@ljoukov/llm 4.0.7 → 4.0.9

package/dist/index.js

@@ -329,6 +329,34 @@ function estimateCallCostUsd({
 import os2 from "os";
 import { TextDecoder } from "util";
 
+// src/utils/runtimeSingleton.ts
+var runtimeSingletonStoreKey = /* @__PURE__ */ Symbol.for("@ljoukov/llm.runtimeSingletonStore");
+function getRuntimeSingletonStore() {
+  const globalObject = globalThis;
+  const existingStore = globalObject[runtimeSingletonStoreKey];
+  if (existingStore) {
+    return existingStore;
+  }
+  const store = /* @__PURE__ */ new Map();
+  Object.defineProperty(globalObject, runtimeSingletonStoreKey, {
+    value: store,
+    enumerable: false,
+    configurable: false,
+    writable: false
+  });
+  return store;
+}
+function getRuntimeSingleton(key, create) {
+  const store = getRuntimeSingletonStore();
+  const existingValue = store.get(key);
+  if (existingValue !== void 0) {
+    return existingValue;
+  }
+  const createdValue = create();
+  store.set(key, createdValue);
+  return createdValue;
+}
+
 // src/openai/chatgpt-auth.ts
 import { Buffer as Buffer2 } from "buffer";
 import fs2 from "fs";
@@ -339,14 +367,16 @@ import { z } from "zod";
 // src/utils/env.ts
 import fs from "fs";
 import path from "path";
-var envLoaded = false;
+var envState = getRuntimeSingleton(/* @__PURE__ */ Symbol.for("@ljoukov/llm.envState"), () => ({
+  envLoaded: false
+}));
 function loadLocalEnv() {
-  if (envLoaded) {
+  if (envState.envLoaded) {
     return;
   }
   const envPath = path.join(process.cwd(), ".env.local");
   loadEnvFromFile(envPath, { override: false });
-  envLoaded = true;
+  envState.envLoaded = true;
 }
 function loadEnvFromFile(filePath, { override = false } = {}) {
   let content;
@@ -432,8 +462,10 @@ var ExchangeResponseSchema = z.object({
   expires_in: z.union([z.number(), z.string()]),
   id_token: z.string().optional()
 });
-var cachedProfile = null;
-var refreshPromise = null;
+var chatGptAuthState = getRuntimeSingleton(/* @__PURE__ */ Symbol.for("@ljoukov/llm.chatGptAuthState"), () => ({
+  cachedProfile: null,
+  refreshPromise: null
+}));
 async function fetchChatGptAuthProfileFromTokenProvider(options) {
   const base = options.baseUrl.replace(/\/+$/u, "");
   const store = options.store?.trim() ? options.store.trim() : "kv";
@@ -534,13 +566,13 @@ async function getChatGptAuthProfile() {
   const tokenProviderUrl = process.env[CHATGPT_AUTH_TOKEN_PROVIDER_URL_ENV];
   const tokenProviderKey = process.env[CHATGPT_AUTH_TOKEN_PROVIDER_API_KEY_ENV] ?? process.env[CHATGPT_AUTH_API_KEY_ENV];
   if (tokenProviderUrl && tokenProviderUrl.trim().length > 0 && tokenProviderKey && tokenProviderKey.trim().length > 0) {
-    if (cachedProfile && !isExpired(cachedProfile)) {
-      return cachedProfile;
+    if (chatGptAuthState.cachedProfile && !isExpired(chatGptAuthState.cachedProfile)) {
+      return chatGptAuthState.cachedProfile;
     }
-    if (refreshPromise) {
-      return refreshPromise;
+    if (chatGptAuthState.refreshPromise) {
+      return chatGptAuthState.refreshPromise;
     }
-    refreshPromise = (async () => {
+    chatGptAuthState.refreshPromise = (async () => {
       try {
         const store = process.env[CHATGPT_AUTH_TOKEN_PROVIDER_STORE_ENV];
         const profile = await fetchChatGptAuthProfileFromTokenProvider({
@@ -548,31 +580,31 @@ async function getChatGptAuthProfile() {
           apiKey: tokenProviderKey,
           store: store ?? void 0
         });
-        cachedProfile = profile;
+        chatGptAuthState.cachedProfile = profile;
         return profile;
       } finally {
-        refreshPromise = null;
+        chatGptAuthState.refreshPromise = null;
       }
     })();
-    return refreshPromise;
+    return chatGptAuthState.refreshPromise;
   }
-  if (cachedProfile && !isExpired(cachedProfile)) {
-    return cachedProfile;
+  if (chatGptAuthState.cachedProfile && !isExpired(chatGptAuthState.cachedProfile)) {
+    return chatGptAuthState.cachedProfile;
   }
-  if (refreshPromise) {
-    return refreshPromise;
+  if (chatGptAuthState.refreshPromise) {
+    return chatGptAuthState.refreshPromise;
   }
-  refreshPromise = (async () => {
+  chatGptAuthState.refreshPromise = (async () => {
     try {
-      const baseProfile = cachedProfile ?? loadAuthProfileFromCodexStore();
+      const baseProfile = chatGptAuthState.cachedProfile ?? loadAuthProfileFromCodexStore();
       const profile = isExpired(baseProfile) ? await refreshAndPersistCodexProfile(baseProfile) : baseProfile;
-      cachedProfile = profile;
+      chatGptAuthState.cachedProfile = profile;
       return profile;
     } finally {
-      refreshPromise = null;
+      chatGptAuthState.refreshPromise = null;
     }
   })();
-  return refreshPromise;
+  return chatGptAuthState.refreshPromise;
 }
 function resolveCodexHome() {
   const codexHome = process.env.CODEX_HOME;
@@ -1304,8 +1336,10 @@ function createAbortError(reason) {
 // src/openai/chatgpt-codex.ts
 var CHATGPT_CODEX_ENDPOINT = "https://chatgpt.com/backend-api/codex/responses";
 var CHATGPT_RESPONSES_EXPERIMENTAL_HEADER = "responses=experimental";
-var cachedResponsesWebSocketMode = null;
-var chatGptResponsesWebSocketDisabled = false;
+var chatGptCodexState = getRuntimeSingleton(/* @__PURE__ */ Symbol.for("@ljoukov/llm.chatGptCodexState"), () => ({
+  cachedResponsesWebSocketMode: null,
+  chatGptResponsesWebSocketDisabled: false
+}));
 async function streamChatGptCodexResponse(options) {
   const { access, accountId } = await getChatGptAuthProfile();
   const mode = resolveChatGptResponsesWebSocketMode();
@@ -1331,7 +1365,7 @@ async function streamChatGptCodexResponse(options) {
       }
     };
   };
-  if (mode === "off" || chatGptResponsesWebSocketDisabled) {
+  if (mode === "off" || chatGptCodexState.chatGptResponsesWebSocketDisabled) {
     return fallbackStreamFactory();
   }
   const websocketHeaders = buildChatGptCodexHeaders({
@@ -1350,7 +1384,7 @@ async function streamChatGptCodexResponse(options) {
     }),
     createFallbackStream: fallbackStreamFactory,
     onWebSocketFallback: () => {
-      chatGptResponsesWebSocketDisabled = true;
+      chatGptCodexState.chatGptResponsesWebSocketDisabled = true;
     }
   });
 }
@@ -1380,14 +1414,14 @@ async function streamChatGptCodexResponseSse(options) {
   return parseEventStream(body);
 }
 function resolveChatGptResponsesWebSocketMode() {
-  if (cachedResponsesWebSocketMode) {
-    return cachedResponsesWebSocketMode;
+  if (chatGptCodexState.cachedResponsesWebSocketMode) {
+    return chatGptCodexState.cachedResponsesWebSocketMode;
   }
-  cachedResponsesWebSocketMode = resolveResponsesWebSocketMode(
+  chatGptCodexState.cachedResponsesWebSocketMode = resolveResponsesWebSocketMode(
     process.env.CHATGPT_RESPONSES_WEBSOCKET_MODE ?? process.env.OPENAI_RESPONSES_WEBSOCKET_MODE,
     "auto"
   );
-  return cachedResponsesWebSocketMode;
+  return chatGptCodexState.cachedResponsesWebSocketMode;
 }
 function buildChatGptCodexHeaders(options) {
   const openAiBeta = options.useWebSocket ? mergeOpenAiBetaHeader(
@@ -1425,8 +1459,8 @@ async function collectChatGptCodexResponse(options) {
         }
       });
     } catch (error) {
-      if (!sawAnyDelta && !retriedViaSseFallback && shouldRetryViaSseFallback(error) && !chatGptResponsesWebSocketDisabled) {
-        chatGptResponsesWebSocketDisabled = true;
+      if (!sawAnyDelta && !retriedViaSseFallback && shouldRetryViaSseFallback(error) && !chatGptCodexState.chatGptResponsesWebSocketDisabled) {
+        chatGptCodexState.chatGptResponsesWebSocketDisabled = true;
         retriedViaSseFallback = true;
         continue;
       }
@@ -1662,7 +1696,12 @@ var MODEL_CONCURRENCY_PROVIDERS = [
   "google",
   "fireworks"
 ];
-var configuredModelConcurrency = normalizeModelConcurrencyConfig({});
+var modelConcurrencyState = getRuntimeSingleton(
+  /* @__PURE__ */ Symbol.for("@ljoukov/llm.modelConcurrencyState"),
+  () => ({
+    configuredModelConcurrency: normalizeModelConcurrencyConfig({})
+  })
+);
 function clampModelConcurrencyCap(value) {
   if (!Number.isFinite(value)) {
     return DEFAULT_MODEL_CONCURRENCY_CAP;
@@ -1736,14 +1775,14 @@ function resolveDefaultProviderCap(provider, modelId) {
   return DEFAULT_FIREWORKS_MODEL_CONCURRENCY_CAP;
 }
 function configureModelConcurrency(config = {}) {
-  configuredModelConcurrency = normalizeModelConcurrencyConfig(config);
+  modelConcurrencyState.configuredModelConcurrency = normalizeModelConcurrencyConfig(config);
 }
 function resetModelConcurrencyConfig() {
-  configuredModelConcurrency = normalizeModelConcurrencyConfig({});
+  modelConcurrencyState.configuredModelConcurrency = normalizeModelConcurrencyConfig({});
 }
 function resolveModelConcurrencyCap(options) {
   const modelId = options.modelId ? normalizeModelIdForConfig(options.modelId) : void 0;
-  const config = options.config ? normalizeModelConcurrencyConfig(options.config) : configuredModelConcurrency;
+  const config = options.config ? normalizeModelConcurrencyConfig(options.config) : modelConcurrencyState.configuredModelConcurrency;
   const providerModelCap = modelId ? config.providerModelCaps[options.provider].get(modelId) : void 0;
   if (providerModelCap !== void 0) {
     return providerModelCap;
@@ -1977,32 +2016,37 @@ import OpenAI from "openai";
 import { Agent, fetch as undiciFetch } from "undici";
 var DEFAULT_FIREWORKS_BASE_URL = "https://api.fireworks.ai/inference/v1";
 var DEFAULT_FIREWORKS_TIMEOUT_MS = 15 * 6e4;
-var cachedClient = null;
-var cachedFetch = null;
-var cachedBaseUrl = null;
-var cachedApiKey = null;
-var cachedTimeoutMs = null;
+var fireworksClientState = getRuntimeSingleton(
+  /* @__PURE__ */ Symbol.for("@ljoukov/llm.fireworksClientState"),
+  () => ({
+    cachedClient: null,
+    cachedFetch: null,
+    cachedBaseUrl: null,
+    cachedApiKey: null,
+    cachedTimeoutMs: null
+  })
+);
 function resolveTimeoutMs() {
-  if (cachedTimeoutMs !== null) {
-    return cachedTimeoutMs;
+  if (fireworksClientState.cachedTimeoutMs !== null) {
+    return fireworksClientState.cachedTimeoutMs;
   }
   const raw = process.env.FIREWORKS_TIMEOUT_MS;
   const parsed = raw ? Number(raw) : Number.NaN;
-  cachedTimeoutMs = Number.isFinite(parsed) && parsed > 0 ? parsed : DEFAULT_FIREWORKS_TIMEOUT_MS;
-  return cachedTimeoutMs;
+  fireworksClientState.cachedTimeoutMs = Number.isFinite(parsed) && parsed > 0 ? parsed : DEFAULT_FIREWORKS_TIMEOUT_MS;
+  return fireworksClientState.cachedTimeoutMs;
 }
 function resolveBaseUrl() {
-  if (cachedBaseUrl !== null) {
-    return cachedBaseUrl;
+  if (fireworksClientState.cachedBaseUrl !== null) {
+    return fireworksClientState.cachedBaseUrl;
   }
   loadLocalEnv();
   const raw = process.env.FIREWORKS_BASE_URL?.trim();
-  cachedBaseUrl = raw && raw.length > 0 ? raw : DEFAULT_FIREWORKS_BASE_URL;
-  return cachedBaseUrl;
+  fireworksClientState.cachedBaseUrl = raw && raw.length > 0 ? raw : DEFAULT_FIREWORKS_BASE_URL;
+  return fireworksClientState.cachedBaseUrl;
 }
 function resolveApiKey() {
-  if (cachedApiKey !== null) {
-    return cachedApiKey;
+  if (fireworksClientState.cachedApiKey !== null) {
+    return fireworksClientState.cachedApiKey;
   }
   loadLocalEnv();
   const raw = process.env.FIREWORKS_TOKEN ?? process.env.FIREWORKS_API_KEY;
@@ -2012,46 +2056,51 @@ function resolveApiKey() {
       "FIREWORKS_TOKEN (or FIREWORKS_API_KEY) must be provided to access Fireworks APIs."
     );
   }
-  cachedApiKey = token;
-  return cachedApiKey;
+  fireworksClientState.cachedApiKey = token;
+  return fireworksClientState.cachedApiKey;
 }
 function getFireworksFetch() {
-  if (cachedFetch) {
-    return cachedFetch;
+  if (fireworksClientState.cachedFetch) {
+    return fireworksClientState.cachedFetch;
   }
   const timeoutMs = resolveTimeoutMs();
   const dispatcher = new Agent({
     bodyTimeout: timeoutMs,
     headersTimeout: timeoutMs
   });
-  cachedFetch = ((input, init) => {
+  fireworksClientState.cachedFetch = ((input, init) => {
     return undiciFetch(input, {
       ...init ?? {},
       dispatcher
     });
   });
-  return cachedFetch;
+  return fireworksClientState.cachedFetch;
 }
 function getFireworksClient() {
-  if (cachedClient) {
-    return cachedClient;
+  if (fireworksClientState.cachedClient) {
+    return fireworksClientState.cachedClient;
   }
-  cachedClient = new OpenAI({
+  fireworksClientState.cachedClient = new OpenAI({
     apiKey: resolveApiKey(),
     baseURL: resolveBaseUrl(),
     timeout: resolveTimeoutMs(),
     fetch: getFireworksFetch()
   });
-  return cachedClient;
+  return fireworksClientState.cachedClient;
 }
 
 // src/fireworks/calls.ts
 var DEFAULT_SCHEDULER_KEY = "__default__";
-var schedulerByModel = /* @__PURE__ */ new Map();
+var fireworksCallState = getRuntimeSingleton(
+  /* @__PURE__ */ Symbol.for("@ljoukov/llm.fireworksCallState"),
+  () => ({
+    schedulerByModel: /* @__PURE__ */ new Map()
+  })
+);
 function getSchedulerForModel(modelId) {
   const normalizedModelId = modelId?.trim();
   const schedulerKey = normalizedModelId && normalizedModelId.length > 0 ? normalizedModelId : DEFAULT_SCHEDULER_KEY;
-  const existing = schedulerByModel.get(schedulerKey);
+  const existing = fireworksCallState.schedulerByModel.get(schedulerKey);
   if (existing) {
     return existing;
   }
@@ -2063,7 +2112,7 @@ function getSchedulerForModel(modelId) {
     minIntervalBetweenStartMs: 200,
     startJitterMs: 200
   });
-  schedulerByModel.set(schedulerKey, created);
+  fireworksCallState.schedulerByModel.set(schedulerKey, created);
   return created;
 }
 async function runFireworksCall(fn, modelId, runOptions) {
@@ -2110,7 +2159,10 @@ var ServiceAccountSchema = z2.object({
   privateKey: private_key.replace(/\\n/g, "\n"),
   tokenUri: token_uri
 }));
-var cachedServiceAccount = null;
+var googleAuthState = getRuntimeSingleton(/* @__PURE__ */ Symbol.for("@ljoukov/llm.googleAuthState"), () => ({
+  cachedServiceAccount: null,
+  authClientCache: /* @__PURE__ */ new Map()
+}));
 function parseGoogleServiceAccount(input) {
   let parsed;
   try {
@@ -2121,16 +2173,16 @@ function parseGoogleServiceAccount(input) {
   return ServiceAccountSchema.parse(parsed);
 }
 function getGoogleServiceAccount() {
-  if (cachedServiceAccount) {
-    return cachedServiceAccount;
+  if (googleAuthState.cachedServiceAccount) {
+    return googleAuthState.cachedServiceAccount;
   }
   loadLocalEnv();
   const raw = process.env.GOOGLE_SERVICE_ACCOUNT_JSON;
   if (!raw || raw.trim().length === 0) {
     throw new Error("GOOGLE_SERVICE_ACCOUNT_JSON must be provided for Google APIs access.");
   }
-  cachedServiceAccount = parseGoogleServiceAccount(raw);
-  return cachedServiceAccount;
+  googleAuthState.cachedServiceAccount = parseGoogleServiceAccount(raw);
+  return googleAuthState.cachedServiceAccount;
 }
 function normaliseScopes(scopes) {
   if (!scopes) {
@@ -2182,8 +2234,10 @@ function isGeminiImageModelId(value) {
 }
 var CLOUD_PLATFORM_SCOPE = "https://www.googleapis.com/auth/cloud-platform";
 var DEFAULT_VERTEX_LOCATION = "global";
-var geminiConfiguration = {};
-var clientPromise;
+var geminiClientState = getRuntimeSingleton(/* @__PURE__ */ Symbol.for("@ljoukov/llm.geminiClientState"), () => ({
+  geminiConfiguration: {},
+  clientPromise: void 0
+}));
 function normaliseConfigValue(value) {
   if (value === void 0 || value === null) {
     return void 0;
@@ -2194,14 +2248,14 @@ function normaliseConfigValue(value) {
 function configureGemini(options = {}) {
   const nextProjectId = normaliseConfigValue(options.projectId);
   const nextLocation = normaliseConfigValue(options.location);
-  geminiConfiguration = {
-    projectId: nextProjectId !== void 0 ? nextProjectId : geminiConfiguration.projectId,
-    location: nextLocation !== void 0 ? nextLocation : geminiConfiguration.location
+  geminiClientState.geminiConfiguration = {
+    projectId: nextProjectId !== void 0 ? nextProjectId : geminiClientState.geminiConfiguration.projectId,
+    location: nextLocation !== void 0 ? nextLocation : geminiClientState.geminiConfiguration.location
   };
-  clientPromise = void 0;
+  geminiClientState.clientPromise = void 0;
 }
 function resolveProjectId() {
-  const override = geminiConfiguration.projectId;
+  const override = geminiClientState.geminiConfiguration.projectId;
   if (override) {
     return override;
   }
@@ -2209,15 +2263,15 @@ function resolveProjectId() {
   return serviceAccount.projectId;
 }
 function resolveLocation() {
-  const override = geminiConfiguration.location;
+  const override = geminiClientState.geminiConfiguration.location;
   if (override) {
     return override;
   }
   return DEFAULT_VERTEX_LOCATION;
 }
 async function getGeminiClient() {
-  if (!clientPromise) {
-    clientPromise = Promise.resolve().then(() => {
+  if (!geminiClientState.clientPromise) {
+    geminiClientState.clientPromise = Promise.resolve().then(() => {
       const projectId = resolveProjectId();
       const location = resolveLocation();
       const googleAuthOptions = getGoogleAuthOptions(CLOUD_PLATFORM_SCOPE);
@@ -2229,7 +2283,7 @@ async function getGeminiClient() {
       });
     });
   }
-  return clientPromise;
+  return geminiClientState.clientPromise;
 }
 
 // src/google/calls.ts
@@ -2425,11 +2479,13 @@ function retryDelayMs(attempt) {
   return base + jitter;
 }
 var DEFAULT_SCHEDULER_KEY2 = "__default__";
-var schedulerByModel2 = /* @__PURE__ */ new Map();
+var googleCallState = getRuntimeSingleton(/* @__PURE__ */ Symbol.for("@ljoukov/llm.googleCallState"), () => ({
+  schedulerByModel: /* @__PURE__ */ new Map()
+}));
 function getSchedulerForModel2(modelId) {
   const normalizedModelId = modelId?.trim();
   const schedulerKey = normalizedModelId && normalizedModelId.length > 0 ? normalizedModelId : DEFAULT_SCHEDULER_KEY2;
-  const existing = schedulerByModel2.get(schedulerKey);
+  const existing = googleCallState.schedulerByModel.get(schedulerKey);
   if (existing) {
     return existing;
   }
@@ -2452,7 +2508,7 @@ function getSchedulerForModel2(modelId) {
       }
     }
   });
-  schedulerByModel2.set(schedulerKey, created);
+  googleCallState.schedulerByModel.set(schedulerKey, created);
   return created;
 }
 async function runGeminiCall(fn, modelId, runOptions) {
@@ -2462,53 +2518,55 @@ async function runGeminiCall(fn, modelId, runOptions) {
 // src/openai/client.ts
 import OpenAI2 from "openai";
 import { Agent as Agent2, fetch as undiciFetch2 } from "undici";
-var cachedApiKey2 = null;
-var cachedClient2 = null;
-var cachedFetch2 = null;
-var cachedTimeoutMs2 = null;
-var openAiResponsesWebSocketMode = null;
-var openAiResponsesWebSocketDisabled = false;
+var openAiClientState = getRuntimeSingleton(/* @__PURE__ */ Symbol.for("@ljoukov/llm.openAiClientState"), () => ({
+  cachedApiKey: null,
+  cachedClient: null,
+  cachedFetch: null,
+  cachedTimeoutMs: null,
+  openAiResponsesWebSocketMode: null,
+  openAiResponsesWebSocketDisabled: false
+}));
 var DEFAULT_OPENAI_TIMEOUT_MS = 15 * 6e4;
 function resolveOpenAiTimeoutMs() {
-  if (cachedTimeoutMs2 !== null) {
-    return cachedTimeoutMs2;
+  if (openAiClientState.cachedTimeoutMs !== null) {
+    return openAiClientState.cachedTimeoutMs;
   }
   const raw = process.env.OPENAI_STREAM_TIMEOUT_MS ?? process.env.OPENAI_TIMEOUT_MS;
   const parsed = raw ? Number(raw) : Number.NaN;
-  cachedTimeoutMs2 = Number.isFinite(parsed) && parsed > 0 ? parsed : DEFAULT_OPENAI_TIMEOUT_MS;
-  return cachedTimeoutMs2;
+  openAiClientState.cachedTimeoutMs = Number.isFinite(parsed) && parsed > 0 ? parsed : DEFAULT_OPENAI_TIMEOUT_MS;
+  return openAiClientState.cachedTimeoutMs;
 }
 function getOpenAiFetch() {
-  if (cachedFetch2) {
-    return cachedFetch2;
+  if (openAiClientState.cachedFetch) {
+    return openAiClientState.cachedFetch;
   }
   const timeoutMs = resolveOpenAiTimeoutMs();
   const dispatcher = new Agent2({
     bodyTimeout: timeoutMs,
     headersTimeout: timeoutMs
   });
-  cachedFetch2 = ((input, init) => {
+  openAiClientState.cachedFetch = ((input, init) => {
     return undiciFetch2(input, {
       ...init ?? {},
       dispatcher
     });
   });
-  return cachedFetch2;
+  return openAiClientState.cachedFetch;
 }
 function resolveOpenAiBaseUrl() {
   loadLocalEnv();
   return process.env.OPENAI_BASE_URL?.trim() || "https://api.openai.com/v1";
 }
 function resolveOpenAiResponsesWebSocketMode() {
-  if (openAiResponsesWebSocketMode) {
-    return openAiResponsesWebSocketMode;
+  if (openAiClientState.openAiResponsesWebSocketMode) {
+    return openAiClientState.openAiResponsesWebSocketMode;
   }
   loadLocalEnv();
-  openAiResponsesWebSocketMode = resolveResponsesWebSocketMode(
+  openAiClientState.openAiResponsesWebSocketMode = resolveResponsesWebSocketMode(
     process.env.OPENAI_RESPONSES_WEBSOCKET_MODE,
     "auto"
   );
-  return openAiResponsesWebSocketMode;
+  return openAiClientState.openAiResponsesWebSocketMode;
 }
 function wrapFallbackStream(stream) {
   return {
@@ -2561,7 +2619,7 @@ function installResponsesWebSocketTransport(client, apiKey) {
   responsesApi.stream = (request, options) => {
     const mode = resolveOpenAiResponsesWebSocketMode();
     const fallbackStreamFactory = () => wrapFallbackStream(originalStream(request, options));
-    if (mode === "off" || openAiResponsesWebSocketDisabled) {
+    if (mode === "off" || openAiClientState.openAiResponsesWebSocketDisabled) {
       return fallbackStreamFactory();
     }
     const signal = options && typeof options === "object" ? options.signal ?? void 0 : void 0;
@@ -2580,15 +2638,15 @@ function installResponsesWebSocketTransport(client, apiKey) {
       createFallbackStream: fallbackStreamFactory,
       onWebSocketFallback: (error) => {
         if (isResponsesWebSocketUnsupportedError(error)) {
-          openAiResponsesWebSocketDisabled = true;
+          openAiClientState.openAiResponsesWebSocketDisabled = true;
         }
       }
     });
   };
 }
 function getOpenAiApiKey() {
-  if (cachedApiKey2 !== null) {
-    return cachedApiKey2;
+  if (openAiClientState.cachedApiKey !== null) {
+    return openAiClientState.cachedApiKey;
   }
   loadLocalEnv();
   const raw = process.env.OPENAI_API_KEY;
@@ -2596,33 +2654,35 @@ function getOpenAiApiKey() {
   if (!value) {
     throw new Error("OPENAI_API_KEY must be provided to access OpenAI APIs.");
   }
-  cachedApiKey2 = value;
-  return cachedApiKey2;
+  openAiClientState.cachedApiKey = value;
+  return openAiClientState.cachedApiKey;
 }
 function getOpenAiClient() {
-  if (cachedClient2) {
-    return cachedClient2;
+  if (openAiClientState.cachedClient) {
+    return openAiClientState.cachedClient;
   }
   loadLocalEnv();
   const apiKey = getOpenAiApiKey();
   const timeoutMs = resolveOpenAiTimeoutMs();
-  cachedClient2 = new OpenAI2({
+  openAiClientState.cachedClient = new OpenAI2({
     apiKey,
     fetch: getOpenAiFetch(),
     timeout: timeoutMs
   });
-  installResponsesWebSocketTransport(cachedClient2, apiKey);
-  return cachedClient2;
+  installResponsesWebSocketTransport(openAiClientState.cachedClient, apiKey);
+  return openAiClientState.cachedClient;
 }
 
 // src/openai/calls.ts
 var DEFAULT_OPENAI_REASONING_EFFORT = "medium";
 var DEFAULT_SCHEDULER_KEY3 = "__default__";
-var schedulerByModel3 = /* @__PURE__ */ new Map();
+var openAiCallState = getRuntimeSingleton(/* @__PURE__ */ Symbol.for("@ljoukov/llm.openAiCallState"), () => ({
+  schedulerByModel: /* @__PURE__ */ new Map()
+}));
 function getSchedulerForModel3(modelId) {
   const normalizedModelId = modelId?.trim();
   const schedulerKey = normalizedModelId && normalizedModelId.length > 0 ? normalizedModelId : DEFAULT_SCHEDULER_KEY3;
-  const existing = schedulerByModel3.get(schedulerKey);
+  const existing = openAiCallState.schedulerByModel.get(schedulerKey);
   if (existing) {
     return existing;
   }
@@ -2634,7 +2694,7 @@ function getSchedulerForModel3(modelId) {
     minIntervalBetweenStartMs: 200,
     startJitterMs: 200
   });
-  schedulerByModel3.set(schedulerKey, created);
+  openAiCallState.schedulerByModel.set(schedulerKey, created);
   return created;
 }
 async function runOpenAiCall(fn, modelId, runOptions) {
@@ -2709,6 +2769,33 @@ function ensureTrailingNewline(value) {
 function hasNonEmptyText(value) {
   return typeof value === "string" && value.length > 0;
 }
+function hasLogArtifactValue(value) {
+  if (value === null || value === void 0) {
+    return false;
+  }
+  if (typeof value === "string") {
+    return value.length > 0;
+  }
+  if (Array.isArray(value)) {
+    return value.length > 0;
+  }
+  if (typeof value === "object") {
+    return Object.keys(value).length > 0;
+  }
+  return true;
+}
+function serialiseJsonArtifact(value) {
+  if (!hasLogArtifactValue(value)) {
+    return void 0;
+  }
+  try {
+    return `${JSON.stringify(sanitiseLogValue(value), null, 2)}
+`;
+  } catch {
+    return `${JSON.stringify(String(value), null, 2)}
+`;
+  }
+}
 function redactDataUrlPayload(value) {
   if (!value.toLowerCase().startsWith("data:")) {
     return value;
@@ -2989,7 +3076,9 @@ var AgentLoggingSessionImpl = class {
     const responsePath = path3.join(baseDir, "response.txt");
     const thoughtsPath = path3.join(baseDir, "thoughts.txt");
     const toolCallPath = path3.join(baseDir, "tool_call.txt");
+    const toolCallJsonPath = path3.join(baseDir, "tool_call.json");
     const toolCallResponsePath = path3.join(baseDir, "tool_call_response.txt");
+    const toolCallResponseJsonPath = path3.join(baseDir, "tool_call_response.json");
     const errorPath = path3.join(baseDir, "error.txt");
     const responseMetadataPath = path3.join(baseDir, "response.metadata.json");
     let chain = this.ensureReady.then(async () => {
@@ -3020,6 +3109,10 @@ var AgentLoggingSessionImpl = class {
           "utf8"
         );
       }
+      const toolCallResponseJson = serialiseJsonArtifact(input.toolCallResponsePayload);
+      if (toolCallResponseJson) {
+        await writeFile(toolCallResponseJsonPath, toolCallResponseJson, "utf8");
+      }
     }).catch(() => void 0);
     this.track(chain);
     let closed = false;
@@ -3058,6 +3151,10 @@ var AgentLoggingSessionImpl = class {
           if (hasNonEmptyText(options?.toolCallText)) {
             await writeFile(toolCallPath, ensureTrailingNewline(options.toolCallText), "utf8");
           }
+          const toolCallJson = serialiseJsonArtifact(options?.toolCallPayload);
+          if (toolCallJson) {
+            await writeFile(toolCallJsonPath, toolCallJson, "utf8");
+          }
           await this.writeAttachments(baseDir, options?.attachments);
           const payload = {
             capturedAt: toIsoNow(),
@@ -3087,6 +3184,10 @@ var AgentLoggingSessionImpl = class {
           if (hasNonEmptyText(options?.toolCallText)) {
             await writeFile(toolCallPath, ensureTrailingNewline(options.toolCallText), "utf8");
           }
+          const toolCallJson = serialiseJsonArtifact(options?.toolCallPayload);
+          if (toolCallJson) {
+            await writeFile(toolCallJsonPath, toolCallJson, "utf8");
+          }
           await this.writeAttachments(baseDir, options?.attachments);
           await writeFile(errorPath, ensureTrailingNewline(toErrorMessage(error)), "utf8");
           const payload = {
@@ -3120,7 +3221,10 @@ var AgentLoggingSessionImpl = class {
     }
   }
 };
-var loggingSessionStorage = new AsyncLocalStorage();
+var loggingSessionStorage = getRuntimeSingleton(
+  /* @__PURE__ */ Symbol.for("@ljoukov/llm.agentLogging.sessionStorage"),
+  () => new AsyncLocalStorage()
+);
 function createAgentLoggingSession(config) {
   return new AgentLoggingSessionImpl(config);
 }
@@ -3135,7 +3239,10 @@ function getCurrentAgentLoggingSession() {
 }
 
 // src/llm.ts
-var toolCallContextStorage = new AsyncLocalStorage2();
+var toolCallContextStorage = getRuntimeSingleton(
+  /* @__PURE__ */ Symbol.for("@ljoukov/llm.toolCallContextStorage"),
+  () => new AsyncLocalStorage2()
+);
 function getCurrentToolCallContext() {
   return toolCallContextStorage.getStore() ?? null;
 }
@@ -5058,6 +5165,10 @@ function collectLoggedAttachmentsFromGeminiParts(parts, prefix) {
   return collectLoggedAttachmentsFromLlmParts(convertGooglePartsToLlmParts(parts), prefix);
 }
 function extractToolCallResponseTextFromOpenAiInput(input) {
+  const responses = extractToolCallResponsePayloadFromOpenAiInput(input);
+  return serialiseLogArtifactText(responses);
+}
+function extractToolCallResponsePayloadFromOpenAiInput(input) {
   if (!Array.isArray(input)) {
     return void 0;
   }
@@ -5074,9 +5185,13 @@ function extractToolCallResponseTextFromOpenAiInput(input) {
       }
     ];
   });
-  return serialiseLogArtifactText(responses);
+  return responses.length > 0 ? responses : void 0;
 }
 function extractToolCallResponseTextFromFireworksMessages(messages) {
+  const responses = extractToolCallResponsePayloadFromFireworksMessages(messages);
+  return serialiseLogArtifactText(responses);
+}
+function extractToolCallResponsePayloadFromFireworksMessages(messages) {
   if (!Array.isArray(messages)) {
     return void 0;
   }
@@ -5091,9 +5206,13 @@ function extractToolCallResponseTextFromFireworksMessages(messages) {
       }
     ];
   });
-  return serialiseLogArtifactText(responses);
+  return responses.length > 0 ? responses : void 0;
 }
 function extractToolCallResponseTextFromGeminiContents(contents) {
+  const responses = extractToolCallResponsePayloadFromGeminiContents(contents);
+  return serialiseLogArtifactText(responses);
+}
+function extractToolCallResponsePayloadFromGeminiContents(contents) {
   if (!Array.isArray(contents)) {
     return void 0;
   }
@@ -5116,40 +5235,36 @@ function extractToolCallResponseTextFromGeminiContents(contents) {
       }
     }
   }
-  return serialiseLogArtifactText(responses);
+  return responses.length > 0 ? responses : void 0;
 }
-function serialiseOpenAiStyleToolCallsForLogging(calls) {
-  return serialiseLogArtifactText(
-    calls.map((call) => {
-      if (call.kind === "custom") {
-        return {
-          kind: call.kind,
-          name: call.name,
-          callId: call.callId,
-          itemId: call.itemId,
-          input: call.input
-        };
-      }
-      const { value, error } = parseOpenAiToolArguments(call.arguments);
+function toLoggedOpenAiStyleToolCalls(calls) {
+  return calls.map((call) => {
+    if (call.kind === "custom") {
       return {
         kind: call.kind,
         name: call.name,
         callId: call.callId,
         itemId: call.itemId,
-        arguments: value,
-        ...error ? { parseError: error, rawArguments: call.arguments } : {}
+        input: call.input
       };
-    })
-  );
+    }
+    const { value, error } = parseOpenAiToolArguments(call.arguments);
+    return {
+      kind: call.kind,
+      name: call.name,
+      callId: call.callId,
+      itemId: call.itemId,
+      arguments: value,
+      ...error ? { parseError: error, rawArguments: call.arguments } : {}
+    };
+  });
 }
-function serialiseGeminiToolCallsForLogging(calls) {
-  return serialiseLogArtifactText(
-    calls.map((call) => ({
-      name: call.name ?? "unknown",
-      callId: typeof call.id === "string" ? call.id : void 0,
-      arguments: sanitiseLogValue(call.args ?? {})
-    }))
-  );
+function toLoggedGeminiToolCalls(calls) {
+  return calls.map((call) => ({
+    name: call.name ?? "unknown",
+    callId: typeof call.id === "string" ? call.id : void 0,
+    arguments: sanitiseLogValue(call.args ?? {})
+  }));
 }
 function startLlmCallLoggerFromContents(options) {
   const session = getCurrentAgentLoggingSession();
@@ -5229,6 +5344,13 @@ function startLlmCallLoggerFromPayload(options) {
   ) : extractToolCallResponseTextFromGeminiContents(
     options.requestPayload.contents
   );
+  const toolCallResponsePayload = options.provider === "openai" || options.provider === "chatgpt" ? extractToolCallResponsePayloadFromOpenAiInput(
+    options.requestPayload.input
+  ) : options.provider === "fireworks" ? extractToolCallResponsePayloadFromFireworksMessages(
+    options.requestPayload.messages
+  ) : extractToolCallResponsePayloadFromGeminiContents(
+    options.requestPayload.contents
+  );
   return session.startLlmCall({
     provider: options.provider,
     modelId: options.modelId,
@@ -5238,7 +5360,8 @@ function startLlmCallLoggerFromPayload(options) {
       ...getCurrentToolCallContext() ? { toolContext: getCurrentToolCallContext() } : {}
     },
     attachments,
-    toolCallResponseText
+    toolCallResponseText,
+    toolCallResponsePayload
   });
 }
 async function runTextCall(params) {
@@ -5803,7 +5926,10 @@ var DEFAULT_TOOL_LOOP_MAX_STEPS = 8;
 function resolveToolLoopContents(input) {
   return resolveTextContents(input);
 }
-var toolLoopSteeringInternals = /* @__PURE__ */ new WeakMap();
+var toolLoopSteeringInternals = getRuntimeSingleton(
+  /* @__PURE__ */ Symbol.for("@ljoukov/llm.toolLoopSteeringInternals"),
+  () => /* @__PURE__ */ new WeakMap()
+);
 function createToolLoopSteeringChannel() {
   const pending = [];
   let closed = false;
@@ -6066,6 +6192,7 @@ async function runToolLoop(request) {
         let responseText = "";
         let reasoningSummary = "";
         let stepToolCallText;
+        let stepToolCallPayload;
         const stepRequestPayload = {
           model: providerInfo.model,
           input,
@@ -6175,7 +6302,7 @@ async function runToolLoop(request) {
             emitEvent({ type: "usage", usage: usageTokens, costUsd: stepCostUsd, modelVersion });
           }
           const responseToolCalls = extractOpenAiToolCalls(finalResponse.output);
-          stepToolCallText = serialiseOpenAiStyleToolCallsForLogging(
+          stepToolCallPayload = toLoggedOpenAiStyleToolCalls(
             responseToolCalls.map(
               (call) => call.kind === "custom" ? {
                 kind: call.kind,
@@ -6192,6 +6319,7 @@ async function runToolLoop(request) {
               }
             )
           );
+          stepToolCallText = serialiseLogArtifactText(stepToolCallPayload);
           const stepToolCalls = [];
           if (responseToolCalls.length === 0) {
             const steeringInput2 = steeringInternal?.drainPendingContents() ?? [];
@@ -6357,6 +6485,7 @@ async function runToolLoop(request) {
           stepCallLogger?.complete({
             responseText,
             toolCallText: stepToolCallText,
+            toolCallPayload: stepToolCallPayload,
             metadata: {
               provider: "openai",
               model: request.model,
@@ -6377,6 +6506,7 @@ async function runToolLoop(request) {
           stepCallLogger?.fail(error, {
             responseText,
             toolCallText: stepToolCallText,
+            toolCallPayload: stepToolCallPayload,
             metadata: {
               provider: "openai",
               model: request.model,
@@ -6411,6 +6541,7 @@ async function runToolLoop(request) {
         let responseText = "";
         let reasoningSummaryText = "";
         let stepToolCallText;
+        let stepToolCallPayload;
         const markFirstModelEvent = () => {
           if (firstModelEventAtMs === void 0) {
             firstModelEventAtMs = Date.now();
@@ -6479,7 +6610,7 @@ async function runToolLoop(request) {
             stepCallLogger?.appendResponseDelta(responseText);
           }
           const responseToolCalls = response.toolCalls ?? [];
-          stepToolCallText = serialiseOpenAiStyleToolCallsForLogging(
+          stepToolCallPayload = toLoggedOpenAiStyleToolCalls(
             responseToolCalls.map(
               (call) => call.kind === "custom" ? {
                 kind: call.kind,
@@ -6496,6 +6627,7 @@ async function runToolLoop(request) {
               }
             )
           );
+          stepToolCallText = serialiseLogArtifactText(stepToolCallPayload);
           if (responseToolCalls.length === 0) {
             const steeringInput2 = steeringInternal?.drainPendingContents() ?? [];
             const steeringItems2 = steeringInput2.length > 0 ? toChatGptInput(steeringInput2).input : [];
@@ -6668,6 +6800,7 @@ async function runToolLoop(request) {
           stepCallLogger?.complete({
             responseText,
             toolCallText: stepToolCallText,
+            toolCallPayload: stepToolCallPayload,
             metadata: {
               provider: "chatgpt",
               model: request.model,
@@ -6686,6 +6819,7 @@ async function runToolLoop(request) {
           stepCallLogger?.fail(error, {
             responseText,
             toolCallText: stepToolCallText,
+            toolCallPayload: stepToolCallPayload,
             metadata: {
               provider: "chatgpt",
               model: request.model,
@@ -6716,6 +6850,7 @@ async function runToolLoop(request) {
         let responseText = "";
         let blocked = false;
         let stepToolCallText;
+        let stepToolCallPayload;
         const stepRequestPayload = {
           model: providerInfo.model,
           messages,
@@ -6780,7 +6915,7 @@ async function runToolLoop(request) {
             });
           }
           const responseToolCalls = extractFireworksToolCalls(message);
-          stepToolCallText = serialiseOpenAiStyleToolCallsForLogging(
+          stepToolCallPayload = toLoggedOpenAiStyleToolCalls(
             responseToolCalls.map((call) => ({
               kind: "function",
               name: call.name,
@@ -6788,6 +6923,7 @@ async function runToolLoop(request) {
               callId: call.id
             }))
           );
+          stepToolCallText = serialiseLogArtifactText(stepToolCallPayload);
           if (responseToolCalls.length === 0) {
             const steeringInput2 = steeringInternal?.drainPendingContents() ?? [];
             const steeringMessages = steeringInput2.length > 0 ? toFireworksMessages(steeringInput2) : [];
@@ -6940,6 +7076,7 @@ async function runToolLoop(request) {
           stepCallLogger?.complete({
             responseText,
             toolCallText: stepToolCallText,
+            toolCallPayload: stepToolCallPayload,
             metadata: {
               provider: "fireworks",
               model: request.model,
@@ -6968,6 +7105,7 @@ async function runToolLoop(request) {
           stepCallLogger?.fail(error, {
             responseText,
             toolCallText: stepToolCallText,
+            toolCallPayload: stepToolCallPayload,
             metadata: {
               provider: "fireworks",
               model: request.model,
@@ -6996,6 +7134,7 @@ async function runToolLoop(request) {
       let responseText = "";
       let thoughtsText = "";
       let stepToolCallText;
+      let stepToolCallPayload;
       const markFirstModelEvent = () => {
         if (firstModelEventAtMs === void 0) {
           firstModelEventAtMs = Date.now();
@@ -7126,7 +7265,8 @@ async function runToolLoop(request) {
           responseImages: 0
         });
         totalCostUsd += stepCostUsd;
-        stepToolCallText = serialiseGeminiToolCallsForLogging(response.functionCalls);
+        stepToolCallPayload = toLoggedGeminiToolCalls(response.functionCalls);
+        stepToolCallText = serialiseLogArtifactText(stepToolCallPayload);
         if (response.functionCalls.length === 0) {
           const steeringInput2 = steeringInternal?.drainPendingContents() ?? [];
           finalText = responseText;
@@ -7295,6 +7435,7 @@ async function runToolLoop(request) {
           responseText,
           attachments: responseOutputAttachments,
           toolCallText: stepToolCallText,
+          toolCallPayload: stepToolCallPayload,
           metadata: {
             provider: "gemini",
             model: request.model,
@@ -7317,6 +7458,7 @@ async function runToolLoop(request) {
         stepCallLogger?.fail(error, {
           responseText,
           toolCallText: stepToolCallText,
+          toolCallPayload: stepToolCallPayload,
           metadata: {
             provider: "gemini",
             model: request.model,