log:f("log"),info:f("info"),warn:f("warn"),error:f("error"),debug:f("debug"),trace:f("trace"),dir:f("dir"),table:f("table"),group:f("group"),groupCollapsed:f("groupCollapsed"),groupEnd:f("groupEnd"),count:f("count"),countReset:f("countReset"),time:f("time"),timeLog:f("timeLog"),timeEnd:f("timeEnd"),assert:(e,...t)=>{e||me("assert",t.length>0?t:["Assertion failed"])}};function Q(e){try{process.stderr.write(e)}catch{}}var Z="https://api.urun.sh/v1";async function he(e){let t=e.fetchImpl??fetch,n=`${e.apiUrl.replace(/\/+$/,"")}/apps`,r=await t(n,{headers:{Authorization:`Bearer ${e.apiKey}`,Accept:"application/json"}});if(!r.ok)throw new Error(`org apps listing failed: GET ${n} \u2192 ${r.status}`);let o=await r.json();if(!Array.isArray(o.apps))throw new Error(`org apps listing returned no "apps" array (GET ${n})`);if(o.truncated===!0)throw new Error(`org apps listing was truncated (GET ${n} returned ${o.apps.length} of more) \u2014 model discovery would silently omit deployed apps`);return o.apps}async function ye(e){let t=await he(e),n=t.filter(r=>r.function_name===e.fnName&&r.deployment_status==="active").map(r=>r.app_slug);if(n.length===0)throw new Error(`urun pi extension: the org has no active deployed app exposing "${e.fnName}" (GET ${e.apiUrl}/apps returned ${t.length} app(s), none servable) \u2014 deploy one with \`urun serve <model>\` first`);return n.sort()}var te="urun-serve",ze="serve",We=131072,Ye=16384,we=3,Ve=42e4,Xe=6e4,_e=2,Qe=15e3;function Ze(e){let t=(e.URUN_API_KEY??"").trim();if(!t)throw new Error("urun pi extension: URUN_API_KEY is required \u2014 model discovery lists the org's deployed serve apps via the org API (URUN_JWT alone cannot list apps)");return{apiKey:t,apiUrl:(e.URUN_API_URL??"").trim()||Z,fnName:(e.URUN_FUNCTION??"").trim()||ze}}function ke(e){let t=(e.URUN_BASE_URL??"").trim(),n=(e.URUN_ORG_ID??"").trim();if(!t)throw new Error("urun pi extension: URUN_BASE_URL is required to open a session");if(/\/v1\/*$/.test(new URL(t).pathname))throw new Error(`urun pi extension: URUN_BASE_URL must be the session-gateway base (e.g. https://api.urun.sh), got ${t} \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 r=(e.URUN_JWT??"").trim();if(r)return{baseUrl:t,orgId:n,auth:{lane:"jwt",jwt:r}};let o=(e.URUN_API_KEY??"").trim();if(!o)throw new Error("urun pi extension: URUN_JWT or URUN_API_KEY is required to open a session");let l=(e.URUN_GATEWAY_URL??"").trim()||void 0;return{baseUrl:t,orgId:n,auth:{lane:"api-key",apiKey:o,gatewayUrl:l}}}function Se(e){let t=null;return n=>{t===null&&((0,M.mkdirSync)((0,Re.dirname)(e),{recursive:!0}),t=(0,M.openSync)(e,"a"));let r=n.detail?` ${JSON.stringify(n.detail)}`:"";(0,M.writeSync)(t,`${new Date().toISOString()} [${n.level}] [urun] ${n.message}${r}
+`)}}function xe(e){let t=Se(H());return async(n,r,o)=>{let{baseUrl:l,orgId:i,auth:u}=ke(n),{App:m,createClientToken:a}=await e(),y=(u.lane==="jwt"?m(r,{baseUrl:l,orgId:i,jwt:u.jwt,diagnosticSink:t}):m(r,{baseUrl:l,orgId:i,diagnosticSink:t,getAccessToken:async()=>(await a(u.apiKey,{baseUrl:u.gatewayUrl,expiresIn:300,allowedFunctions:[`${r}/${o}`]})).token}))[o];if(typeof y!="function")throw new Error(`urun pi extension: app "${r}" has no function "${o}"`);let _=y();if(typeof _.end!="function")throw new Error(`urun pi extension: app "${r}" function "${o}" returned a session without end()`);let P=_.connect;return typeof P=="function"&&await P.call(_),_}}var et=xe(async()=>{let{createRequire:e}=await import("module"),t=e(g);if(typeof globalThis.RTCPeerConnection>"u")try{t.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 t("@urun-sh/core")});function tt(e,t){try{return Promise.resolve(e.end()).then(()=>{},n=>t(n))}catch(n){return t(n),Promise.resolve()}}function nt(e){if(e?.hasUI===!1)return!0;if(e?.hasUI===!0)return!1;let t=process.argv.slice(2);if(t.includes("-p")||t.includes("--print"))return!0;let n=t.indexOf("--mode");return n!==-1&&(t[n+1]==="json"||t[n+1]==="rpc")?!0:!process.stdin.isTTY||!process.stdout.isTTY}function Te(e,t){return`uRun: releasing the pooled "${e}" session failed \u2014 the backend session may stay allocated until the platform reclaims it: ${t instanceof Error?t.message:String(t)}`}var G=class{constructor(t,n,r,o=(l,i)=>X.warn(Te(l,i))){this.open=t;this.env=n;this.fnName=r;this.onReleaseError=o}open;env;fnName;onReleaseError;pool=new Map;releasing=new Set;acquire(t){let n=this.pool.get(t);if(!n){let r=Promise.resolve(this.open(this.env,t,this.fnName)).then(o=>({session:o,responses:new J(o)}));n=r,this.pool.set(t,r),r.catch(()=>{this.pool.get(t)===r&&this.pool.delete(t)})}return n}evict(t){let n=this.pool.get(t);if(!n)return Promise.resolve();this.pool.delete(t);let r=n.then(o=>tt(o.session,l=>this.onReleaseError(t,l)),()=>{});return this.releasing.add(r),r.then(()=>this.releasing.delete(r)),r}closeAll(){let t=[...this.pool.keys()].map(n=>this.evict(n));return Promise.all([...t,...this.releasing]).then(()=>{})}};function Ee(e){let t=[];e.systemPrompt&&t.push({role:"system",content:e.systemPrompt});for(let n of e.messages){if(n.role==="toolResult"){t.push({type:"function_call_output",call_id:n.toolCallId,output:ot(n.content)});continue}if(n.role==="assistant"&&Array.isArray(n.content)){let o="",l=()=>{o&&(t.push({role:"assistant",content:o}),o="")};for(let i of n.content)if(i.type==="text")o+=String(i.text??"");else if(i.type==="toolCall")l(),t.push({type:"function_call",call_id:String(i.id??""),name:String(i.name??""),arguments:JSON.stringify(i.arguments??{})});else{if(i.type==="thinking")continue;throw new Error(`urun pi extension: unsupported assistant content part "${String(i.type)}"`)}l();continue}let r=n.role==="assistant"?"assistant":"user";t.push({role:r,content:rt(n.content)})}return Ae(t)}function Ae(e){let t=[];for(let n of e){let r=t[t.length-1];r&&"role"in r&&"role"in n&&r.role===n.role?r.content=[r.content,n.content].filter(o=>o.length>0).join(`
 
-`):t.push({...n})}return t}function ot(e){return typeof e=="string"?e:Array.isArray(e)?e.map(t=>t&&typeof t=="object"&&"text"in t?String(t.text):"").join(""):""}function rt(e){if(typeof e=="string")return e;if(!Array.isArray(e))return"";let t="";for(let n of e){let o=String(n.type??"");if(o!=="text")throw new Error(`urun pi extension: unsupported toolResult content part "${o}" \u2014 the serve lane carries text tool results only`);t+=String(n.text??"")}return t}function st(e){if(!(!Array.isArray(e)||e.length===0))return e.map(t=>({type:"function",name:t.name,description:t.description,parameters:t.parameters}))}function ve(e,t){if(!t.trim())return{};try{return JSON.parse(t)}catch{throw new Error(`uRun serve error: tool call "${e}" arguments are not valid JSON: ${t}`)}}function it(e){let t=e?.output;return Array.isArray(t)?t.filter(n=>n.type==="function_call"):[]}function Pe(e,t){let n=t?.usage;if(!n||typeof n!="object")return;let o=typeof n.input_tokens=="number"?n.input_tokens:void 0,r=typeof n.output_tokens=="number"?n.output_tokens:void 0;o===void 0||r===void 0||(e.input=o,e.output=r,e.totalTokens=typeof n.total_tokens=="number"?n.total_tokens:o+r)}function at(e){let t=e?.output;if(!Array.isArray(t))return"";let n="";for(let o of t)if(!(o.type!=="reasoning"||!Array.isArray(o.content)))for(let r of o.content)r.type==="reasoning_text"&&(n+=String(r.text??""));return n}var lt=["<tool_call>","<function="];function ut(e,t){let n=lt.find(o=>t.includes(o));if(n)return new Error(`serving row emitted a prose tool call (literal ${JSON.stringify(n)} in the assistant text) for a request that carried tools, and no structured tool_calls arrived \u2014 the tool-call parser is not configured on the model row "${e}" (see catalog parser_defaults)`)}var z=class extends Error{};function ct(e){let t=e instanceof Error?e.message:String(e),n=e?.code;return/\b529\b/.test(t)||/overloaded/i.test(t)||n===529||n==="529"}function dt(e){return new Promise(t=>setTimeout(t,e))}function pt(e,t,n,o){return new Promise((r,l)=>{let i=!1,u=s=>{i||(i=!0,clearTimeout(f),o?.removeEventListener("abort",a),s())},f=setTimeout(()=>u(()=>l(n())),t),a=()=>u(()=>l(new Error("aborted")));if(o?.aborted){a();return}o?.addEventListener("abort",a,{once:!0}),Promise.resolve(e).then(s=>u(()=>r(s)),s=>u(()=>l(s)))})}async function*gt(e,t){let n=e[Symbol.asyncIterator](),o,r=new Promise((l,i)=>{o=()=>i(new Error("aborted")),t.addEventListener("abort",o,{once:!0})});try{for(;;){if(t.aborted)throw new Error("aborted");let l=await Promise.race([n.next(),r]);if(l.done)return;yield l.value}}finally{o&&t.removeEventListener("abort",o),n.return?.(void 0)}}function ee(e,t){if(e?.hasUI===!1){Q(`${t}
-`);return}let n=e?.sendMessage;if(typeof n=="function")try{n.call(e,{customType:"urun-status",content:t,display:!0},{triggerTurn:!1})}catch{}}function Ce(e,t={},n){let o=t.connectDeadlineMs??Ve,r=t.stallTimeoutMs??Xe,l=t.phaseHeartbeatMs??Qe;return(i,u,f)=>{let a=(0,ke.createAssistantMessageEventStream)(),s={role:"assistant",content:[],api:i.api,provider:i.provider,model:i.id,usage:{input:0,output:0,cacheRead:0,cacheWrite:0,totalTokens:0,cost:{input:0,output:0,cacheRead:0,cacheWrite:0,total:0}},stopReason:"stop",timestamp:Date.now()},y=new AbortController,_=f?.signal,P=!1,L=()=>{P=!0,y.abort()};_&&(_.aborted?L():_.addEventListener("abort",L,{once:!0}));let O=!1,C,ne=()=>{C&&clearTimeout(C),C=setTimeout(()=>{O=!0,y.abort()},r)},D=()=>{C&&clearTimeout(C),C=void 0},S=null,x=null,v=null,oe="",U=!1,W=()=>{if(!x)return;let p=x;x=null,a.push({type:"thinking_end",contentIndex:s.content.indexOf(p),content:p.thinking,partial:s})},re=()=>{if(!S)return;let p=S;S=null,a.push({type:"text_end",contentIndex:s.content.indexOf(p),content:p.text,partial:s})},Y=()=>{if(!v)return;let{call:p,args:k,contentIndex:T}=v;v=null,p.arguments=ve(p.name,k),a.push({type:"toolcall_end",contentIndex:T,toolCall:p,partial:s})},Ie=p=>{let k={type:"toolCall",id:String(p.call_id??p.id??`call_${s.content.length}`),name:String(p.name??""),arguments:{}};s.content.push(k);let T=s.content.length-1;a.push({type:"toolcall_start",contentIndex:T,partial:s});let N=typeof p.arguments=="string"?p.arguments:"";N&&a.push({type:"toolcall_delta",contentIndex:T,delta:N,partial:s}),k.arguments=ve(k.name,N),a.push({type:"toolcall_end",contentIndex:T,toolCall:k,partial:s})};return(async()=>{a.push({type:"start",partial:s});let p=!1,k,T=()=>{k&&clearInterval(k),k=void 0},N=Date.now();k=setInterval(()=>{let b=Math.round((Date.now()-N)/1e3);ee(n,`uRun: still opening the "${i.id}" session\u2026 (${b}s \u2014 serve apps scale to zero, first turn can cold-start ~6min)`)},l);let Me=async b=>{let{responses:R}=await pt(e.acquire(i.id),o,()=>new Error(`uRun session for "${i.id}" did not connect within ${Math.round(o/1e3)}s \u2014 likely cold-starting (scaled to zero) or queued behind capacity. Retry shortly.`),y.signal);T();let A,E;for(let c=1;c<=we;c++)try{let w=await R.responses.create({model:i.id,input:b,stream:!0,tools:st(u.tools),max_output_tokens:i.maxTokens,...f?.reasoning!=null&&f.reasoning!=="off"&&i.reasoning!==!1?{}:{chat_template_kwargs:{enable_thinking:!1}}});ne();for await(let $ of gt(w,y.signal)){ne();let h=$;if(h.type==="response.output_text.delta"){let d=typeof h.delta=="string"?h.delta:"";d&&(U=!0,W(),Y(),S||(S={type:"text",text:""},s.content.push(S),a.push({type:"text_start",contentIndex:s.content.length-1,partial:s})),S.text+=d,oe+=d,a.push({type:"text_delta",contentIndex:s.content.indexOf(S),delta:d,partial:s}))}else if(h.type==="response.reasoning_text.delta"){let d=typeof h.delta=="string"?h.delta:"";d&&(U=!0,x||(x={type:"thinking",thinking:""},s.content.push(x),a.push({type:"thinking_start",contentIndex:s.content.length-1,partial:s})),x.thinking+=d,a.push({type:"thinking_delta",contentIndex:s.content.indexOf(x),delta:d,partial:s}))}else if(h.type==="response.function_call_arguments.delta"&&typeof h.tool_index=="number"){let d=h;if(U=!0,W(),re(),!v||v.toolIndex!==d.tool_index){Y();let se={type:"toolCall",id:d.call_id??`call_${d.tool_index}`,name:d.name??"",arguments:{}};s.content.push(se),v={call:se,args:"",contentIndex:s.content.length-1,toolIndex:d.tool_index},a.push({type:"toolcall_start",contentIndex:v.contentIndex,partial:s})}d.call_id&&(v.call.id=d.call_id),d.name&&(v.call.name=d.name);let I=typeof d.delta=="string"?d.delta:"";I&&(v.args+=I,a.push({type:"toolcall_delta",contentIndex:v.contentIndex,delta:I,partial:s}))}else if(h.type==="response.completed"){A=h.response;break}else if(h.type==="error"){let d=h.error,I=new Error(`uRun serve error: ${d?.message??"unknown error"}`);throw I.code=d?.code??null,I}}if(!A)throw new z(`uRun backhaul for "${i.id}" closed before the turn completed \u2014 the serving session ended mid-turn (its pod was restarted, drained or deleted).`);E=void 0;break}catch(w){if(E=w,y.signal.aborted)throw w;if(c<we&&ct(w)&&!U){await dt(c*500);continue}throw w}if(E)throw E;return A};try{let b=Ee(u),R;for(let c=1;c<=_e;c++)try{R=await Me(b);break}catch(w){let $=O||w instanceof z;if(c>=_e||!$||U||P)throw w;D(),e.evict(i.id),O=!1,y=new AbortController,ee(n,`uRun: the "${i.id}" session ended mid-turn \u2014 re-homing to a fresh assignment and replaying the turn\u2026`)}if(D(),W(),re(),Y(),!s.content.some(c=>c.type==="toolCall"))for(let c of it(R))Ie(c);if(!s.content.some(c=>c.type==="thinking")){let c=at(R);if(c){let w={type:"thinking",thinking:""};s.content.push(w);let $=s.content.length-1;a.push({type:"thinking_start",contentIndex:$,partial:s}),w.thinking=c,a.push({type:"thinking_delta",contentIndex:$,delta:c,partial:s}),a.push({type:"thinking_end",contentIndex:$,content:c,partial:s})}}let A=s.content.filter(c=>c.type==="toolCall").length;if((u.tools?.length??0)>0&&A===0){let c=ut(i.id,oe);if(c)throw c}let E=A>0?"toolUse":R?.status==="incomplete"?"length":"stop";s.stopReason=E,Pe(s.usage,R),a.push({type:"done",reason:E,message:s}),a.end(s)}catch(b){if(p=!0,D(),P||_?.aborted&&!O)s.stopReason="aborted",s.errorMessage="aborted by user",a.push({type:"error",reason:"aborted",error:s});else{let R=O?`uRun stream stalled \u2014 no output for ${Math.round(r/1e3)}s; the model may be hung. Retry, or Ctrl-C to abort.`:b instanceof Error?b.message:String(b);s.stopReason="error",s.errorMessage=R,a.push({type:"error",reason:"error",error:s})}a.end(s)}finally{D(),T(),_&&_.removeEventListener("abort",L),p&&e.evict(i.id)}})(),a}}function ft(e){return{id:e,name:`uRun ${e}`,api:te,reasoning:!0,input:["text"],cost:{input:0,output:0,cacheRead:0,cacheWrite:0},contextWindow:We,maxTokens:Ye}}function Ae(e={}){return async t=>{let n=e.env??process.env,{apiKey:o,apiUrl:r,fnName:l}=Ze(n),i=await ye({apiUrl:r,apiKey:o,fnName:l,fetchImpl:e.fetchImpl}),u=new G(e.openSession??et,n,l,(a,s)=>{let y=Te(a,s);nt(t)?Q(`${y}
-`):X.warn(y)});t.on("session_shutdown",()=>u.closeAll());let f={name:"uRun",api:te,baseUrl:"urun://session",apiKey:"$URUN_API_KEY",models:i.map(ft),streamSimple:Ce(u,e.timing??{},t)};t.registerProvider("urun",f)}}var mt=Ae();0&&(module.exports={SessionPool,URUN_API,applyServeUsage,coalesceSameRole,createFileDiagnosticSink,createUrunExtension,makeSessionFactory,makeStreamSimple,notify,piDiagnosticLogPath,resolveSessionEnv,toResponsesInput});
+`):t.push({...n})}return t}function rt(e){return typeof e=="string"?e:Array.isArray(e)?e.map(t=>t&&typeof t=="object"&&"text"in t?String(t.text):"").join(""):""}function ot(e){if(typeof e=="string")return e;if(!Array.isArray(e))return"";let t="";for(let n of e){let r=String(n.type??"");if(r!=="text")throw new Error(`urun pi extension: unsupported toolResult content part "${r}" \u2014 the serve lane carries text tool results only`);t+=String(n.text??"")}return t}function st(e){if(!(!Array.isArray(e)||e.length===0))return e.map(t=>({type:"function",name:t.name,description:t.description,parameters:t.parameters}))}function ve(e,t){if(!t.trim())return{};try{return JSON.parse(t)}catch{throw new Error(`uRun serve error: tool call "${e}" arguments are not valid JSON: ${t}`)}}function it(e){let t=e?.output;return Array.isArray(t)?t.filter(n=>n.type==="function_call"):[]}function Pe(e,t){let n=t?.usage;if(!n||typeof n!="object")return;let r=typeof n.input_tokens=="number"?n.input_tokens:void 0,o=typeof n.output_tokens=="number"?n.output_tokens:void 0;r===void 0||o===void 0||(e.input=r,e.output=o,e.totalTokens=typeof n.total_tokens=="number"?n.total_tokens:r+o)}function at(e){let t=e?.output;if(!Array.isArray(t))return"";let n="";for(let r of t)if(!(r.type!=="reasoning"||!Array.isArray(r.content)))for(let o of r.content)o.type==="reasoning_text"&&(n+=String(o.text??""));return n}var lt=["<tool_call>","<function="];function ut(e,t){let n=lt.find(r=>t.includes(r));if(n)return new Error(`serving row emitted a prose tool call (literal ${JSON.stringify(n)} in the assistant text) for a request that carried tools, and no structured tool_calls arrived \u2014 the tool-call parser is not configured on the model row "${e}" (see catalog parser_defaults)`)}var z=class extends Error{};function ct(e){let t=e instanceof Error?e.message:String(e),n=e?.code;return/\b529\b/.test(t)||/overloaded/i.test(t)||n===529||n==="529"}function dt(e){return new Promise(t=>setTimeout(t,e))}function pt(e,t,n,r){return new Promise((o,l)=>{let i=!1,u=s=>{i||(i=!0,clearTimeout(m),r?.removeEventListener("abort",a),s())},m=setTimeout(()=>u(()=>l(n())),t),a=()=>u(()=>l(new Error("aborted")));if(r?.aborted){a();return}r?.addEventListener("abort",a,{once:!0}),Promise.resolve(e).then(s=>u(()=>o(s)),s=>u(()=>l(s)))})}async function*gt(e,t){let n=e[Symbol.asyncIterator](),r,o=new Promise((l,i)=>{r=()=>i(new Error("aborted")),t.addEventListener("abort",r,{once:!0})});try{for(;;){if(t.aborted)throw new Error("aborted");let l=await Promise.race([n.next(),o]);if(l.done)return;yield l.value}}finally{r&&t.removeEventListener("abort",r),n.return?.(void 0)}}function ee(e,t){if(e?.hasUI===!1){Q(`${t}
+`);return}let n=e?.sendMessage;if(typeof n=="function")try{n.call(e,{customType:"urun-status",content:t,display:!0},{triggerTurn:!1})}catch{}}function Ce(e,t={},n){let r=t.connectDeadlineMs??Ve,o=t.stallTimeoutMs??Xe,l=t.phaseHeartbeatMs??Qe;return(i,u,m)=>{let a=(0,be.createAssistantMessageEventStream)(),s={role:"assistant",content:[],api:i.api,provider:i.provider,model:i.id,usage:{input:0,output:0,cacheRead:0,cacheWrite:0,totalTokens:0,cost:{input:0,output:0,cacheRead:0,cacheWrite:0,total:0}},stopReason:"stop",timestamp:Date.now()},y=new AbortController,_=m?.signal,P=!1,L=()=>{P=!0,y.abort()};_&&(_.aborted?L():_.addEventListener("abort",L,{once:!0}));let O=!1,C,ne=()=>{C&&clearTimeout(C),C=setTimeout(()=>{O=!0,y.abort()},o)},D=()=>{C&&clearTimeout(C),C=void 0},S=null,x=null,v=null,re="",U=!1,W=()=>{if(!x)return;let p=x;x=null,a.push({type:"thinking_end",contentIndex:s.content.indexOf(p),content:p.thinking,partial:s})},oe=()=>{if(!S)return;let p=S;S=null,a.push({type:"text_end",contentIndex:s.content.indexOf(p),content:p.text,partial:s})},Y=()=>{if(!v)return;let{call:p,args:b,contentIndex:T}=v;v=null,p.arguments=ve(p.name,b),a.push({type:"toolcall_end",contentIndex:T,toolCall:p,partial:s})},Ie=p=>{let b={type:"toolCall",id:String(p.call_id??p.id??`call_${s.content.length}`),name:String(p.name??""),arguments:{}};s.content.push(b);let T=s.content.length-1;a.push({type:"toolcall_start",contentIndex:T,partial:s});let N=typeof p.arguments=="string"?p.arguments:"";N&&a.push({type:"toolcall_delta",contentIndex:T,delta:N,partial:s}),b.arguments=ve(b.name,N),a.push({type:"toolcall_end",contentIndex:T,toolCall:b,partial:s})};return(async()=>{a.push({type:"start",partial:s});let p=!1,b,T=()=>{b&&clearInterval(b),b=void 0},N=Date.now();b=setInterval(()=>{let R=Math.round((Date.now()-N)/1e3);ee(n,`uRun: still opening the "${i.id}" session\u2026 (${R}s \u2014 serve apps scale to zero, first turn can cold-start ~6min)`)},l);let Me=async R=>{let{responses:k}=await pt(e.acquire(i.id),r,()=>new Error(`uRun session for "${i.id}" did not connect within ${Math.round(r/1e3)}s \u2014 likely cold-starting (scaled to zero) or queued behind capacity. Retry shortly.`),y.signal);T();let $,E;for(let c=1;c<=we;c++)try{let w=await k.responses.create({model:i.id,input:R,stream:!0,tools:st(u.tools),max_output_tokens:i.maxTokens,...m?.reasoning!=null&&m.reasoning!=="off"&&i.reasoning!==!1?{}:{chat_template_kwargs:{enable_thinking:!1}}});ne();for await(let A of gt(w,y.signal)){ne();let h=A;if(h.type==="response.output_text.delta"){let d=typeof h.delta=="string"?h.delta:"";d&&(U=!0,W(),Y(),S||(S={type:"text",text:""},s.content.push(S),a.push({type:"text_start",contentIndex:s.content.length-1,partial:s})),S.text+=d,re+=d,a.push({type:"text_delta",contentIndex:s.content.indexOf(S),delta:d,partial:s}))}else if(h.type==="response.reasoning_text.delta"){let d=typeof h.delta=="string"?h.delta:"";d&&(U=!0,x||(x={type:"thinking",thinking:""},s.content.push(x),a.push({type:"thinking_start",contentIndex:s.content.length-1,partial:s})),x.thinking+=d,a.push({type:"thinking_delta",contentIndex:s.content.indexOf(x),delta:d,partial:s}))}else if(h.type==="response.function_call_arguments.delta"&&typeof h.tool_index=="number"){let d=h;if(U=!0,W(),oe(),!v||v.toolIndex!==d.tool_index){Y();let se={type:"toolCall",id:d.call_id??`call_${d.tool_index}`,name:d.name??"",arguments:{}};s.content.push(se),v={call:se,args:"",contentIndex:s.content.length-1,toolIndex:d.tool_index},a.push({type:"toolcall_start",contentIndex:v.contentIndex,partial:s})}d.call_id&&(v.call.id=d.call_id),d.name&&(v.call.name=d.name);let I=typeof d.delta=="string"?d.delta:"";I&&(v.args+=I,a.push({type:"toolcall_delta",contentIndex:v.contentIndex,delta:I,partial:s}))}else if(h.type==="response.completed"){$=h.response;break}else if(h.type==="error"){let d=h.error,I=new Error(`uRun serve error: ${d?.message??"unknown error"}`);throw I.code=d?.code??null,I}}if(!$)throw new z(`uRun backhaul for "${i.id}" closed before the turn completed \u2014 the serving session ended mid-turn (its pod was restarted, drained or deleted).`);E=void 0;break}catch(w){if(E=w,y.signal.aborted)throw w;if(c<we&&ct(w)&&!U){await dt(c*500);continue}throw w}if(E)throw E;return $};try{let R=Ee(u),k;for(let c=1;c<=_e;c++)try{k=await Me(R);break}catch(w){let A=O||w instanceof z;if(c>=_e||!A||U||P)throw w;D(),e.evict(i.id),O=!1,y=new AbortController,ee(n,`uRun: the "${i.id}" session ended mid-turn \u2014 re-homing to a fresh assignment and replaying the turn\u2026`)}if(D(),W(),oe(),Y(),!s.content.some(c=>c.type==="toolCall"))for(let c of it(k))Ie(c);if(!s.content.some(c=>c.type==="thinking")){let c=at(k);if(c){let w={type:"thinking",thinking:""};s.content.push(w);let A=s.content.length-1;a.push({type:"thinking_start",contentIndex:A,partial:s}),w.thinking=c,a.push({type:"thinking_delta",contentIndex:A,delta:c,partial:s}),a.push({type:"thinking_end",contentIndex:A,content:c,partial:s})}}let $=s.content.filter(c=>c.type==="toolCall").length;if((u.tools?.length??0)>0&&$===0){let c=ut(i.id,re);if(c)throw c}let E=$>0?"toolUse":k?.status==="incomplete"?"length":"stop";s.stopReason=E,Pe(s.usage,k),a.push({type:"done",reason:E,message:s}),a.end(s)}catch(R){if(p=!0,D(),P||_?.aborted&&!O)s.stopReason="aborted",s.errorMessage="aborted by user",a.push({type:"error",reason:"aborted",error:s});else{let k=O?`uRun stream stalled \u2014 no output for ${Math.round(o/1e3)}s; the model may be hung. Retry, or Ctrl-C to abort.`:R instanceof Error?R.message:String(R);s.stopReason="error",s.errorMessage=k,a.push({type:"error",reason:"error",error:s})}a.end(s)}finally{D(),T(),_&&_.removeEventListener("abort",L),p&&e.evict(i.id)}})(),a}}function ft(e){return{id:e,name:`uRun ${e}`,api:te,reasoning:!0,input:["text"],cost:{input:0,output:0,cacheRead:0,cacheWrite:0},contextWindow:We,maxTokens:Ye}}function $e(e={}){return async t=>{let n=e.env??process.env,{apiKey:r,apiUrl:o,fnName:l}=Ze(n),i=await ye({apiUrl:o,apiKey:r,fnName:l,fetchImpl:e.fetchImpl}),u=new G(e.openSession??et,n,l,(a,s)=>{let y=Te(a,s);nt(t)?Q(`${y}
+`):X.warn(y)});t.on("session_shutdown",()=>u.closeAll());let m={name:"uRun",api:te,baseUrl:"urun://session",apiKey:"$URUN_API_KEY",models:i.map(ft),streamSimple:Ce(u,e.timing??{},t)};t.registerProvider("urun",m)}}var mt=$e();0&&(module.exports={SessionPool,URUN_API,applyServeUsage,coalesceSameRole,createFileDiagnosticSink,createUrunExtension,makeSessionFactory,makeStreamSimple,notify,piDiagnosticLogPath,resolveSessionEnv,toResponsesInput});

package/dist/pi-extension/index.js

@@ -1,4 +1,4 @@
-import{a as z}from"../chunk-76BZLCGT.js";import{c as W,f as Q}from"../chunk-TMQJY56F.js";import"../chunk-7V2JKPIG.js";import{c as C}from"../chunk-YSFSRI3D.js";C();import{createAssistantMessageEventStream as he}from"@earendil-works/pi-ai";import{mkdirSync as we,openSync as _e,writeSync as ke}from"fs";import{dirname as ve}from"path";C();import{appendFileSync as de,mkdirSync as pe}from"fs";import fe from"os";import{dirname as ge,join as me}from"path";import{format as ye}from"util";function J(t=fe.homedir()){return me(t,".urun","logs","pi-extension.log")}var V=!1;function te(t,e){let n=J();try{pe(ge(n),{recursive:!0}),de(n,`${new Date().toISOString()} [console.${t}] [urun] ${ye(...e)}
+import{a as z}from"../chunk-76BZLCGT.js";import{e as W,h as Q}from"../chunk-NYSZ3RGB.js";import"../chunk-7V2JKPIG.js";import{c as C}from"../chunk-YSFSRI3D.js";C();import{createAssistantMessageEventStream as he}from"@earendil-works/pi-ai";import{mkdirSync as we,openSync as _e,writeSync as ke}from"fs";import{dirname as ve}from"path";C();import{appendFileSync as de,mkdirSync as pe}from"fs";import fe from"os";import{dirname as ge,join as me}from"path";import{format as ye}from"util";function J(t=fe.homedir()){return me(t,".urun","logs","pi-extension.log")}var V=!1;function te(t,e){let n=J();try{pe(ge(n),{recursive:!0}),de(n,`${new Date().toISOString()} [console.${t}] [urun] ${ye(...e)}
 `)}catch(o){if(!V){V=!0;try{process.stderr.write(`[urun] pi-extension quiet console: cannot write ${n} (${String(o)}); further urun-side console output will be dropped
 `)}catch{}}}}var f=t=>(...e)=>te(t,e),j={log:f("log"),info:f("info"),warn:f("warn"),error:f("error"),debug:f("debug"),trace:f("trace"),dir:f("dir"),table:f("table"),group:f("group"),groupCollapsed:f("groupCollapsed"),groupEnd:f("groupEnd"),count:f("count"),countReset:f("countReset"),time:f("time"),timeLog:f("timeLog"),timeEnd:f("timeEnd"),assert:(t,...e)=>{t||te("assert",e.length>0?e:["Assertion failed"])}};function q(t){try{process.stderr.write(t)}catch{}}C();async function ne(t){let e=await Q(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 ${e.length} app(s), none servable) \u2014 deploy one with \`urun serve <model>\` first`);return n.sort()}var ae="urun-serve",xe="serve",Re=131072,be=16384,oe=3,Te=42e4,Ee=6e4,re=2,Se=15e3;function Ae(t){let e=(t.URUN_API_KEY??"").trim();if(!e)throw new Error("urun pi extension: URUN_API_KEY is required \u2014 model discovery lists the org's deployed serve apps via the org API (URUN_JWT alone cannot list apps)");return{apiKey:e,apiUrl:(t.URUN_API_URL??"").trim()||W,fnName:(t.URUN_FUNCTION??"").trim()||xe}}function Ue(t){let e=(t.URUN_BASE_URL??"").trim(),n=(t.URUN_ORG_ID??"").trim();if(!e)throw new Error("urun pi extension: URUN_BASE_URL is required to open a session");if(/\/v1\/*$/.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 s=(t.URUN_API_KEY??"").trim();if(!s)throw new Error("urun pi extension: URUN_JWT or URUN_API_KEY is required to open a session");let l=(t.URUN_GATEWAY_URL??"").trim()||void 0;return{baseUrl:e,orgId:n,auth:{lane:"api-key",apiKey:s,gatewayUrl:l}}}function Ie(t){let e=null;return n=>{e===null&&(we(ve(t),{recursive:!0}),e=_e(t,"a"));let o=n.detail?` ${JSON.stringify(n.detail)}`:"";ke(e,`${new Date().toISOString()} [${n.level}] [urun] ${n.message}${o}
 `)}}function Pe(t){let e=Ie(J());return async(n,o,s)=>{let{baseUrl:l,orgId:i,auth:p}=Ue(n),{App:k,createClientToken:a}=await t(),m=(p.lane==="jwt"?k(o,{baseUrl:l,orgId:i,jwt:p.jwt,diagnosticSink:e}):k(o,{baseUrl:l,orgId:i,diagnosticSink:e,getAccessToken:async()=>(await a(p.apiKey,{baseUrl:p.gatewayUrl,expiresIn:300,allowedFunctions:[`${o}/${s}`]})).token}))[s];if(typeof m!="function")throw new Error(`urun pi extension: app "${o}" has no function "${s}"`);let h=m();if(typeof h.end!="function")throw new Error(`urun pi extension: app "${o}" function "${s}" returned a session without end()`);let A=h.connect;return typeof A=="function"&&await A.call(h),h}}var Ce=Pe(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 Ne(t,e){try{return Promise.resolve(t.end()).then(()=>{},n=>e(n))}catch(n){return e(n),Promise.resolve()}}function $e(t){if(t?.hasUI===!1)return!0;if(t?.hasUI===!0)return!1;let e=process.argv.slice(2);if(e.includes("-p")||e.includes("--print"))return!0;let n=e.indexOf("--mode");return n!==-1&&(e[n+1]==="json"||e[n+1]==="rpc")?!0:!process.stdin.isTTY||!process.stdout.isTTY}function le(t,e){return`uRun: releasing the pooled "${t}" session failed \u2014 the backend session may stay allocated until the platform reclaims it: ${e instanceof Error?e.message:String(e)}`}var H=class{constructor(e,n,o,s=(l,i)=>j.warn(le(l,i))){this.open=e;this.env=n;this.fnName=o;this.onReleaseError=s}open;env;fnName;onReleaseError;pool=new Map;releasing=new Set;acquire(e){let n=this.pool.get(e);if(!n){let o=Promise.resolve(this.open(this.env,e,this.fnName)).then(s=>({session:s,responses:new z(s)}));n=o,this.pool.set(e,o),o.catch(()=>{this.pool.get(e)===o&&this.pool.delete(e)})}return n}evict(e){let n=this.pool.get(e);if(!n)return Promise.resolve();this.pool.delete(e);let o=n.then(s=>Ne(s.session,l=>this.onReleaseError(e,l)),()=>{});return this.releasing.add(o),o.then(()=>this.releasing.delete(o)),o}closeAll(){let e=[...this.pool.keys()].map(n=>this.evict(n));return Promise.all([...e,...this.releasing]).then(()=>{})}};function Le(t){let e=[];t.systemPrompt&&e.push({role:"system",content:t.systemPrompt});for(let n of t.messages){if(n.role==="toolResult"){e.push({type:"function_call_output",call_id:n.toolCallId,output:De(n.content)});continue}if(n.role==="assistant"&&Array.isArray(n.content)){let s="",l=()=>{s&&(e.push({role:"assistant",content:s}),s="")};for(let i of n.content)if(i.type==="text")s+=String(i.text??"");else if(i.type==="toolCall")l(),e.push({type:"function_call",call_id:String(i.id??""),name:String(i.name??""),arguments:JSON.stringify(i.arguments??{})});else{if(i.type==="thinking")continue;throw new Error(`urun pi extension: unsupported assistant content part "${String(i.type)}"`)}l();continue}let o=n.role==="assistant"?"assistant":"user";e.push({role:o,content:Me(n.content)})}return Oe(e)}function Oe(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(s=>s.length>0).join(`