@earendil-works/pi-ai 1.5.0 → 1.5.1

package/dist/api/openai-responses.js

@@ -67,6 +67,7 @@ function getCompat(model) {
         supportsDeveloperRole: model.compat?.supportsDeveloperRole ?? true,
         supportsMidConvoSystemMessages: model.compat?.supportsMidConvoSystemMessages ?? false,
         sessionAffinityFormat: model.compat?.sessionAffinityFormat ?? detectSessionAffinityFormat(model),
+        promptCacheKeyMode: model.compat?.promptCacheKeyMode ?? "auto",
         supportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? true,
         supportsStrictMode: model.compat?.supportsStrictMode ?? false,
         supportsOpenAIGrammarTools: model.compat?.supportsOpenAIGrammarTools ?? false,
@@ -221,7 +222,7 @@ function createClient(model, context, apiKey, optionsHeaders, identity, fetch, s
         }
         else {
             if (compat.sessionAffinityFormat === "openai") {
-                headers.session_id = sessionId;
+                headers["session-id"] = sessionId;
             }
             headers["x-client-request-id"] = sessionId;
         }
@@ -252,13 +253,14 @@ function buildParams(model, context, options, compat, grammarToolInputProperties
         configurationUpdateEffort,
     });
     const cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);
+    const sendPromptCacheKey = cacheRetention !== "none" && compat.promptCacheKeyMode !== "disabled";
     // Sign in with ChatGPT rejects these request fields.
     const omitUnsupportedFields = isChatGPTSignIn(model, options?.apiKey);
     const params = {
         model: model.id,
         input: messages,
         stream: true,
-        prompt_cache_key: cacheRetention === "none" ? undefined : clampOpenAIPromptCacheKey(options?.sessionId),
+        prompt_cache_key: sendPromptCacheKey ? clampOpenAIPromptCacheKey(options?.sessionId) : undefined,
         prompt_cache_retention: omitUnsupportedFields ? undefined : getPromptCacheRetention(compat, cacheRetention),
         prompt_cache_options: omitUnsupportedFields ? undefined : getPromptCacheOptions(compat, cacheRetention),
         store: false,

package/dist/api/openai-responses.js.map

@@ -1 +1 @@
-{"version":3,"file":"openai-responses.js","sourceRoot":"","sources":["../../src/api/openai-responses.ts"],"names":[],"mappings":"AAAA,OAAO,MAAM,MAAM,QAAQ,CAAC;AAE5B,OAAO,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAelD,OAAO,EAAE,mBAAmB,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AACrF,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACtD,OAAO,EAAuB,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC/E,OAAO,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AAC/D,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAClE,OAAO,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AACrG,OAAO,EAAE,gCAAgC,EAAE,MAAM,2BAA2B,CAAC;AAC7E,OAAO,EAAE,0BAA0B,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AAChG,OAAO,EAAE,yBAAyB,EAAE,MAAM,0BAA0B,CAAC;AACrE,OAAO,EACN,wBAAwB,EACxB,qBAAqB,EACrB,uBAAuB,EACvB,sBAAsB,GACtB,MAAM,8BAA8B,CAAC;AACtC,OAAO,EAAE,gBAAgB,EAAE,qBAAqB,EAAE,MAAM,qBAAqB,CAAC;AAE9E,MAAM,0BAA0B,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,cAAc,EAAE,UAAU,CAAC,CAAC,CAAC;AACnF,wGAAwG;AACxG,MAAM,kCAAkC,GAAG,EAAE,CAAC;AAC9C,MAAM,iBAAiB,GAAG,oCAAoC,CAAC;AAC/D,MAAM,4BAA4B,GACjC,8IAA8I,CAAC;AAChJ,oFAAoF;AACpF,MAAM,mCAAmC,GAAG,QAAQ,CAAC;AAErD;;;GAGG;AACH,SAAS,eAAe,CAAC,KAAgC,EAAE,MAA0B;IACpF,OAAO,CACN,KAAK,CAAC,QAAQ,KAAK,QAAQ;QAC3B,KAAK,CAAC,OAAO,KAAK,2BAA2B;QAC7C,MAAM,KAAK,SAAS;QACpB,CAAC,MAAM,CAAC,UAAU,CAAC,KAAK,CAAC,CACzB,CAAC;AACH,CAAC;AAED,SAAS,SAAS,CAAC,OAAoC,EAAE,IAAY;IACpE,IAAI,CAAC,OAAO;QAAE,OAAO,KAAK,CAAC;IAC3B,MAAM,QAAQ,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;IACpC,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QACpD,IAAI,GAAG,CAAC,WAAW,EAAE,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,IAAI,CAAC;IAC9F,CAAC;IACD,OAAO,KAAK,CAAC;AACd,CAAC;AAED,SAAS,eAAe,CAAC,QAAgB,EAAE,MAA0B,EAAE,OAAoC;IAC1G,IAAI,MAAM;QAAE,OAAO,MAAM,CAAC;IAC1B,IAAI,SAAS,CAAC,OAAO,EAAE,eAAe,CAAC,IAAI,SAAS,CAAC,OAAO,EAAE,sBAAsB,CAAC;QAAE,OAAO,QAAQ,CAAC;IACvG,MAAM,IAAI,KAAK,CAAC,4BAA4B,QAAQ,EAAE,CAAC,CAAC;AACzD,CAAC;AAED,SAAS,2BAA2B,CAAC,KAA8D;IAClG,OAAO,KAAK,CAAC,QAAQ,KAAK,YAAY,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,QAAQ,CAAC;AAC7G,CAAC;AAED;;;GAGG;AACH,SAAS,qBAAqB,CAAC,cAA+B,EAAE,GAAiB;IAChF,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,SAAS,CAAC,KAAgC;IAClD,OAAO;QACN,qBAAqB,EAAE,KAAK,CAAC,MAAM,EAAE,qBAAqB,IAAI,IAAI;QAClE,8BAA8B,EAAE,KAAK,CAAC,MAAM,EAAE,8BAA8B,IAAI,KAAK;QACrF,qBAAqB,EAAE,KAAK,CAAC,MAAM,EAAE,qBAAqB,IAAI,2BAA2B,CAAC,KAAK,CAAC;QAChG,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,IAAI;QAC5E,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,KAAK;QAC7D,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,KAAK;QAC7E,uBAAuB,EAAE,KAAK,CAAC,MAAM,EAAE,uBAAuB,IAAI,KAAK;QACvE,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,KAAK;QAC7D,+BAA+B,EAAE,KAAK,CAAC,MAAM,EAAE,+BAA+B,IAAI,KAAK;QACvF,uBAAuB,EAAE,KAAK,CAAC,MAAM,EAAE,uBAAuB,IAAI,IAAI;QACtE,sBAAsB,EAAE,KAAK,CAAC,MAAM,EAAE,sBAAsB,IAAI,KAAK;KACrE,CAAC;AACH,CAAC;AAED,iGAAiG;AACjG,SAAS,4BAA4B,CACpC,KAAgC,EAChC,OAA2C,EAC3C,MAAuC;IAEvC,IACC,CAAC,MAAM,CAAC,sBAAsB;QAC9B,CAAC,OAAO,EAAE,eAAe;QACzB,2DAA2D;QAC3D,eAAe,CAAC,KAAK,EAAE,OAAO,CAAC,MAAM,CAAC,EACrC,CAAC;QACF,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,OAAO,KAAK,CAAC,gBAAgB,EAAE,CAAC,OAAO,CAAC,eAAe,CAAC,IAAI,OAAO,CAAC,eAAe,CAAC;AACrF,CAAC;AAED,SAAS,uBAAuB,CAC/B,MAAuC,EACvC,cAA8B;IAE9B,OAAO,cAAc,KAAK,MAAM,IAAI,MAAM,CAAC,0BAA0B,IAAI,CAAC,MAAM,CAAC,+BAA+B;QAC/G,CAAC,CAAC,KAAK;QACP,CAAC,CAAC,SAAS,CAAC;AACd,CAAC;AAED,SAAS,qBAAqB,CAC7B,MAAuC,EACvC,cAA8B;IAE9B,IAAI,CAAC,MAAM,CAAC,+BAA+B;QAAE,OAAO,SAAS,CAAC;IAC9D,IAAI,cAAc,KAAK,MAAM;QAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC;IAC3D,IAAI,cAAc,KAAK,MAAM,IAAI,MAAM,CAAC,0BAA0B;QAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC;IAC1F,OAAO,SAAS,CAAC;AAClB,CAAC;AAUD;;GAEG;AACH,MAAM,CAAC,MAAM,MAAM,GAA+D,CACjF,KAAgC,EAChC,OAA0B,EAC1B,OAAgC,EACF,EAAE;IAChC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IACjD,MAAM,iBAAiB,GAAG,iBAAiB,CAAC,OAAO,EAAE,SAAS,CAAC,KAAK,CAAC,CAAC,8BAA8B,CAAC,CAAC;IAEtG,yBAAyB;IACzB,CAAC,KAAK,IAAI,EAAE;QACX,MAAM,MAAM,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;QAChC,MAAM,yBAAyB,GAAG,4BAA4B,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QACvF,MAAM,MAAM,GAAqB;YAChC,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,EAAE;YACX,GAAG,EAAE,KAAK,CAAC,GAAU;YACrB,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,KAAK,CAAC,EAAE;YACf,GAAG,CAAC,yBAAyB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,qBAAqB,EAAE,yBAAyB,EAAE,CAAC;YACxG,KAAK,EAAE;gBACN,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;gBACT,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;aACpE;YACD,UAAU,EAAE,SAAS;YACrB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,IAAI,CAAC;YACJ,uBAAuB;YACvB,MAAM,MAAM,GAAG,eAAe,CAAC,KAAK,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YAClF,MAAM,cAAc,GAAG,qBAAqB,CAAC,OAAO,EAAE,cAAc,EAAE,OAAO,EAAE,GAAG,CAAC,CAAC;YACpF,MAAM,cAAc,GAAG,cAAc,KAAK,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC;YAClF,MAAM,0BAA0B,GAAG,gCAAgC,CAClE,gBAAgB,CAAC,iBAAiB,CAAC,QAAQ,CAAC,EAC5C,MAAM,CAAC,0BAA0B,CACjC,CAAC;YACF,MAAM,MAAM,GAAG,YAAY,CAC1B,KAAK,EACL,iBAAiB,EACjB,MAAM,EACN,OAAO,EAAE,OAAO,EAChB,OAAO,EAAE,QAAQ,EACjB,OAAO,EAAE,KAAK,EACd,cAAc,CACd,CAAC;YACF,IAAI,MAAM,GAAG,WAAW,CACvB,KAAK,EACL,iBAAiB,EACjB,OAAO,EACP,MAAM,EACN,0BAA0B,EAC1B,yBAAyB,CACzB,CAAC;YACF,MAAM,UAAU,GAAG,MAAM,OAAO,EAAE,SAAS,EAAE,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YAC7D,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;gBAC9B,MAAM,GAAG,UAA2C,CAAC;YACtD,CAAC;YACD,MAAM,cAAc,GAAG;gBACtB,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACtD,GAAG,CAAC,OAAO,EAAE,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC3E,UAAU,EAAE,CAAC;aACb,CAAC;YACF,MAAM,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,GAAG,MAAM,oBAAoB,CAClE,GAAG,EAAE,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC,YAAY,EAAE,EACpE;gBACC,UAAU,EAAE,OAAO,EAAE,UAAU;gBAC/B,eAAe,EAAE,OAAO,EAAE,eAAe;gBACzC,MAAM,EAAE,OAAO,EAAE,MAAM;aACvB,CACD,CAAC;YACF,MAAM,OAAO,EAAE,UAAU,EAAE,CAAC,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,OAAO,EAAE,eAAe,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,EAAE,KAAK,CAAC,CAAC;YAC5G,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAEhD,MAAM,sBAAsB,CAAC,YAAY,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE;gBACjE,qBAAqB,EAAE,OAAO,EAAE,qBAAqB;gBACrD,WAAW,EAAE,OAAO,EAAE,WAAW;gBACjC,0BAA0B;gBAC1B,uBAAuB,EAAE,CAAC,KAAK,EAAE,WAAW,EAAE,EAAE,CAAC,uBAAuB,CAAC,KAAK,EAAE,WAAW,EAAE,KAAK,CAAC;aACnG,CAAC,CAAC;YAEH,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;gBACrC,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;YACxE,CAAC;YACD,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,YAAY,IAAI,2BAA2B,CAAC,CAAC;YACrE,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,MAAM,CAAC,UAAU,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;YACnE,MAAM,YAAY,GAAG,mBAAmB,CACvC,sBAAsB,CAAC,KAAK,CAAC,EAC7B,GAAG,KAAK,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,QAAQ,YAAY,CACtE,CAAC;YACF,8EAA8E;YAC9E,6EAA6E;YAC7E,IAAI,YAAY,CAAC,QAAQ,CAAC,2CAA2C,CAAC,EAAE,CAAC;gBACxE,MAAM,CAAC,YAAY,GAAG,GAAG,YAAY,+BAA+B,iBAAiB,EAAE,CAAC;YACzF,CAAC;iBAAM,IAAI,YAAY,CAAC,QAAQ,CAAC,gEAAgE,CAAC,EAAE,CAAC;gBACpG,MAAM,CAAC,YAAY,GAAG,GAAG,YAAY,KAAK,4BAA4B,EAAE,CAAC;YAC1E,CAAC;iBAAM,CAAC;gBACP,MAAM,CAAC,YAAY,GAAG,YAAY,CAAC;YACpC,CAAC;YACD,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;YACzE,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;IACF,CAAC,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AACf,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,YAAY,GAA4D,CACpF,KAAgC,EAChC,OAA0B,EAC1B,OAA6B,EACC,EAAE;IAChC,eAAe,CAAC,KAAK,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;IAEnE,MAAM,IAAI,GAAG;QACZ,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC;QAC7D,UAAU,EAAE,OAAO,EAAE,UAAU;KACE,CAAC;IACnC,MAAM,gBAAgB,GAAG,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,kBAAkB,CAAC,KAAK,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvG,MAAM,eAAe,GAAG,gBAAgB,KAAK,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,CAAC;IAElF,OAAO,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE;QAC7B,GAAG,IAAI;QACP,eAAe;KACkB,CAAC,CAAC;AACrC,CAAC,CAAC;AAEF,SAAS,YAAY,CACpB,KAAgC,EAChC,OAA0B,EAC1B,MAAc,EACd,cAAgC,EAChC,QAAyB,EACzB,KAA+B,EAC/B,SAAkB;IAElB,MAAM,MAAM,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;IAChC,MAAM,OAAO,GAAoB,EAAE,GAAG,kBAAkB,CAAC,KAAK,EAAE,QAAQ,CAAC,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC;IAC9F,IAAI,KAAK,CAAC,QAAQ,KAAK,gBAAgB,EAAE,CAAC;QACzC,MAAM,SAAS,GAAG,qBAAqB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;QAC1D,MAAM,cAAc,GAAG,0BAA0B,CAAC;YACjD,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,SAAS;SACT,CAAC,CAAC;QACH,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;IACxC,CAAC;IAED,IAAI,SAAS,EAAE,CAAC;QACf,IAAI,MAAM,CAAC,qBAAqB,KAAK,YAAY,EAAE,CAAC;YACnD,OAAO,CAAC,cAAc,CAAC,GAAG,SAAS,CAAC;QACrC,CAAC;aAAM,CAAC;YACP,IAAI,MAAM,CAAC,qBAAqB,KAAK,QAAQ,EAAE,CAAC;gBAC/C,OAAO,CAAC,UAAU,GAAG,SAAS,CAAC;YAChC,CAAC;YACD,OAAO,CAAC,qBAAqB,CAAC,GAAG,SAAS,CAAC;QAC5C,CAAC;IACF,CAAC;IAED,2DAA2D;IAC3D,IAAI,cAAc,EAAE,CAAC;QACpB,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;IACxC,CAAC;IAED,OAAO,IAAI,MAAM,CAAC;QACjB,MAAM;QACN,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,uBAAuB,EAAE,IAAI;QAC7B,KAAK;QACL,cAAc,EAAE,OAAO;KACvB,CAAC,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CACnB,KAAgC,EAChC,OAA0B,EAC1B,OAA2C,EAC3C,MAAuC,EACvC,0BAAuD,EACvD,yBAA6C;IAE7C,MAAM,eAAe,GAAG,sBAAsB,CAAC,OAAO,CAAC,QAAQ,EAAE,uBAAuB,CAAC,MAAM,CAAC,CAAC,CAAC;IAClG,MAAM,QAAQ,GAAG,wBAAwB,CAAC,KAAK,EAAE,OAAO,EAAE,0BAA0B,EAAE;QACrF,0BAA0B;QAC1B,8BAA8B,EAAE,MAAM,CAAC,8BAA8B;QACrE,uBAAuB,EAAE,MAAM,CAAC,uBAAuB;QACvD,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;QAC7C,WAAW,EAAE;YACZ,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;YAC7C,0BAA0B,EAAE,MAAM,CAAC,0BAA0B;SAC7D;QACD,yBAAyB;KACzB,CAAC,CAAC;IAEH,MAAM,cAAc,GAAG,qBAAqB,CAAC,OAAO,EAAE,cAAc,EAAE,OAAO,EAAE,GAAG,CAAC,CAAC;IACpF,qDAAqD;IACrD,MAAM,qBAAqB,GAAG,eAAe,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACtE,MAAM,MAAM,GAAkC;QAC7C,KAAK,EAAE,KAAK,CAAC,EAAE;QACf,KAAK,EAAE,QAAQ;QACf,MAAM,EAAE,IAAI;QACZ,gBAAgB,EAAE,cAAc,KAAK,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,yBAAyB,CAAC,OAAO,EAAE,SAAS,CAAC;QACvG,sBAAsB,EAAE,qBAAqB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,uBAAuB,CAAC,MAAM,EAAE,cAAc,CAAC;QAC3G,oBAAoB,EAAE,qBAAqB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,qBAAqB,CAAC,MAAM,EAAE,cAAc,CAAC;QACvG,KAAK,EAAE,KAAK;KACZ,CAAC;IAEF,IAAI,OAAO,EAAE,SAAS,IAAI,MAAM,CAAC,uBAAuB,IAAI,CAAC,qBAAqB,EAAE,CAAC;QACpF,MAAM,CAAC,iBAAiB,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,SAAS,EAAE,kCAAkC,CAAC,CAAC;IAC5F,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,IAAI,CAAC,qBAAqB,EAAE,CAAC;QAClE,MAAM,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,YAAY,GAAG,OAAO,CAAC,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,eAAe,CAAC,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC7C,MAAM,CAAC,KAAK,GAAG,qBAAqB,CAAC,eAAe,CAAC,YAAY,EAAE;YAClE,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;YAC7C,0BAA0B,EAAE,MAAM,CAAC,0BAA0B;SAC7D,CAAC,CAAC;IACJ,CAAC;IAED,IAAI,OAAO,EAAE,UAAU,KAAK,SAAS,EAAE,CAAC;QACvC,MAAM,CAAC,WAAW,GAAG,OAAO,CAAC,UAAU,CAAC;IACzC,CAAC;IAED,MAAM,eAAe,GAAG,OAAO,EAAE,eAAe,IAAI,CAAC,OAAO,EAAE,gBAAgB,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;IACvG,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QACrB,IAAI,eAAe,EAAE,CAAC;YACrB,MAAM,eAAe,GAAG,OAAO,EAAE,eAAe;gBAC/C,CAAC,CAAC,CAAC,KAAK,CAAC,gBAAgB,EAAE,CAAC,OAAO,CAAC,eAAe,CAAC,IAAI,OAAO,CAAC,eAAe,CAAC;gBAChF,CAAC,CAAC,eAAe,CAAC;YACnB,MAAM,MAAM,GAAG,yBAAyB,KAAK,SAAS,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,mCAAmC,CAAC;YAC/G,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,MAAwD;gBAChE,OAAO,EAAE,OAAO,EAAE,gBAAgB,IAAI,MAAM;aAC5C,CAAC;YACF,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAClD,CAAC;aAAM,IAAI,KAAK,CAAC,QAAQ,KAAK,gBAAgB,IAAI,KAAK,CAAC,gBAAgB,EAAE,GAAG,KAAK,IAAI,EAAE,CAAC;YACxF,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,CAAC,KAAK,CAAC,gBAAgB,EAAE,GAAG,IAAI,MAAM,CAAmD;aACjG,CAAC;QACH,CAAC;QACD,IAAI,KAAK,CAAC,QAAQ,KAAK,KAAK;YAAE,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;IAChF,CAAC;IAED,+EAA+E;IAC/E,MAAM,cAAc,GAAG,qBAAqB,CAAC,KAAK,EAAE,eAAe,IAAI,KAAK,EAAE,OAAO,EAAE,cAAc,CAAC,CAAC;IACvG,IAAI,cAAc,EAAE,CAAC;QACpB,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IACvC,CAAC;IAED,OAAO,MAAM,CAAC;AACf,CAAC;AAED,SAAS,4BAA4B,CACpC,KAA4C,EAC5C,WAAsE;IAEtE,QAAQ,WAAW,EAAE,CAAC;QACrB,KAAK,MAAM;YACV,OAAO,GAAG,CAAC;QACZ,KAAK,UAAU,CAAC;QAChB,KAAK,MAAM;YACV,OAAO,KAAK,CAAC,EAAE,KAAK,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QACzC;YACC,OAAO,CAAC,CAAC;IACX,CAAC;AACF,CAAC;AAED,SAAS,uBAAuB,CAC/B,KAAY,EACZ,WAAsE,EACtE,KAA4C;IAE5C,MAAM,UAAU,GAAG,4BAA4B,CAAC,KAAK,EAAE,WAAW,CAAC,CAAC;IACpE,IAAI,UAAU,KAAK,CAAC;QAAE,OAAO;IAE7B,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,UAAU,CAAC;IAC/B,KAAK,CAAC,IAAI,CAAC,MAAM,IAAI,UAAU,CAAC;IAChC,KAAK,CAAC,IAAI,CAAC,SAAS,IAAI,UAAU,CAAC;IACnC,KAAK,CAAC,IAAI,CAAC,UAAU,IAAI,UAAU,CAAC;IACpC,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC;AACxG,CAAC","sourcesContent":["import OpenAI from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { clampThinkingLevel } from \"../models.ts\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tModel,\n\tOpenAIResponsesCompat,\n\tProviderEnv,\n\tProviderHeaders,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tTranscriptContext,\n\tUsage,\n} from \"../types.ts\";\nimport { formatProviderError, normalizeProviderError } from \"../utils/error-body.ts\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { headersToRecord } from \"../utils/headers.ts\";\nimport { type ClientIdentity, getIdentityHeaders } from \"../utils/identity.ts\";\nimport { getProviderEnvValue } from \"../utils/provider-env.ts\";\nimport { retryProviderRequest } from \"../utils/provider-retry.ts\";\nimport { getDeclaredTools, resolveTranscript, resolveTranscriptTools } from \"../utils/transcript.ts\";\nimport { createGrammarToolInputProperties } from \"./constrained-sampling.ts\";\nimport { buildCopilotDynamicHeaders, hasCopilotVisionInput } from \"./github-copilot-headers.ts\";\nimport { clampOpenAIPromptCacheKey } from \"./openai-prompt-cache.ts\";\nimport {\n\tconvertResponsesMessages,\n\tconvertResponsesTools,\n\tgetResponsesToolSupport,\n\tprocessResponsesStream,\n} from \"./openai-responses-shared.ts\";\nimport { buildBaseOptions, resolveSamplingParams } from \"./simple-options.ts\";\n\nconst OPENAI_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\"]);\n// OpenAI Responses rejects max_output_tokens below 16: https://github.com/earendil-works/pi/issues/6265\nconst OPENAI_RESPONSES_MIN_OUTPUT_TOKENS = 16;\nconst CHATGPT_USAGE_URL = \"https://chatgpt.com/settings/usage\";\nconst CHATGPT_WORKSPACE_LIMIT_HINT =\n\t\"You've reached your ChatGPT workspace's usage limit. Ask your workspace admin to increase it at admin.openai.com → Workspace → Usage limits.\";\n// Fixed reasoning.effort while configuration_update items carry the thinking level.\nconst CONFIGURATION_UPDATE_REQUEST_EFFORT = \"medium\";\n\n/**\n * OpenAI API keys start with `sk-`; a different credential sent directly to OpenAI\n * is a Sign in with ChatGPT access token.\n */\nfunction isChatGPTSignIn(model: Model<\"openai-responses\">, apiKey: string | undefined): boolean {\n\treturn (\n\t\tmodel.provider === \"openai\" &&\n\t\tmodel.baseUrl === \"https://api.openai.com/v1\" &&\n\t\tapiKey !== undefined &&\n\t\t!apiKey.startsWith(\"sk-\")\n\t);\n}\n\nfunction hasHeader(headers: ProviderHeaders | undefined, name: string): boolean {\n\tif (!headers) return false;\n\tconst expected = name.toLowerCase();\n\tfor (const [key, value] of Object.entries(headers)) {\n\t\tif (key.toLowerCase() === expected && value !== null && value.trim().length > 0) return true;\n\t}\n\treturn false;\n}\n\nfunction getClientApiKey(provider: string, apiKey: string | undefined, headers: ProviderHeaders | undefined): string {\n\tif (apiKey) return apiKey;\n\tif (hasHeader(headers, \"authorization\") || hasHeader(headers, \"cf-aig-authorization\")) return \"unused\";\n\tthrow new Error(`No API key for provider: ${provider}`);\n}\n\nfunction detectSessionAffinityFormat(model: Pick<Model<\"openai-responses\">, \"provider\" | \"baseUrl\">) {\n\treturn model.provider === \"openrouter\" || model.baseUrl.includes(\"openrouter.ai\") ? \"openrouter\" : \"openai\";\n}\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses PI_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction getCompat(model: Model<\"openai-responses\">): Required<OpenAIResponsesCompat> {\n\treturn {\n\t\tsupportsDeveloperRole: model.compat?.supportsDeveloperRole ?? true,\n\t\tsupportsMidConvoSystemMessages: model.compat?.supportsMidConvoSystemMessages ?? false,\n\t\tsessionAffinityFormat: model.compat?.sessionAffinityFormat ?? detectSessionAffinityFormat(model),\n\t\tsupportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? true,\n\t\tsupportsStrictMode: model.compat?.supportsStrictMode ?? false,\n\t\tsupportsOpenAIGrammarTools: model.compat?.supportsOpenAIGrammarTools ?? false,\n\t\tsupportsAdditionalTools: model.compat?.supportsAdditionalTools ?? false,\n\t\tsupportsToolSearch: model.compat?.supportsToolSearch ?? false,\n\t\tsupportsExplicitPromptCacheMode: model.compat?.supportsExplicitPromptCacheMode ?? false,\n\t\tsupportsMaxOutputTokens: model.compat?.supportsMaxOutputTokens ?? true,\n\t\tsupportsMidConvoEffort: model.compat?.supportsMidConvoEffort ?? false,\n\t};\n}\n\n/** Effort to send in configuration_update items, or undefined to send it as reasoning.effort. */\nfunction getConfigurationUpdateEffort(\n\tmodel: Model<\"openai-responses\">,\n\toptions: OpenAIResponsesOptions | undefined,\n\tcompat: Required<OpenAIResponsesCompat>,\n): string | undefined {\n\tif (\n\t\t!compat.supportsMidConvoEffort ||\n\t\t!options?.reasoningEffort ||\n\t\t// Sign in with ChatGPT rejects configuration_update items.\n\t\tisChatGPTSignIn(model, options.apiKey)\n\t) {\n\t\treturn undefined;\n\t}\n\treturn model.thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort;\n}\n\nfunction getPromptCacheRetention(\n\tcompat: Required<OpenAIResponsesCompat>,\n\tcacheRetention: CacheRetention,\n): \"24h\" | undefined {\n\treturn cacheRetention === \"long\" && compat.supportsLongCacheRetention && !compat.supportsExplicitPromptCacheMode\n\t\t? \"24h\"\n\t\t: undefined;\n}\n\nfunction getPromptCacheOptions(\n\tcompat: Required<OpenAIResponsesCompat>,\n\tcacheRetention: CacheRetention,\n): ResponseCreateParamsStreaming[\"prompt_cache_options\"] {\n\tif (!compat.supportsExplicitPromptCacheMode) return undefined;\n\tif (cacheRetention === \"none\") return { mode: \"explicit\" };\n\tif (cacheRetention === \"long\" && compat.supportsLongCacheRetention) return { ttl: \"30m\" };\n\treturn undefined;\n}\n\n// OpenAI Responses-specific options\nexport interface OpenAIResponsesOptions extends StreamOptions {\n\treasoningEffort?: \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\" | \"max\";\n\treasoningSummary?: \"auto\" | \"detailed\" | \"concise\" | null;\n\tserviceTier?: ResponseCreateParamsStreaming[\"service_tier\"];\n\ttoolChoice?: ResponseCreateParamsStreaming[\"tool_choice\"];\n}\n\n/**\n * Generate function for OpenAI Responses API\n */\nexport const stream: StreamFunction<\"openai-responses\", OpenAIResponsesOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: TranscriptContext,\n\toptions?: OpenAIResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\tconst normalizedContext = resolveTranscript(context, getCompat(model).supportsMidConvoSystemMessages);\n\n\t// Start async processing\n\t(async () => {\n\t\tconst compat = getCompat(model);\n\t\tconst configurationUpdateEffort = getConfigurationUpdateEffort(model, options, compat);\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: model.api as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\t...(configurationUpdateEffort === undefined ? {} : { providerThinkingLevel: configurationUpdateEffort }),\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"pending\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// Create OpenAI client\n\t\t\tconst apiKey = getClientApiKey(model.provider, options?.apiKey, options?.headers);\n\t\t\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);\n\t\t\tconst cacheSessionId = cacheRetention === \"none\" ? undefined : options?.sessionId;\n\t\t\tconst grammarToolInputProperties = createGrammarToolInputProperties(\n\t\t\t\tgetDeclaredTools(normalizedContext.messages),\n\t\t\t\tcompat.supportsOpenAIGrammarTools,\n\t\t\t);\n\t\t\tconst client = createClient(\n\t\t\t\tmodel,\n\t\t\t\tnormalizedContext,\n\t\t\t\tapiKey,\n\t\t\t\toptions?.headers,\n\t\t\t\toptions?.identity,\n\t\t\t\toptions?.fetch,\n\t\t\t\tcacheSessionId,\n\t\t\t);\n\t\t\tlet params = buildParams(\n\t\t\t\tmodel,\n\t\t\t\tnormalizedContext,\n\t\t\t\toptions,\n\t\t\t\tcompat,\n\t\t\t\tgrammarToolInputProperties,\n\t\t\t\tconfigurationUpdateEffort,\n\t\t\t);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as ResponseCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst requestOptions = {\n\t\t\t\t...(options?.signal ? { signal: options.signal } : {}),\n\t\t\t\t...(options?.timeoutMs !== undefined ? { timeout: options.timeoutMs } : {}),\n\t\t\t\tmaxRetries: 0,\n\t\t\t};\n\t\t\tconst { data: openaiStream, response } = await retryProviderRequest(\n\t\t\t\t() => client.responses.create(params, requestOptions).withResponse(),\n\t\t\t\t{\n\t\t\t\t\tmaxRetries: options?.maxRetries,\n\t\t\t\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\t\t\t\tsignal: options?.signal,\n\t\t\t\t},\n\t\t\t);\n\t\t\tawait options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\tawait processResponsesStream(openaiStream, output, stream, model, {\n\t\t\t\tonProviderStreamEvent: options?.onProviderStreamEvent,\n\t\t\t\tserviceTier: options?.serviceTier,\n\t\t\t\tgrammarToolInputProperties,\n\t\t\t\tapplyServiceTierPricing: (usage, serviceTier) => applyServiceTierPricing(usage, serviceTier, model),\n\t\t\t});\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"pending\") {\n\t\t\t\tthrow new Error(\"OpenAI Responses stream ended without a stop reason\");\n\t\t\t}\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(output.errorMessage || \"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\tconst errorMessage = formatProviderError(\n\t\t\t\tnormalizeProviderError(error),\n\t\t\t\t`${model.provider === \"openai\" ? \"OpenAI\" : model.provider} API error`,\n\t\t\t);\n\t\t\t// Sign in with ChatGPT shares the subscription's usage limit with other apps.\n\t\t\t// On ChatGPT Business plans, the workspace admin also sets a per-user limit.\n\t\t\tif (errorMessage.includes(\"subscription_sharing_usage_limit_exceeded\")) {\n\t\t\t\toutput.errorMessage = `${errorMessage}\\nCheck your ChatGPT usage: ${CHATGPT_USAGE_URL}`;\n\t\t\t} else if (errorMessage.includes(\"subscription_sharing_workplace_individual_usage_limit_exceeded\")) {\n\t\t\t\toutput.errorMessage = `${errorMessage}\\n${CHATGPT_WORKSPACE_LIMIT_HINT}`;\n\t\t\t} else {\n\t\t\t\toutput.errorMessage = errorMessage;\n\t\t\t}\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimple: StreamFunction<\"openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: TranscriptContext,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tgetClientApiKey(model.provider, options?.apiKey, options?.headers);\n\n\tconst base = {\n\t\t...buildBaseOptions(model, context, options, options?.apiKey),\n\t\ttoolChoice: options?.toolChoice,\n\t} satisfies OpenAIResponsesOptions;\n\tconst clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : undefined;\n\tconst reasoningEffort = clampedReasoning === \"off\" ? undefined : clampedReasoning;\n\n\treturn stream(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies OpenAIResponsesOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: TranscriptContext,\n\tapiKey: string,\n\toptionsHeaders?: ProviderHeaders,\n\tidentity?: ClientIdentity,\n\tfetch?: typeof globalThis.fetch,\n\tsessionId?: string,\n) {\n\tconst compat = getCompat(model);\n\tconst headers: ProviderHeaders = { ...getIdentityHeaders(model, identity), ...model.headers };\n\tif (model.provider === \"github-copilot\") {\n\t\tconst hasImages = hasCopilotVisionInput(context.messages);\n\t\tconst copilotHeaders = buildCopilotDynamicHeaders({\n\t\t\tmessages: context.messages,\n\t\t\thasImages,\n\t\t});\n\t\tObject.assign(headers, copilotHeaders);\n\t}\n\n\tif (sessionId) {\n\t\tif (compat.sessionAffinityFormat === \"openrouter\") {\n\t\t\theaders[\"x-session-id\"] = sessionId;\n\t\t} else {\n\t\t\tif (compat.sessionAffinityFormat === \"openai\") {\n\t\t\t\theaders.session_id = sessionId;\n\t\t\t}\n\t\t\theaders[\"x-client-request-id\"] = sessionId;\n\t\t}\n\t}\n\n\t// Merge options headers last so they can override defaults\n\tif (optionsHeaders) {\n\t\tObject.assign(headers, optionsHeaders);\n\t}\n\n\treturn new OpenAI({\n\t\tapiKey,\n\t\tbaseURL: model.baseUrl,\n\t\tdangerouslyAllowBrowser: true,\n\t\tfetch,\n\t\tdefaultHeaders: headers,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: TranscriptContext,\n\toptions: OpenAIResponsesOptions | undefined,\n\tcompat: Required<OpenAIResponsesCompat>,\n\tgrammarToolInputProperties: ReadonlyMap<string, string>,\n\tconfigurationUpdateEffort: string | undefined,\n) {\n\tconst transcriptTools = resolveTranscriptTools(context.messages, getResponsesToolSupport(compat));\n\tconst messages = convertResponsesMessages(model, context, OPENAI_TOOL_CALL_PROVIDERS, {\n\t\tgrammarToolInputProperties,\n\t\tsupportsMidConvoSystemMessages: compat.supportsMidConvoSystemMessages,\n\t\tsupportsAdditionalTools: compat.supportsAdditionalTools,\n\t\tsupportsToolSearch: compat.supportsToolSearch,\n\t\ttoolOptions: {\n\t\t\tsupportsStrictMode: compat.supportsStrictMode,\n\t\t\tsupportsOpenAIGrammarTools: compat.supportsOpenAIGrammarTools,\n\t\t},\n\t\tconfigurationUpdateEffort,\n\t});\n\n\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);\n\t// Sign in with ChatGPT rejects these request fields.\n\tconst omitUnsupportedFields = isChatGPTSignIn(model, options?.apiKey);\n\tconst params: ResponseCreateParamsStreaming = {\n\t\tmodel: model.id,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: cacheRetention === \"none\" ? undefined : clampOpenAIPromptCacheKey(options?.sessionId),\n\t\tprompt_cache_retention: omitUnsupportedFields ? undefined : getPromptCacheRetention(compat, cacheRetention),\n\t\tprompt_cache_options: omitUnsupportedFields ? undefined : getPromptCacheOptions(compat, cacheRetention),\n\t\tstore: false,\n\t};\n\n\tif (options?.maxTokens && compat.supportsMaxOutputTokens && !omitUnsupportedFields) {\n\t\tparams.max_output_tokens = Math.max(options.maxTokens, OPENAI_RESPONSES_MIN_OUTPUT_TOKENS);\n\t}\n\n\tif (options?.temperature !== undefined && !omitUnsupportedFields) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (options?.serviceTier !== undefined) {\n\t\tparams.service_tier = options.serviceTier;\n\t}\n\n\tif (transcriptTools.requestTools.length > 0) {\n\t\tparams.tools = convertResponsesTools(transcriptTools.requestTools, {\n\t\t\tsupportsStrictMode: compat.supportsStrictMode,\n\t\t\tsupportsOpenAIGrammarTools: compat.supportsOpenAIGrammarTools,\n\t\t});\n\t}\n\n\tif (options?.toolChoice !== undefined) {\n\t\tparams.tool_choice = options.toolChoice;\n\t}\n\n\tconst reasoningEffort = options?.reasoningEffort ?? (options?.reasoningSummary ? \"medium\" : undefined);\n\tif (model.reasoning) {\n\t\tif (reasoningEffort) {\n\t\t\tconst requestedEffort = options?.reasoningEffort\n\t\t\t\t? (model.thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort)\n\t\t\t\t: reasoningEffort;\n\t\t\tconst effort = configurationUpdateEffort === undefined ? requestedEffort : CONFIGURATION_UPDATE_REQUEST_EFFORT;\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: effort as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t\tsummary: options?.reasoningSummary || \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else if (model.provider !== \"github-copilot\" && model.thinkingLevelMap?.off !== null) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: (model.thinkingLevelMap?.off ?? \"none\") as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t};\n\t\t}\n\t\tif (model.provider === \"xai\") params.include = [\"reasoning.encrypted_content\"];\n\t}\n\n\t// Last so model and request sampling parameters override named request fields.\n\tconst samplingParams = resolveSamplingParams(model, reasoningEffort ?? \"off\", options?.samplingParams);\n\tif (samplingParams) {\n\t\tObject.assign(params, samplingParams);\n\t}\n\n\treturn params;\n}\n\nfunction getServiceTierCostMultiplier(\n\tmodel: Pick<Model<\"openai-responses\">, \"id\">,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n): number {\n\tswitch (serviceTier) {\n\t\tcase \"flex\":\n\t\t\treturn 0.5;\n\t\tcase \"priority\":\n\t\tcase \"fast\":\n\t\t\treturn model.id === \"gpt-5.5\" ? 2.5 : 2;\n\t\tdefault:\n\t\t\treturn 1;\n\t}\n}\n\nfunction applyServiceTierPricing(\n\tusage: Usage,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\tmodel: Pick<Model<\"openai-responses\">, \"id\">,\n) {\n\tconst multiplier = getServiceTierCostMultiplier(model, serviceTier);\n\tif (multiplier === 1) return;\n\n\tusage.cost.input *= multiplier;\n\tusage.cost.output *= multiplier;\n\tusage.cost.cacheRead *= multiplier;\n\tusage.cost.cacheWrite *= multiplier;\n\tusage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;\n}\n"]}
+{"version":3,"file":"openai-responses.js","sourceRoot":"","sources":["../../src/api/openai-responses.ts"],"names":[],"mappings":"AAAA,OAAO,MAAM,MAAM,QAAQ,CAAC;AAE5B,OAAO,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAelD,OAAO,EAAE,mBAAmB,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AACrF,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACtD,OAAO,EAAuB,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC/E,OAAO,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AAC/D,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAClE,OAAO,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AACrG,OAAO,EAAE,gCAAgC,EAAE,MAAM,2BAA2B,CAAC;AAC7E,OAAO,EAAE,0BAA0B,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AAChG,OAAO,EAAE,yBAAyB,EAAE,MAAM,0BAA0B,CAAC;AACrE,OAAO,EACN,wBAAwB,EACxB,qBAAqB,EACrB,uBAAuB,EACvB,sBAAsB,GACtB,MAAM,8BAA8B,CAAC;AACtC,OAAO,EAAE,gBAAgB,EAAE,qBAAqB,EAAE,MAAM,qBAAqB,CAAC;AAE9E,MAAM,0BAA0B,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,cAAc,EAAE,UAAU,CAAC,CAAC,CAAC;AACnF,wGAAwG;AACxG,MAAM,kCAAkC,GAAG,EAAE,CAAC;AAC9C,MAAM,iBAAiB,GAAG,oCAAoC,CAAC;AAC/D,MAAM,4BAA4B,GACjC,8IAA8I,CAAC;AAChJ,oFAAoF;AACpF,MAAM,mCAAmC,GAAG,QAAQ,CAAC;AAErD;;;GAGG;AACH,SAAS,eAAe,CAAC,KAAgC,EAAE,MAA0B;IACpF,OAAO,CACN,KAAK,CAAC,QAAQ,KAAK,QAAQ;QAC3B,KAAK,CAAC,OAAO,KAAK,2BAA2B;QAC7C,MAAM,KAAK,SAAS;QACpB,CAAC,MAAM,CAAC,UAAU,CAAC,KAAK,CAAC,CACzB,CAAC;AACH,CAAC;AAED,SAAS,SAAS,CAAC,OAAoC,EAAE,IAAY;IACpE,IAAI,CAAC,OAAO;QAAE,OAAO,KAAK,CAAC;IAC3B,MAAM,QAAQ,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;IACpC,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QACpD,IAAI,GAAG,CAAC,WAAW,EAAE,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,IAAI,CAAC;IAC9F,CAAC;IACD,OAAO,KAAK,CAAC;AACd,CAAC;AAED,SAAS,eAAe,CAAC,QAAgB,EAAE,MAA0B,EAAE,OAAoC;IAC1G,IAAI,MAAM;QAAE,OAAO,MAAM,CAAC;IAC1B,IAAI,SAAS,CAAC,OAAO,EAAE,eAAe,CAAC,IAAI,SAAS,CAAC,OAAO,EAAE,sBAAsB,CAAC;QAAE,OAAO,QAAQ,CAAC;IACvG,MAAM,IAAI,KAAK,CAAC,4BAA4B,QAAQ,EAAE,CAAC,CAAC;AACzD,CAAC;AAED,SAAS,2BAA2B,CAAC,KAA8D;IAClG,OAAO,KAAK,CAAC,QAAQ,KAAK,YAAY,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,QAAQ,CAAC;AAC7G,CAAC;AAED;;;GAGG;AACH,SAAS,qBAAqB,CAAC,cAA+B,EAAE,GAAiB;IAChF,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,SAAS,CAAC,KAAgC;IAClD,OAAO;QACN,qBAAqB,EAAE,KAAK,CAAC,MAAM,EAAE,qBAAqB,IAAI,IAAI;QAClE,8BAA8B,EAAE,KAAK,CAAC,MAAM,EAAE,8BAA8B,IAAI,KAAK;QACrF,qBAAqB,EAAE,KAAK,CAAC,MAAM,EAAE,qBAAqB,IAAI,2BAA2B,CAAC,KAAK,CAAC;QAChG,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,MAAM;QAC9D,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,IAAI;QAC5E,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,KAAK;QAC7D,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,KAAK;QAC7E,uBAAuB,EAAE,KAAK,CAAC,MAAM,EAAE,uBAAuB,IAAI,KAAK;QACvE,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,KAAK;QAC7D,+BAA+B,EAAE,KAAK,CAAC,MAAM,EAAE,+BAA+B,IAAI,KAAK;QACvF,uBAAuB,EAAE,KAAK,CAAC,MAAM,EAAE,uBAAuB,IAAI,IAAI;QACtE,sBAAsB,EAAE,KAAK,CAAC,MAAM,EAAE,sBAAsB,IAAI,KAAK;KACrE,CAAC;AACH,CAAC;AAED,iGAAiG;AACjG,SAAS,4BAA4B,CACpC,KAAgC,EAChC,OAA2C,EAC3C,MAAuC;IAEvC,IACC,CAAC,MAAM,CAAC,sBAAsB;QAC9B,CAAC,OAAO,EAAE,eAAe;QACzB,2DAA2D;QAC3D,eAAe,CAAC,KAAK,EAAE,OAAO,CAAC,MAAM,CAAC,EACrC,CAAC;QACF,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,OAAO,KAAK,CAAC,gBAAgB,EAAE,CAAC,OAAO,CAAC,eAAe,CAAC,IAAI,OAAO,CAAC,eAAe,CAAC;AACrF,CAAC;AAED,SAAS,uBAAuB,CAC/B,MAAuC,EACvC,cAA8B;IAE9B,OAAO,cAAc,KAAK,MAAM,IAAI,MAAM,CAAC,0BAA0B,IAAI,CAAC,MAAM,CAAC,+BAA+B;QAC/G,CAAC,CAAC,KAAK;QACP,CAAC,CAAC,SAAS,CAAC;AACd,CAAC;AAED,SAAS,qBAAqB,CAC7B,MAAuC,EACvC,cAA8B;IAE9B,IAAI,CAAC,MAAM,CAAC,+BAA+B;QAAE,OAAO,SAAS,CAAC;IAC9D,IAAI,cAAc,KAAK,MAAM;QAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC;IAC3D,IAAI,cAAc,KAAK,MAAM,IAAI,MAAM,CAAC,0BAA0B;QAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC;IAC1F,OAAO,SAAS,CAAC;AAClB,CAAC;AAUD;;GAEG;AACH,MAAM,CAAC,MAAM,MAAM,GAA+D,CACjF,KAAgC,EAChC,OAA0B,EAC1B,OAAgC,EACF,EAAE;IAChC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IACjD,MAAM,iBAAiB,GAAG,iBAAiB,CAAC,OAAO,EAAE,SAAS,CAAC,KAAK,CAAC,CAAC,8BAA8B,CAAC,CAAC;IAEtG,yBAAyB;IACzB,CAAC,KAAK,IAAI,EAAE;QACX,MAAM,MAAM,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;QAChC,MAAM,yBAAyB,GAAG,4BAA4B,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QACvF,MAAM,MAAM,GAAqB;YAChC,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,EAAE;YACX,GAAG,EAAE,KAAK,CAAC,GAAU;YACrB,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,KAAK,CAAC,EAAE;YACf,GAAG,CAAC,yBAAyB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,qBAAqB,EAAE,yBAAyB,EAAE,CAAC;YACxG,KAAK,EAAE;gBACN,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;gBACT,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;aACpE;YACD,UAAU,EAAE,SAAS;YACrB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,IAAI,CAAC;YACJ,uBAAuB;YACvB,MAAM,MAAM,GAAG,eAAe,CAAC,KAAK,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YAClF,MAAM,cAAc,GAAG,qBAAqB,CAAC,OAAO,EAAE,cAAc,EAAE,OAAO,EAAE,GAAG,CAAC,CAAC;YACpF,MAAM,cAAc,GAAG,cAAc,KAAK,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC;YAClF,MAAM,0BAA0B,GAAG,gCAAgC,CAClE,gBAAgB,CAAC,iBAAiB,CAAC,QAAQ,CAAC,EAC5C,MAAM,CAAC,0BAA0B,CACjC,CAAC;YACF,MAAM,MAAM,GAAG,YAAY,CAC1B,KAAK,EACL,iBAAiB,EACjB,MAAM,EACN,OAAO,EAAE,OAAO,EAChB,OAAO,EAAE,QAAQ,EACjB,OAAO,EAAE,KAAK,EACd,cAAc,CACd,CAAC;YACF,IAAI,MAAM,GAAG,WAAW,CACvB,KAAK,EACL,iBAAiB,EACjB,OAAO,EACP,MAAM,EACN,0BAA0B,EAC1B,yBAAyB,CACzB,CAAC;YACF,MAAM,UAAU,GAAG,MAAM,OAAO,EAAE,SAAS,EAAE,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YAC7D,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;gBAC9B,MAAM,GAAG,UAA2C,CAAC;YACtD,CAAC;YACD,MAAM,cAAc,GAAG;gBACtB,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACtD,GAAG,CAAC,OAAO,EAAE,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC3E,UAAU,EAAE,CAAC;aACb,CAAC;YACF,MAAM,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,GAAG,MAAM,oBAAoB,CAClE,GAAG,EAAE,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC,YAAY,EAAE,EACpE;gBACC,UAAU,EAAE,OAAO,EAAE,UAAU;gBAC/B,eAAe,EAAE,OAAO,EAAE,eAAe;gBACzC,MAAM,EAAE,OAAO,EAAE,MAAM;aACvB,CACD,CAAC;YACF,MAAM,OAAO,EAAE,UAAU,EAAE,CAAC,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,OAAO,EAAE,eAAe,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,EAAE,KAAK,CAAC,CAAC;YAC5G,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAEhD,MAAM,sBAAsB,CAAC,YAAY,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE;gBACjE,qBAAqB,EAAE,OAAO,EAAE,qBAAqB;gBACrD,WAAW,EAAE,OAAO,EAAE,WAAW;gBACjC,0BAA0B;gBAC1B,uBAAuB,EAAE,CAAC,KAAK,EAAE,WAAW,EAAE,EAAE,CAAC,uBAAuB,CAAC,KAAK,EAAE,WAAW,EAAE,KAAK,CAAC;aACnG,CAAC,CAAC;YAEH,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;gBACrC,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;YACxE,CAAC;YACD,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,YAAY,IAAI,2BAA2B,CAAC,CAAC;YACrE,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,MAAM,CAAC,UAAU,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;YACnE,MAAM,YAAY,GAAG,mBAAmB,CACvC,sBAAsB,CAAC,KAAK,CAAC,EAC7B,GAAG,KAAK,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,QAAQ,YAAY,CACtE,CAAC;YACF,8EAA8E;YAC9E,6EAA6E;YAC7E,IAAI,YAAY,CAAC,QAAQ,CAAC,2CAA2C,CAAC,EAAE,CAAC;gBACxE,MAAM,CAAC,YAAY,GAAG,GAAG,YAAY,+BAA+B,iBAAiB,EAAE,CAAC;YACzF,CAAC;iBAAM,IAAI,YAAY,CAAC,QAAQ,CAAC,gEAAgE,CAAC,EAAE,CAAC;gBACpG,MAAM,CAAC,YAAY,GAAG,GAAG,YAAY,KAAK,4BAA4B,EAAE,CAAC;YAC1E,CAAC;iBAAM,CAAC;gBACP,MAAM,CAAC,YAAY,GAAG,YAAY,CAAC;YACpC,CAAC;YACD,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;YACzE,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;IACF,CAAC,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AACf,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,YAAY,GAA4D,CACpF,KAAgC,EAChC,OAA0B,EAC1B,OAA6B,EACC,EAAE;IAChC,eAAe,CAAC,KAAK,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;IAEnE,MAAM,IAAI,GAAG;QACZ,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC;QAC7D,UAAU,EAAE,OAAO,EAAE,UAAU;KACE,CAAC;IACnC,MAAM,gBAAgB,GAAG,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,kBAAkB,CAAC,KAAK,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvG,MAAM,eAAe,GAAG,gBAAgB,KAAK,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,CAAC;IAElF,OAAO,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE;QAC7B,GAAG,IAAI;QACP,eAAe;KACkB,CAAC,CAAC;AACrC,CAAC,CAAC;AAEF,SAAS,YAAY,CACpB,KAAgC,EAChC,OAA0B,EAC1B,MAAc,EACd,cAAgC,EAChC,QAAyB,EACzB,KAA+B,EAC/B,SAAkB;IAElB,MAAM,MAAM,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;IAChC,MAAM,OAAO,GAAoB,EAAE,GAAG,kBAAkB,CAAC,KAAK,EAAE,QAAQ,CAAC,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC;IAC9F,IAAI,KAAK,CAAC,QAAQ,KAAK,gBAAgB,EAAE,CAAC;QACzC,MAAM,SAAS,GAAG,qBAAqB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;QAC1D,MAAM,cAAc,GAAG,0BAA0B,CAAC;YACjD,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,SAAS;SACT,CAAC,CAAC;QACH,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;IACxC,CAAC;IAED,IAAI,SAAS,EAAE,CAAC;QACf,IAAI,MAAM,CAAC,qBAAqB,KAAK,YAAY,EAAE,CAAC;YACnD,OAAO,CAAC,cAAc,CAAC,GAAG,SAAS,CAAC;QACrC,CAAC;aAAM,CAAC;YACP,IAAI,MAAM,CAAC,qBAAqB,KAAK,QAAQ,EAAE,CAAC;gBAC/C,OAAO,CAAC,YAAY,CAAC,GAAG,SAAS,CAAC;YACnC,CAAC;YACD,OAAO,CAAC,qBAAqB,CAAC,GAAG,SAAS,CAAC;QAC5C,CAAC;IACF,CAAC;IAED,2DAA2D;IAC3D,IAAI,cAAc,EAAE,CAAC;QACpB,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;IACxC,CAAC;IAED,OAAO,IAAI,MAAM,CAAC;QACjB,MAAM;QACN,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,uBAAuB,EAAE,IAAI;QAC7B,KAAK;QACL,cAAc,EAAE,OAAO;KACvB,CAAC,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CACnB,KAAgC,EAChC,OAA0B,EAC1B,OAA2C,EAC3C,MAAuC,EACvC,0BAAuD,EACvD,yBAA6C;IAE7C,MAAM,eAAe,GAAG,sBAAsB,CAAC,OAAO,CAAC,QAAQ,EAAE,uBAAuB,CAAC,MAAM,CAAC,CAAC,CAAC;IAClG,MAAM,QAAQ,GAAG,wBAAwB,CAAC,KAAK,EAAE,OAAO,EAAE,0BAA0B,EAAE;QACrF,0BAA0B;QAC1B,8BAA8B,EAAE,MAAM,CAAC,8BAA8B;QACrE,uBAAuB,EAAE,MAAM,CAAC,uBAAuB;QACvD,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;QAC7C,WAAW,EAAE;YACZ,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;YAC7C,0BAA0B,EAAE,MAAM,CAAC,0BAA0B;SAC7D;QACD,yBAAyB;KACzB,CAAC,CAAC;IAEH,MAAM,cAAc,GAAG,qBAAqB,CAAC,OAAO,EAAE,cAAc,EAAE,OAAO,EAAE,GAAG,CAAC,CAAC;IACpF,MAAM,kBAAkB,GAAG,cAAc,KAAK,MAAM,IAAI,MAAM,CAAC,kBAAkB,KAAK,UAAU,CAAC;IACjG,qDAAqD;IACrD,MAAM,qBAAqB,GAAG,eAAe,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACtE,MAAM,MAAM,GAAkC;QAC7C,KAAK,EAAE,KAAK,CAAC,EAAE;QACf,KAAK,EAAE,QAAQ;QACf,MAAM,EAAE,IAAI;QACZ,gBAAgB,EAAE,kBAAkB,CAAC,CAAC,CAAC,yBAAyB,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS;QAChG,sBAAsB,EAAE,qBAAqB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,uBAAuB,CAAC,MAAM,EAAE,cAAc,CAAC;QAC3G,oBAAoB,EAAE,qBAAqB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,qBAAqB,CAAC,MAAM,EAAE,cAAc,CAAC;QACvG,KAAK,EAAE,KAAK;KACZ,CAAC;IAEF,IAAI,OAAO,EAAE,SAAS,IAAI,MAAM,CAAC,uBAAuB,IAAI,CAAC,qBAAqB,EAAE,CAAC;QACpF,MAAM,CAAC,iBAAiB,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,SAAS,EAAE,kCAAkC,CAAC,CAAC;IAC5F,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,IAAI,CAAC,qBAAqB,EAAE,CAAC;QAClE,MAAM,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,YAAY,GAAG,OAAO,CAAC,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,eAAe,CAAC,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC7C,MAAM,CAAC,KAAK,GAAG,qBAAqB,CAAC,eAAe,CAAC,YAAY,EAAE;YAClE,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;YAC7C,0BAA0B,EAAE,MAAM,CAAC,0BAA0B;SAC7D,CAAC,CAAC;IACJ,CAAC;IAED,IAAI,OAAO,EAAE,UAAU,KAAK,SAAS,EAAE,CAAC;QACvC,MAAM,CAAC,WAAW,GAAG,OAAO,CAAC,UAAU,CAAC;IACzC,CAAC;IAED,MAAM,eAAe,GAAG,OAAO,EAAE,eAAe,IAAI,CAAC,OAAO,EAAE,gBAAgB,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;IACvG,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QACrB,IAAI,eAAe,EAAE,CAAC;YACrB,MAAM,eAAe,GAAG,OAAO,EAAE,eAAe;gBAC/C,CAAC,CAAC,CAAC,KAAK,CAAC,gBAAgB,EAAE,CAAC,OAAO,CAAC,eAAe,CAAC,IAAI,OAAO,CAAC,eAAe,CAAC;gBAChF,CAAC,CAAC,eAAe,CAAC;YACnB,MAAM,MAAM,GAAG,yBAAyB,KAAK,SAAS,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,mCAAmC,CAAC;YAC/G,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,MAAwD;gBAChE,OAAO,EAAE,OAAO,EAAE,gBAAgB,IAAI,MAAM;aAC5C,CAAC;YACF,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAClD,CAAC;aAAM,IAAI,KAAK,CAAC,QAAQ,KAAK,gBAAgB,IAAI,KAAK,CAAC,gBAAgB,EAAE,GAAG,KAAK,IAAI,EAAE,CAAC;YACxF,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,CAAC,KAAK,CAAC,gBAAgB,EAAE,GAAG,IAAI,MAAM,CAAmD;aACjG,CAAC;QACH,CAAC;QACD,IAAI,KAAK,CAAC,QAAQ,KAAK,KAAK;YAAE,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;IAChF,CAAC;IAED,+EAA+E;IAC/E,MAAM,cAAc,GAAG,qBAAqB,CAAC,KAAK,EAAE,eAAe,IAAI,KAAK,EAAE,OAAO,EAAE,cAAc,CAAC,CAAC;IACvG,IAAI,cAAc,EAAE,CAAC;QACpB,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IACvC,CAAC;IAED,OAAO,MAAM,CAAC;AACf,CAAC;AAED,SAAS,4BAA4B,CACpC,KAA4C,EAC5C,WAAsE;IAEtE,QAAQ,WAAW,EAAE,CAAC;QACrB,KAAK,MAAM;YACV,OAAO,GAAG,CAAC;QACZ,KAAK,UAAU,CAAC;QAChB,KAAK,MAAM;YACV,OAAO,KAAK,CAAC,EAAE,KAAK,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QACzC;YACC,OAAO,CAAC,CAAC;IACX,CAAC;AACF,CAAC;AAED,SAAS,uBAAuB,CAC/B,KAAY,EACZ,WAAsE,EACtE,KAA4C;IAE5C,MAAM,UAAU,GAAG,4BAA4B,CAAC,KAAK,EAAE,WAAW,CAAC,CAAC;IACpE,IAAI,UAAU,KAAK,CAAC;QAAE,OAAO;IAE7B,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,UAAU,CAAC;IAC/B,KAAK,CAAC,IAAI,CAAC,MAAM,IAAI,UAAU,CAAC;IAChC,KAAK,CAAC,IAAI,CAAC,SAAS,IAAI,UAAU,CAAC;IACnC,KAAK,CAAC,IAAI,CAAC,UAAU,IAAI,UAAU,CAAC;IACpC,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC;AACxG,CAAC","sourcesContent":["import OpenAI from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { clampThinkingLevel } from \"../models.ts\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tModel,\n\tOpenAIResponsesCompat,\n\tProviderEnv,\n\tProviderHeaders,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tTranscriptContext,\n\tUsage,\n} from \"../types.ts\";\nimport { formatProviderError, normalizeProviderError } from \"../utils/error-body.ts\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { headersToRecord } from \"../utils/headers.ts\";\nimport { type ClientIdentity, getIdentityHeaders } from \"../utils/identity.ts\";\nimport { getProviderEnvValue } from \"../utils/provider-env.ts\";\nimport { retryProviderRequest } from \"../utils/provider-retry.ts\";\nimport { getDeclaredTools, resolveTranscript, resolveTranscriptTools } from \"../utils/transcript.ts\";\nimport { createGrammarToolInputProperties } from \"./constrained-sampling.ts\";\nimport { buildCopilotDynamicHeaders, hasCopilotVisionInput } from \"./github-copilot-headers.ts\";\nimport { clampOpenAIPromptCacheKey } from \"./openai-prompt-cache.ts\";\nimport {\n\tconvertResponsesMessages,\n\tconvertResponsesTools,\n\tgetResponsesToolSupport,\n\tprocessResponsesStream,\n} from \"./openai-responses-shared.ts\";\nimport { buildBaseOptions, resolveSamplingParams } from \"./simple-options.ts\";\n\nconst OPENAI_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\"]);\n// OpenAI Responses rejects max_output_tokens below 16: https://github.com/earendil-works/pi/issues/6265\nconst OPENAI_RESPONSES_MIN_OUTPUT_TOKENS = 16;\nconst CHATGPT_USAGE_URL = \"https://chatgpt.com/settings/usage\";\nconst CHATGPT_WORKSPACE_LIMIT_HINT =\n\t\"You've reached your ChatGPT workspace's usage limit. Ask your workspace admin to increase it at admin.openai.com → Workspace → Usage limits.\";\n// Fixed reasoning.effort while configuration_update items carry the thinking level.\nconst CONFIGURATION_UPDATE_REQUEST_EFFORT = \"medium\";\n\n/**\n * OpenAI API keys start with `sk-`; a different credential sent directly to OpenAI\n * is a Sign in with ChatGPT access token.\n */\nfunction isChatGPTSignIn(model: Model<\"openai-responses\">, apiKey: string | undefined): boolean {\n\treturn (\n\t\tmodel.provider === \"openai\" &&\n\t\tmodel.baseUrl === \"https://api.openai.com/v1\" &&\n\t\tapiKey !== undefined &&\n\t\t!apiKey.startsWith(\"sk-\")\n\t);\n}\n\nfunction hasHeader(headers: ProviderHeaders | undefined, name: string): boolean {\n\tif (!headers) return false;\n\tconst expected = name.toLowerCase();\n\tfor (const [key, value] of Object.entries(headers)) {\n\t\tif (key.toLowerCase() === expected && value !== null && value.trim().length > 0) return true;\n\t}\n\treturn false;\n}\n\nfunction getClientApiKey(provider: string, apiKey: string | undefined, headers: ProviderHeaders | undefined): string {\n\tif (apiKey) return apiKey;\n\tif (hasHeader(headers, \"authorization\") || hasHeader(headers, \"cf-aig-authorization\")) return \"unused\";\n\tthrow new Error(`No API key for provider: ${provider}`);\n}\n\nfunction detectSessionAffinityFormat(model: Pick<Model<\"openai-responses\">, \"provider\" | \"baseUrl\">) {\n\treturn model.provider === \"openrouter\" || model.baseUrl.includes(\"openrouter.ai\") ? \"openrouter\" : \"openai\";\n}\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses PI_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction getCompat(model: Model<\"openai-responses\">): Required<OpenAIResponsesCompat> {\n\treturn {\n\t\tsupportsDeveloperRole: model.compat?.supportsDeveloperRole ?? true,\n\t\tsupportsMidConvoSystemMessages: model.compat?.supportsMidConvoSystemMessages ?? false,\n\t\tsessionAffinityFormat: model.compat?.sessionAffinityFormat ?? detectSessionAffinityFormat(model),\n\t\tpromptCacheKeyMode: model.compat?.promptCacheKeyMode ?? \"auto\",\n\t\tsupportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? true,\n\t\tsupportsStrictMode: model.compat?.supportsStrictMode ?? false,\n\t\tsupportsOpenAIGrammarTools: model.compat?.supportsOpenAIGrammarTools ?? false,\n\t\tsupportsAdditionalTools: model.compat?.supportsAdditionalTools ?? false,\n\t\tsupportsToolSearch: model.compat?.supportsToolSearch ?? false,\n\t\tsupportsExplicitPromptCacheMode: model.compat?.supportsExplicitPromptCacheMode ?? false,\n\t\tsupportsMaxOutputTokens: model.compat?.supportsMaxOutputTokens ?? true,\n\t\tsupportsMidConvoEffort: model.compat?.supportsMidConvoEffort ?? false,\n\t};\n}\n\n/** Effort to send in configuration_update items, or undefined to send it as reasoning.effort. */\nfunction getConfigurationUpdateEffort(\n\tmodel: Model<\"openai-responses\">,\n\toptions: OpenAIResponsesOptions | undefined,\n\tcompat: Required<OpenAIResponsesCompat>,\n): string | undefined {\n\tif (\n\t\t!compat.supportsMidConvoEffort ||\n\t\t!options?.reasoningEffort ||\n\t\t// Sign in with ChatGPT rejects configuration_update items.\n\t\tisChatGPTSignIn(model, options.apiKey)\n\t) {\n\t\treturn undefined;\n\t}\n\treturn model.thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort;\n}\n\nfunction getPromptCacheRetention(\n\tcompat: Required<OpenAIResponsesCompat>,\n\tcacheRetention: CacheRetention,\n): \"24h\" | undefined {\n\treturn cacheRetention === \"long\" && compat.supportsLongCacheRetention && !compat.supportsExplicitPromptCacheMode\n\t\t? \"24h\"\n\t\t: undefined;\n}\n\nfunction getPromptCacheOptions(\n\tcompat: Required<OpenAIResponsesCompat>,\n\tcacheRetention: CacheRetention,\n): ResponseCreateParamsStreaming[\"prompt_cache_options\"] {\n\tif (!compat.supportsExplicitPromptCacheMode) return undefined;\n\tif (cacheRetention === \"none\") return { mode: \"explicit\" };\n\tif (cacheRetention === \"long\" && compat.supportsLongCacheRetention) return { ttl: \"30m\" };\n\treturn undefined;\n}\n\n// OpenAI Responses-specific options\nexport interface OpenAIResponsesOptions extends StreamOptions {\n\treasoningEffort?: \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\" | \"max\";\n\treasoningSummary?: \"auto\" | \"detailed\" | \"concise\" | null;\n\tserviceTier?: ResponseCreateParamsStreaming[\"service_tier\"];\n\ttoolChoice?: ResponseCreateParamsStreaming[\"tool_choice\"];\n}\n\n/**\n * Generate function for OpenAI Responses API\n */\nexport const stream: StreamFunction<\"openai-responses\", OpenAIResponsesOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: TranscriptContext,\n\toptions?: OpenAIResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\tconst normalizedContext = resolveTranscript(context, getCompat(model).supportsMidConvoSystemMessages);\n\n\t// Start async processing\n\t(async () => {\n\t\tconst compat = getCompat(model);\n\t\tconst configurationUpdateEffort = getConfigurationUpdateEffort(model, options, compat);\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: model.api as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\t...(configurationUpdateEffort === undefined ? {} : { providerThinkingLevel: configurationUpdateEffort }),\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"pending\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// Create OpenAI client\n\t\t\tconst apiKey = getClientApiKey(model.provider, options?.apiKey, options?.headers);\n\t\t\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);\n\t\t\tconst cacheSessionId = cacheRetention === \"none\" ? undefined : options?.sessionId;\n\t\t\tconst grammarToolInputProperties = createGrammarToolInputProperties(\n\t\t\t\tgetDeclaredTools(normalizedContext.messages),\n\t\t\t\tcompat.supportsOpenAIGrammarTools,\n\t\t\t);\n\t\t\tconst client = createClient(\n\t\t\t\tmodel,\n\t\t\t\tnormalizedContext,\n\t\t\t\tapiKey,\n\t\t\t\toptions?.headers,\n\t\t\t\toptions?.identity,\n\t\t\t\toptions?.fetch,\n\t\t\t\tcacheSessionId,\n\t\t\t);\n\t\t\tlet params = buildParams(\n\t\t\t\tmodel,\n\t\t\t\tnormalizedContext,\n\t\t\t\toptions,\n\t\t\t\tcompat,\n\t\t\t\tgrammarToolInputProperties,\n\t\t\t\tconfigurationUpdateEffort,\n\t\t\t);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as ResponseCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst requestOptions = {\n\t\t\t\t...(options?.signal ? { signal: options.signal } : {}),\n\t\t\t\t...(options?.timeoutMs !== undefined ? { timeout: options.timeoutMs } : {}),\n\t\t\t\tmaxRetries: 0,\n\t\t\t};\n\t\t\tconst { data: openaiStream, response } = await retryProviderRequest(\n\t\t\t\t() => client.responses.create(params, requestOptions).withResponse(),\n\t\t\t\t{\n\t\t\t\t\tmaxRetries: options?.maxRetries,\n\t\t\t\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\t\t\t\tsignal: options?.signal,\n\t\t\t\t},\n\t\t\t);\n\t\t\tawait options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\tawait processResponsesStream(openaiStream, output, stream, model, {\n\t\t\t\tonProviderStreamEvent: options?.onProviderStreamEvent,\n\t\t\t\tserviceTier: options?.serviceTier,\n\t\t\t\tgrammarToolInputProperties,\n\t\t\t\tapplyServiceTierPricing: (usage, serviceTier) => applyServiceTierPricing(usage, serviceTier, model),\n\t\t\t});\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"pending\") {\n\t\t\t\tthrow new Error(\"OpenAI Responses stream ended without a stop reason\");\n\t\t\t}\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(output.errorMessage || \"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\tconst errorMessage = formatProviderError(\n\t\t\t\tnormalizeProviderError(error),\n\t\t\t\t`${model.provider === \"openai\" ? \"OpenAI\" : model.provider} API error`,\n\t\t\t);\n\t\t\t// Sign in with ChatGPT shares the subscription's usage limit with other apps.\n\t\t\t// On ChatGPT Business plans, the workspace admin also sets a per-user limit.\n\t\t\tif (errorMessage.includes(\"subscription_sharing_usage_limit_exceeded\")) {\n\t\t\t\toutput.errorMessage = `${errorMessage}\\nCheck your ChatGPT usage: ${CHATGPT_USAGE_URL}`;\n\t\t\t} else if (errorMessage.includes(\"subscription_sharing_workplace_individual_usage_limit_exceeded\")) {\n\t\t\t\toutput.errorMessage = `${errorMessage}\\n${CHATGPT_WORKSPACE_LIMIT_HINT}`;\n\t\t\t} else {\n\t\t\t\toutput.errorMessage = errorMessage;\n\t\t\t}\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimple: StreamFunction<\"openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: TranscriptContext,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tgetClientApiKey(model.provider, options?.apiKey, options?.headers);\n\n\tconst base = {\n\t\t...buildBaseOptions(model, context, options, options?.apiKey),\n\t\ttoolChoice: options?.toolChoice,\n\t} satisfies OpenAIResponsesOptions;\n\tconst clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : undefined;\n\tconst reasoningEffort = clampedReasoning === \"off\" ? undefined : clampedReasoning;\n\n\treturn stream(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies OpenAIResponsesOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: TranscriptContext,\n\tapiKey: string,\n\toptionsHeaders?: ProviderHeaders,\n\tidentity?: ClientIdentity,\n\tfetch?: typeof globalThis.fetch,\n\tsessionId?: string,\n) {\n\tconst compat = getCompat(model);\n\tconst headers: ProviderHeaders = { ...getIdentityHeaders(model, identity), ...model.headers };\n\tif (model.provider === \"github-copilot\") {\n\t\tconst hasImages = hasCopilotVisionInput(context.messages);\n\t\tconst copilotHeaders = buildCopilotDynamicHeaders({\n\t\t\tmessages: context.messages,\n\t\t\thasImages,\n\t\t});\n\t\tObject.assign(headers, copilotHeaders);\n\t}\n\n\tif (sessionId) {\n\t\tif (compat.sessionAffinityFormat === \"openrouter\") {\n\t\t\theaders[\"x-session-id\"] = sessionId;\n\t\t} else {\n\t\t\tif (compat.sessionAffinityFormat === \"openai\") {\n\t\t\t\theaders[\"session-id\"] = sessionId;\n\t\t\t}\n\t\t\theaders[\"x-client-request-id\"] = sessionId;\n\t\t}\n\t}\n\n\t// Merge options headers last so they can override defaults\n\tif (optionsHeaders) {\n\t\tObject.assign(headers, optionsHeaders);\n\t}\n\n\treturn new OpenAI({\n\t\tapiKey,\n\t\tbaseURL: model.baseUrl,\n\t\tdangerouslyAllowBrowser: true,\n\t\tfetch,\n\t\tdefaultHeaders: headers,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: TranscriptContext,\n\toptions: OpenAIResponsesOptions | undefined,\n\tcompat: Required<OpenAIResponsesCompat>,\n\tgrammarToolInputProperties: ReadonlyMap<string, string>,\n\tconfigurationUpdateEffort: string | undefined,\n) {\n\tconst transcriptTools = resolveTranscriptTools(context.messages, getResponsesToolSupport(compat));\n\tconst messages = convertResponsesMessages(model, context, OPENAI_TOOL_CALL_PROVIDERS, {\n\t\tgrammarToolInputProperties,\n\t\tsupportsMidConvoSystemMessages: compat.supportsMidConvoSystemMessages,\n\t\tsupportsAdditionalTools: compat.supportsAdditionalTools,\n\t\tsupportsToolSearch: compat.supportsToolSearch,\n\t\ttoolOptions: {\n\t\t\tsupportsStrictMode: compat.supportsStrictMode,\n\t\t\tsupportsOpenAIGrammarTools: compat.supportsOpenAIGrammarTools,\n\t\t},\n\t\tconfigurationUpdateEffort,\n\t});\n\n\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);\n\tconst sendPromptCacheKey = cacheRetention !== \"none\" && compat.promptCacheKeyMode !== \"disabled\";\n\t// Sign in with ChatGPT rejects these request fields.\n\tconst omitUnsupportedFields = isChatGPTSignIn(model, options?.apiKey);\n\tconst params: ResponseCreateParamsStreaming = {\n\t\tmodel: model.id,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: sendPromptCacheKey ? clampOpenAIPromptCacheKey(options?.sessionId) : undefined,\n\t\tprompt_cache_retention: omitUnsupportedFields ? undefined : getPromptCacheRetention(compat, cacheRetention),\n\t\tprompt_cache_options: omitUnsupportedFields ? undefined : getPromptCacheOptions(compat, cacheRetention),\n\t\tstore: false,\n\t};\n\n\tif (options?.maxTokens && compat.supportsMaxOutputTokens && !omitUnsupportedFields) {\n\t\tparams.max_output_tokens = Math.max(options.maxTokens, OPENAI_RESPONSES_MIN_OUTPUT_TOKENS);\n\t}\n\n\tif (options?.temperature !== undefined && !omitUnsupportedFields) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (options?.serviceTier !== undefined) {\n\t\tparams.service_tier = options.serviceTier;\n\t}\n\n\tif (transcriptTools.requestTools.length > 0) {\n\t\tparams.tools = convertResponsesTools(transcriptTools.requestTools, {\n\t\t\tsupportsStrictMode: compat.supportsStrictMode,\n\t\t\tsupportsOpenAIGrammarTools: compat.supportsOpenAIGrammarTools,\n\t\t});\n\t}\n\n\tif (options?.toolChoice !== undefined) {\n\t\tparams.tool_choice = options.toolChoice;\n\t}\n\n\tconst reasoningEffort = options?.reasoningEffort ?? (options?.reasoningSummary ? \"medium\" : undefined);\n\tif (model.reasoning) {\n\t\tif (reasoningEffort) {\n\t\t\tconst requestedEffort = options?.reasoningEffort\n\t\t\t\t? (model.thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort)\n\t\t\t\t: reasoningEffort;\n\t\t\tconst effort = configurationUpdateEffort === undefined ? requestedEffort : CONFIGURATION_UPDATE_REQUEST_EFFORT;\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: effort as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t\tsummary: options?.reasoningSummary || \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else if (model.provider !== \"github-copilot\" && model.thinkingLevelMap?.off !== null) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: (model.thinkingLevelMap?.off ?? \"none\") as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t};\n\t\t}\n\t\tif (model.provider === \"xai\") params.include = [\"reasoning.encrypted_content\"];\n\t}\n\n\t// Last so model and request sampling parameters override named request fields.\n\tconst samplingParams = resolveSamplingParams(model, reasoningEffort ?? \"off\", options?.samplingParams);\n\tif (samplingParams) {\n\t\tObject.assign(params, samplingParams);\n\t}\n\n\treturn params;\n}\n\nfunction getServiceTierCostMultiplier(\n\tmodel: Pick<Model<\"openai-responses\">, \"id\">,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n): number {\n\tswitch (serviceTier) {\n\t\tcase \"flex\":\n\t\t\treturn 0.5;\n\t\tcase \"priority\":\n\t\tcase \"fast\":\n\t\t\treturn model.id === \"gpt-5.5\" ? 2.5 : 2;\n\t\tdefault:\n\t\t\treturn 1;\n\t}\n}\n\nfunction applyServiceTierPricing(\n\tusage: Usage,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\tmodel: Pick<Model<\"openai-responses\">, \"id\">,\n) {\n\tconst multiplier = getServiceTierCostMultiplier(model, serviceTier);\n\tif (multiplier === 1) return;\n\n\tusage.cost.input *= multiplier;\n\tusage.cost.output *= multiplier;\n\tusage.cost.cacheRead *= multiplier;\n\tusage.cost.cacheWrite *= multiplier;\n\tusage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;\n}\n"]}

package/dist/api/simple-options.d.ts

@@ -1,4 +1,5 @@
 import type { Api, Model, ModelThinkingLevel, SamplingParams, SimpleStreamOptions, StreamOptions, ThinkingBudgets, ThinkingLevel, TranscriptContext } from "../types.ts";
+/** Fit maxTokens into the context window, leaving room for the input being counted higher than estimated. */
 export declare function clampMaxTokensToContext(model: Model<Api>, context: TranscriptContext, maxTokens: number): number;
 export declare function resolveSamplingParams(model: Model<Api>, thinkingLevel: ModelThinkingLevel, requestParams?: SamplingParams): SamplingParams | undefined;
 export declare function buildBaseOptions(model: Model<Api>, context: TranscriptContext, options?: SimpleStreamOptions, apiKey?: string): StreamOptions;

package/dist/api/simple-options.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"simple-options.d.ts","sourceRoot":"","sources":["../../src/api/simple-options.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACX,GAAG,EACH,KAAK,EACL,kBAAkB,EAClB,cAAc,EACd,mBAAmB,EACnB,aAAa,EACb,eAAe,EACf,aAAa,EACb,iBAAiB,EACjB,MAAM,aAAa,CAAC;AAMrB,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,OAAO,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,CAIhH;AAED,wBAAgB,qBAAqB,CACpC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,aAAa,EAAE,kBAAkB,EACjC,aAAa,CAAC,EAAE,cAAc,GAC5B,cAAc,GAAG,SAAS,CAM5B;AAED,wBAAgB,gBAAgB,CAC/B,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,OAAO,EAAE,iBAAiB,EAC1B,OAAO,CAAC,EAAE,mBAAmB,EAC7B,MAAM,CAAC,EAAE,MAAM,GACb,aAAa,CAyBf;AAED,4FAA4F;AAC5F,eAAO,MAAM,iBAAiB,OAAO,CAAC;AAEtC,eAAO,MAAM,wBAAwB,EAAE,eAKtC,CAAC;AAEF,wBAAgB,cAAc,CAAC,MAAM,EAAE,aAAa,GAAG,SAAS,GAAG,OAAO,CAAC,aAAa,EAAE,OAAO,GAAG,KAAK,CAAC,GAAG,SAAS,CAErH;AAED,wBAAgB,sBAAsB,CAAC,cAAc,EAAE,aAAa,EAAE,aAAa,CAAC,EAAE,eAAe,GAAG,MAAM,CAI7G;AAED,kGAAkG;AAClG,wBAAgB,+BAA+B,CAAC,cAAc,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,MAAM,CAE/F;AAED,wBAAgB,0BAA0B,CAEzC,aAAa,EAAE,MAAM,GAAG,SAAS,EACjC,cAAc,EAAE,MAAM,EACtB,cAAc,EAAE,aAAa,EAC7B,aAAa,CAAC,EAAE,eAAe,GAC7B;IAAE,SAAS,EAAE,MAAM,CAAC;IAAC,cAAc,EAAE,MAAM,CAAA;CAAE,CAU/C"}
+{"version":3,"file":"simple-options.d.ts","sourceRoot":"","sources":["../../src/api/simple-options.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACX,GAAG,EACH,KAAK,EACL,kBAAkB,EAClB,cAAc,EACd,mBAAmB,EACnB,aAAa,EACb,eAAe,EACf,aAAa,EACb,iBAAiB,EACjB,MAAM,aAAa,CAAC;AAarB,6GAA6G;AAC7G,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,OAAO,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,CAShH;AAED,wBAAgB,qBAAqB,CACpC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,aAAa,EAAE,kBAAkB,EACjC,aAAa,CAAC,EAAE,cAAc,GAC5B,cAAc,GAAG,SAAS,CAM5B;AAED,wBAAgB,gBAAgB,CAC/B,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,OAAO,EAAE,iBAAiB,EAC1B,OAAO,CAAC,EAAE,mBAAmB,EAC7B,MAAM,CAAC,EAAE,MAAM,GACb,aAAa,CA0Bf;AAED,4FAA4F;AAC5F,eAAO,MAAM,iBAAiB,OAAO,CAAC;AAEtC,eAAO,MAAM,wBAAwB,EAAE,eAKtC,CAAC;AAEF,wBAAgB,cAAc,CAAC,MAAM,EAAE,aAAa,GAAG,SAAS,GAAG,OAAO,CAAC,aAAa,EAAE,OAAO,GAAG,KAAK,CAAC,GAAG,SAAS,CAErH;AAED,wBAAgB,sBAAsB,CAAC,cAAc,EAAE,aAAa,EAAE,aAAa,CAAC,EAAE,eAAe,GAAG,MAAM,CAI7G;AAED,kGAAkG;AAClG,wBAAgB,+BAA+B,CAAC,cAAc,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,MAAM,CAE/F;AAED,wBAAgB,0BAA0B,CAEzC,aAAa,EAAE,MAAM,GAAG,SAAS,EACjC,cAAc,EAAE,MAAM,EACtB,cAAc,EAAE,aAAa,EAC7B,aAAa,CAAC,EAAE,eAAe,GAC7B;IAAE,SAAS,EAAE,MAAM,CAAC;IAAC,cAAc,EAAE,MAAM,CAAA;CAAE,CAU/C"}

package/dist/api/simple-options.js

@@ -1,12 +1,25 @@
 import { clampThinkingLevel } from "../models.js";
 import { estimateContextTokens } from "../utils/estimate.js";
 const CONTEXT_SAFETY_TOKENS = 4096;
+// OpenRouter counts the input with its own tokenizer and rejects input + max_tokens above the
+// window before routing. It counted prose up to 12% above the provider's count (#10497).
+const CONTEXT_SAFETY_RATIO = 0.15;
+// The 15% margin is kept only while it leaves at least this much output. Otherwise (a nearly full
+// large window, or any small window such as a 32k local model) only the fixed margin is kept, and
+// output is capped at this size so an OpenRouter overcount still fits for a while before compaction.
+const MIN_OUTPUT_WITH_RATIO_MARGIN = 32768;
 const MIN_MAX_TOKENS = 1;
+/** Fit maxTokens into the context window, leaving room for the input being counted higher than estimated. */
 export function clampMaxTokensToContext(model, context, maxTokens) {
     if (model.contextWindow <= 0)
         return Math.max(MIN_MAX_TOKENS, maxTokens);
-    const available = model.contextWindow - estimateContextTokens(context).tokens - CONTEXT_SAFETY_TOKENS;
-    return Math.min(maxTokens, Math.max(MIN_MAX_TOKENS, available));
+    const input = estimateContextTokens(context).tokens;
+    const margin = Math.max(CONTEXT_SAFETY_TOKENS, Math.ceil(input * CONTEXT_SAFETY_RATIO));
+    let room = model.contextWindow - input - margin;
+    if (room < MIN_OUTPUT_WITH_RATIO_MARGIN) {
+        room = Math.min(MIN_OUTPUT_WITH_RATIO_MARGIN, model.contextWindow - input - CONTEXT_SAFETY_TOKENS);
+    }
+    return Math.min(maxTokens, Math.max(MIN_MAX_TOKENS, room));
 }
 export function resolveSamplingParams(model, thinkingLevel, requestParams) {
     const effectiveThinkingLevel = clampThinkingLevel(model, thinkingLevel);
@@ -28,6 +41,7 @@ export function buildBaseOptions(model, context, options, apiKey) {
         transport: options?.transport,
         cacheRetention: options?.cacheRetention,
         sessionId: options?.sessionId,
+        disableContinuation: options?.disableContinuation,
         headers: options?.headers,
         identity: options?.identity,
         onPayload: options?.onPayload,

package/dist/api/simple-options.js.map

@@ -1 +1 @@
-{"version":3,"file":"simple-options.js","sourceRoot":"","sources":["../../src/api/simple-options.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAYlD,OAAO,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAE7D,MAAM,qBAAqB,GAAG,IAAI,CAAC;AACnC,MAAM,cAAc,GAAG,CAAC,CAAC;AAEzB,MAAM,UAAU,uBAAuB,CAAC,KAAiB,EAAE,OAA0B,EAAE,SAAiB;IACvG,IAAI,KAAK,CAAC,aAAa,IAAI,CAAC;QAAE,OAAO,IAAI,CAAC,GAAG,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;IACzE,MAAM,SAAS,GAAG,KAAK,CAAC,aAAa,GAAG,qBAAqB,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,qBAAqB,CAAC;IACtG,OAAO,IAAI,CAAC,GAAG,CAAC,SAAS,EAAE,IAAI,CAAC,GAAG,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC,CAAC;AACjE,CAAC;AAED,MAAM,UAAU,qBAAqB,CACpC,KAAiB,EACjB,aAAiC,EACjC,aAA8B;IAE9B,MAAM,sBAAsB,GAAG,kBAAkB,CAAC,KAAK,EAAE,aAAa,CAAC,CAAC;IACxE,MAAM,mBAAmB,GAAG,KAAK,CAAC,6BAA6B,EAAE,CAAC,sBAAsB,CAAC,CAAC;IAC1F,OAAO,KAAK,CAAC,cAAc,IAAI,mBAAmB,IAAI,aAAa;QAClE,CAAC,CAAC,EAAE,GAAG,KAAK,CAAC,cAAc,EAAE,GAAG,mBAAmB,EAAE,GAAG,aAAa,EAAE;QACvE,CAAC,CAAC,SAAS,CAAC;AACd,CAAC;AAED,MAAM,UAAU,gBAAgB,CAC/B,KAAiB,EACjB,OAA0B,EAC1B,OAA6B,EAC7B,MAAe;IAEf,MAAM,cAAc,GAAG,qBAAqB,CAAC,KAAK,EAAE,OAAO,EAAE,SAAS,IAAI,KAAK,EAAE,OAAO,EAAE,cAAc,CAAC,CAAC;IAC1G,OAAO;QACN,WAAW,EAAE,OAAO,EAAE,WAAW;QACjC,cAAc;QACd,SAAS,EAAE,uBAAuB,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,KAAK,CAAC,SAAS,CAAC;QACzF,MAAM,EAAE,OAAO,EAAE,MAAM;QACvB,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,MAAM,EAAE,MAAM,IAAI,OAAO,EAAE,MAAM;QACjC,KAAK,EAAE,OAAO,EAAE,KAAK;QACrB,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,cAAc,EAAE,OAAO,EAAE,cAAc;QACvC,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,OAAO,EAAE,OAAO,EAAE,OAAO;QACzB,QAAQ,EAAE,OAAO,EAAE,QAAQ;QAC3B,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,UAAU,EAAE,OAAO,EAAE,UAAU;QAC/B,qBAAqB,EAAE,OAAO,EAAE,qBAAqB;QACrD,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,yBAAyB,EAAE,OAAO,EAAE,yBAAyB;QAC7D,UAAU,EAAE,OAAO,EAAE,UAAU;QAC/B,eAAe,EAAE,OAAO,EAAE,eAAe;QACzC,QAAQ,EAAE,OAAO,EAAE,QAAQ;QAC3B,GAAG,EAAE,OAAO,EAAE,GAAG;KACjB,CAAC;AACH,CAAC;AAED,4FAA4F;AAC5F,MAAM,CAAC,MAAM,iBAAiB,GAAG,IAAI,CAAC;AAEtC,MAAM,CAAC,MAAM,wBAAwB,GAAoB;IACxD,OAAO,EAAE,IAAI;IACb,GAAG,EAAE,IAAI;IACT,MAAM,EAAE,IAAI;IACZ,IAAI,EAAE,KAAK;CACX,CAAC;AAEF,MAAM,UAAU,cAAc,CAAC,MAAiC;IAC/D,OAAO,MAAM,KAAK,OAAO,IAAI,MAAM,KAAK,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;AACjE,CAAC;AAED,MAAM,UAAU,sBAAsB,CAAC,cAA6B,EAAE,aAA+B;IACpG,MAAM,OAAO,GAAG,EAAE,GAAG,wBAAwB,EAAE,GAAG,aAAa,EAAE,CAAC;IAClE,MAAM,KAAK,GAAG,cAAc,CAAC,cAAc,CAAE,CAAC;IAC9C,OAAO,OAAO,CAAC,KAAK,CAAE,CAAC;AACxB,CAAC;AAED,kGAAkG;AAClG,MAAM,UAAU,+BAA+B,CAAC,cAAsB,EAAE,OAAe;IACtF,OAAO,IAAI,CAAC,GAAG,CAAC,cAAc,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,GAAG,iBAAiB,CAAC,CAAC,CAAC;AAC3E,CAAC;AAED,MAAM,UAAU,0BAA0B;AACzC,wFAAwF;AACxF,aAAiC,EACjC,cAAsB,EACtB,cAA6B,EAC7B,aAA+B;IAE/B,IAAI,cAAc,GAAG,sBAAsB,CAAC,cAAc,EAAE,aAAa,CAAC,CAAC;IAC3E,MAAM,SAAS,GACd,aAAa,KAAK,SAAS,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,aAAa,GAAG,cAAc,EAAE,cAAc,CAAC,CAAC;IAEzG,IAAI,SAAS,IAAI,cAAc,EAAE,CAAC;QACjC,cAAc,GAAG,+BAA+B,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;IAC7E,CAAC;IAED,OAAO,EAAE,SAAS,EAAE,cAAc,EAAE,CAAC;AACtC,CAAC","sourcesContent":["import { clampThinkingLevel } from \"../models.ts\";\nimport type {\n\tApi,\n\tModel,\n\tModelThinkingLevel,\n\tSamplingParams,\n\tSimpleStreamOptions,\n\tStreamOptions,\n\tThinkingBudgets,\n\tThinkingLevel,\n\tTranscriptContext,\n} from \"../types.ts\";\nimport { estimateContextTokens } from \"../utils/estimate.ts\";\n\nconst CONTEXT_SAFETY_TOKENS = 4096;\nconst MIN_MAX_TOKENS = 1;\n\nexport function clampMaxTokensToContext(model: Model<Api>, context: TranscriptContext, maxTokens: number): number {\n\tif (model.contextWindow <= 0) return Math.max(MIN_MAX_TOKENS, maxTokens);\n\tconst available = model.contextWindow - estimateContextTokens(context).tokens - CONTEXT_SAFETY_TOKENS;\n\treturn Math.min(maxTokens, Math.max(MIN_MAX_TOKENS, available));\n}\n\nexport function resolveSamplingParams(\n\tmodel: Model<Api>,\n\tthinkingLevel: ModelThinkingLevel,\n\trequestParams?: SamplingParams,\n): SamplingParams | undefined {\n\tconst effectiveThinkingLevel = clampThinkingLevel(model, thinkingLevel);\n\tconst thinkingLevelParams = model.samplingParamsByThinkingLevel?.[effectiveThinkingLevel];\n\treturn model.samplingParams || thinkingLevelParams || requestParams\n\t\t? { ...model.samplingParams, ...thinkingLevelParams, ...requestParams }\n\t\t: undefined;\n}\n\nexport function buildBaseOptions(\n\tmodel: Model<Api>,\n\tcontext: TranscriptContext,\n\toptions?: SimpleStreamOptions,\n\tapiKey?: string,\n): StreamOptions {\n\tconst samplingParams = resolveSamplingParams(model, options?.reasoning ?? \"off\", options?.samplingParams);\n\treturn {\n\t\ttemperature: options?.temperature,\n\t\tsamplingParams,\n\t\tmaxTokens: clampMaxTokensToContext(model, context, options?.maxTokens ?? model.maxTokens),\n\t\tsignal: options?.signal,\n\t\ttelemetry: options?.telemetry,\n\t\tapiKey: apiKey || options?.apiKey,\n\t\tfetch: options?.fetch,\n\t\ttransport: options?.transport,\n\t\tcacheRetention: options?.cacheRetention,\n\t\tsessionId: options?.sessionId,\n\t\theaders: options?.headers,\n\t\tidentity: options?.identity,\n\t\tonPayload: options?.onPayload,\n\t\tonResponse: options?.onResponse,\n\t\tonProviderStreamEvent: options?.onProviderStreamEvent,\n\t\ttimeoutMs: options?.timeoutMs,\n\t\twebsocketConnectTimeoutMs: options?.websocketConnectTimeoutMs,\n\t\tmaxRetries: options?.maxRetries,\n\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\tmetadata: options?.metadata,\n\t\tenv: options?.env,\n\t};\n}\n\n/** Tokens always left for the answer when a thinking budget shares the response ceiling. */\nexport const MIN_ANSWER_TOKENS = 1024;\n\nexport const DEFAULT_THINKING_BUDGETS: ThinkingBudgets = {\n\tminimal: 1024,\n\tlow: 2048,\n\tmedium: 8192,\n\thigh: 16384,\n};\n\nexport function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, \"xhigh\" | \"max\"> | undefined {\n\treturn effort === \"xhigh\" || effort === \"max\" ? \"high\" : effort;\n}\n\nexport function thinkingBudgetForLevel(reasoningLevel: ThinkingLevel, customBudgets?: ThinkingBudgets): number {\n\tconst budgets = { ...DEFAULT_THINKING_BUDGETS, ...customBudgets };\n\tconst level = clampReasoning(reasoningLevel)!;\n\treturn budgets[level]!;\n}\n\n/** Cap a thinking budget so at least MIN_ANSWER_TOKENS remain under a shared response ceiling. */\nexport function clampThinkingBudgetToAnswerRoom(thinkingBudget: number, ceiling: number): number {\n\treturn Math.min(thinkingBudget, Math.max(0, ceiling - MIN_ANSWER_TOKENS));\n}\n\nexport function adjustMaxTokensForThinking(\n\t// Undefined means no explicit caller cap. Use the model cap and fit thinking inside it.\n\tbaseMaxTokens: number | undefined,\n\tmodelMaxTokens: number,\n\treasoningLevel: ThinkingLevel,\n\tcustomBudgets?: ThinkingBudgets,\n): { maxTokens: number; thinkingBudget: number } {\n\tlet thinkingBudget = thinkingBudgetForLevel(reasoningLevel, customBudgets);\n\tconst maxTokens =\n\t\tbaseMaxTokens === undefined ? modelMaxTokens : Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);\n\n\tif (maxTokens <= thinkingBudget) {\n\t\tthinkingBudget = clampThinkingBudgetToAnswerRoom(thinkingBudget, maxTokens);\n\t}\n\n\treturn { maxTokens, thinkingBudget };\n}\n"]}
+{"version":3,"file":"simple-options.js","sourceRoot":"","sources":["../../src/api/simple-options.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAYlD,OAAO,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAE7D,MAAM,qBAAqB,GAAG,IAAI,CAAC;AACnC,8FAA8F;AAC9F,yFAAyF;AACzF,MAAM,oBAAoB,GAAG,IAAI,CAAC;AAClC,kGAAkG;AAClG,kGAAkG;AAClG,qGAAqG;AACrG,MAAM,4BAA4B,GAAG,KAAK,CAAC;AAC3C,MAAM,cAAc,GAAG,CAAC,CAAC;AAEzB,6GAA6G;AAC7G,MAAM,UAAU,uBAAuB,CAAC,KAAiB,EAAE,OAA0B,EAAE,SAAiB;IACvG,IAAI,KAAK,CAAC,aAAa,IAAI,CAAC;QAAE,OAAO,IAAI,CAAC,GAAG,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;IACzE,MAAM,KAAK,GAAG,qBAAqB,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC;IACpD,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,qBAAqB,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,oBAAoB,CAAC,CAAC,CAAC;IACxF,IAAI,IAAI,GAAG,KAAK,CAAC,aAAa,GAAG,KAAK,GAAG,MAAM,CAAC;IAChD,IAAI,IAAI,GAAG,4BAA4B,EAAE,CAAC;QACzC,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,4BAA4B,EAAE,KAAK,CAAC,aAAa,GAAG,KAAK,GAAG,qBAAqB,CAAC,CAAC;IACpG,CAAC;IACD,OAAO,IAAI,CAAC,GAAG,CAAC,SAAS,EAAE,IAAI,CAAC,GAAG,CAAC,cAAc,EAAE,IAAI,CAAC,CAAC,CAAC;AAC5D,CAAC;AAED,MAAM,UAAU,qBAAqB,CACpC,KAAiB,EACjB,aAAiC,EACjC,aAA8B;IAE9B,MAAM,sBAAsB,GAAG,kBAAkB,CAAC,KAAK,EAAE,aAAa,CAAC,CAAC;IACxE,MAAM,mBAAmB,GAAG,KAAK,CAAC,6BAA6B,EAAE,CAAC,sBAAsB,CAAC,CAAC;IAC1F,OAAO,KAAK,CAAC,cAAc,IAAI,mBAAmB,IAAI,aAAa;QAClE,CAAC,CAAC,EAAE,GAAG,KAAK,CAAC,cAAc,EAAE,GAAG,mBAAmB,EAAE,GAAG,aAAa,EAAE;QACvE,CAAC,CAAC,SAAS,CAAC;AACd,CAAC;AAED,MAAM,UAAU,gBAAgB,CAC/B,KAAiB,EACjB,OAA0B,EAC1B,OAA6B,EAC7B,MAAe;IAEf,MAAM,cAAc,GAAG,qBAAqB,CAAC,KAAK,EAAE,OAAO,EAAE,SAAS,IAAI,KAAK,EAAE,OAAO,EAAE,cAAc,CAAC,CAAC;IAC1G,OAAO;QACN,WAAW,EAAE,OAAO,EAAE,WAAW;QACjC,cAAc;QACd,SAAS,EAAE,uBAAuB,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,KAAK,CAAC,SAAS,CAAC;QACzF,MAAM,EAAE,OAAO,EAAE,MAAM;QACvB,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,MAAM,EAAE,MAAM,IAAI,OAAO,EAAE,MAAM;QACjC,KAAK,EAAE,OAAO,EAAE,KAAK;QACrB,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,cAAc,EAAE,OAAO,EAAE,cAAc;QACvC,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,mBAAmB,EAAE,OAAO,EAAE,mBAAmB;QACjD,OAAO,EAAE,OAAO,EAAE,OAAO;QACzB,QAAQ,EAAE,OAAO,EAAE,QAAQ;QAC3B,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,UAAU,EAAE,OAAO,EAAE,UAAU;QAC/B,qBAAqB,EAAE,OAAO,EAAE,qBAAqB;QACrD,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,yBAAyB,EAAE,OAAO,EAAE,yBAAyB;QAC7D,UAAU,EAAE,OAAO,EAAE,UAAU;QAC/B,eAAe,EAAE,OAAO,EAAE,eAAe;QACzC,QAAQ,EAAE,OAAO,EAAE,QAAQ;QAC3B,GAAG,EAAE,OAAO,EAAE,GAAG;KACjB,CAAC;AACH,CAAC;AAED,4FAA4F;AAC5F,MAAM,CAAC,MAAM,iBAAiB,GAAG,IAAI,CAAC;AAEtC,MAAM,CAAC,MAAM,wBAAwB,GAAoB;IACxD,OAAO,EAAE,IAAI;IACb,GAAG,EAAE,IAAI;IACT,MAAM,EAAE,IAAI;IACZ,IAAI,EAAE,KAAK;CACX,CAAC;AAEF,MAAM,UAAU,cAAc,CAAC,MAAiC;IAC/D,OAAO,MAAM,KAAK,OAAO,IAAI,MAAM,KAAK,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;AACjE,CAAC;AAED,MAAM,UAAU,sBAAsB,CAAC,cAA6B,EAAE,aAA+B;IACpG,MAAM,OAAO,GAAG,EAAE,GAAG,wBAAwB,EAAE,GAAG,aAAa,EAAE,CAAC;IAClE,MAAM,KAAK,GAAG,cAAc,CAAC,cAAc,CAAE,CAAC;IAC9C,OAAO,OAAO,CAAC,KAAK,CAAE,CAAC;AACxB,CAAC;AAED,kGAAkG;AAClG,MAAM,UAAU,+BAA+B,CAAC,cAAsB,EAAE,OAAe;IACtF,OAAO,IAAI,CAAC,GAAG,CAAC,cAAc,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,GAAG,iBAAiB,CAAC,CAAC,CAAC;AAC3E,CAAC;AAED,MAAM,UAAU,0BAA0B;AACzC,wFAAwF;AACxF,aAAiC,EACjC,cAAsB,EACtB,cAA6B,EAC7B,aAA+B;IAE/B,IAAI,cAAc,GAAG,sBAAsB,CAAC,cAAc,EAAE,aAAa,CAAC,CAAC;IAC3E,MAAM,SAAS,GACd,aAAa,KAAK,SAAS,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,aAAa,GAAG,cAAc,EAAE,cAAc,CAAC,CAAC;IAEzG,IAAI,SAAS,IAAI,cAAc,EAAE,CAAC;QACjC,cAAc,GAAG,+BAA+B,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;IAC7E,CAAC;IAED,OAAO,EAAE,SAAS,EAAE,cAAc,EAAE,CAAC;AACtC,CAAC","sourcesContent":["import { clampThinkingLevel } from \"../models.ts\";\nimport type {\n\tApi,\n\tModel,\n\tModelThinkingLevel,\n\tSamplingParams,\n\tSimpleStreamOptions,\n\tStreamOptions,\n\tThinkingBudgets,\n\tThinkingLevel,\n\tTranscriptContext,\n} from \"../types.ts\";\nimport { estimateContextTokens } from \"../utils/estimate.ts\";\n\nconst CONTEXT_SAFETY_TOKENS = 4096;\n// OpenRouter counts the input with its own tokenizer and rejects input + max_tokens above the\n// window before routing. It counted prose up to 12% above the provider's count (#10497).\nconst CONTEXT_SAFETY_RATIO = 0.15;\n// The 15% margin is kept only while it leaves at least this much output. Otherwise (a nearly full\n// large window, or any small window such as a 32k local model) only the fixed margin is kept, and\n// output is capped at this size so an OpenRouter overcount still fits for a while before compaction.\nconst MIN_OUTPUT_WITH_RATIO_MARGIN = 32768;\nconst MIN_MAX_TOKENS = 1;\n\n/** Fit maxTokens into the context window, leaving room for the input being counted higher than estimated. */\nexport function clampMaxTokensToContext(model: Model<Api>, context: TranscriptContext, maxTokens: number): number {\n\tif (model.contextWindow <= 0) return Math.max(MIN_MAX_TOKENS, maxTokens);\n\tconst input = estimateContextTokens(context).tokens;\n\tconst margin = Math.max(CONTEXT_SAFETY_TOKENS, Math.ceil(input * CONTEXT_SAFETY_RATIO));\n\tlet room = model.contextWindow - input - margin;\n\tif (room < MIN_OUTPUT_WITH_RATIO_MARGIN) {\n\t\troom = Math.min(MIN_OUTPUT_WITH_RATIO_MARGIN, model.contextWindow - input - CONTEXT_SAFETY_TOKENS);\n\t}\n\treturn Math.min(maxTokens, Math.max(MIN_MAX_TOKENS, room));\n}\n\nexport function resolveSamplingParams(\n\tmodel: Model<Api>,\n\tthinkingLevel: ModelThinkingLevel,\n\trequestParams?: SamplingParams,\n): SamplingParams | undefined {\n\tconst effectiveThinkingLevel = clampThinkingLevel(model, thinkingLevel);\n\tconst thinkingLevelParams = model.samplingParamsByThinkingLevel?.[effectiveThinkingLevel];\n\treturn model.samplingParams || thinkingLevelParams || requestParams\n\t\t? { ...model.samplingParams, ...thinkingLevelParams, ...requestParams }\n\t\t: undefined;\n}\n\nexport function buildBaseOptions(\n\tmodel: Model<Api>,\n\tcontext: TranscriptContext,\n\toptions?: SimpleStreamOptions,\n\tapiKey?: string,\n): StreamOptions {\n\tconst samplingParams = resolveSamplingParams(model, options?.reasoning ?? \"off\", options?.samplingParams);\n\treturn {\n\t\ttemperature: options?.temperature,\n\t\tsamplingParams,\n\t\tmaxTokens: clampMaxTokensToContext(model, context, options?.maxTokens ?? model.maxTokens),\n\t\tsignal: options?.signal,\n\t\ttelemetry: options?.telemetry,\n\t\tapiKey: apiKey || options?.apiKey,\n\t\tfetch: options?.fetch,\n\t\ttransport: options?.transport,\n\t\tcacheRetention: options?.cacheRetention,\n\t\tsessionId: options?.sessionId,\n\t\tdisableContinuation: options?.disableContinuation,\n\t\theaders: options?.headers,\n\t\tidentity: options?.identity,\n\t\tonPayload: options?.onPayload,\n\t\tonResponse: options?.onResponse,\n\t\tonProviderStreamEvent: options?.onProviderStreamEvent,\n\t\ttimeoutMs: options?.timeoutMs,\n\t\twebsocketConnectTimeoutMs: options?.websocketConnectTimeoutMs,\n\t\tmaxRetries: options?.maxRetries,\n\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\tmetadata: options?.metadata,\n\t\tenv: options?.env,\n\t};\n}\n\n/** Tokens always left for the answer when a thinking budget shares the response ceiling. */\nexport const MIN_ANSWER_TOKENS = 1024;\n\nexport const DEFAULT_THINKING_BUDGETS: ThinkingBudgets = {\n\tminimal: 1024,\n\tlow: 2048,\n\tmedium: 8192,\n\thigh: 16384,\n};\n\nexport function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, \"xhigh\" | \"max\"> | undefined {\n\treturn effort === \"xhigh\" || effort === \"max\" ? \"high\" : effort;\n}\n\nexport function thinkingBudgetForLevel(reasoningLevel: ThinkingLevel, customBudgets?: ThinkingBudgets): number {\n\tconst budgets = { ...DEFAULT_THINKING_BUDGETS, ...customBudgets };\n\tconst level = clampReasoning(reasoningLevel)!;\n\treturn budgets[level]!;\n}\n\n/** Cap a thinking budget so at least MIN_ANSWER_TOKENS remain under a shared response ceiling. */\nexport function clampThinkingBudgetToAnswerRoom(thinkingBudget: number, ceiling: number): number {\n\treturn Math.min(thinkingBudget, Math.max(0, ceiling - MIN_ANSWER_TOKENS));\n}\n\nexport function adjustMaxTokensForThinking(\n\t// Undefined means no explicit caller cap. Use the model cap and fit thinking inside it.\n\tbaseMaxTokens: number | undefined,\n\tmodelMaxTokens: number,\n\treasoningLevel: ThinkingLevel,\n\tcustomBudgets?: ThinkingBudgets,\n): { maxTokens: number; thinkingBudget: number } {\n\tlet thinkingBudget = thinkingBudgetForLevel(reasoningLevel, customBudgets);\n\tconst maxTokens =\n\t\tbaseMaxTokens === undefined ? modelMaxTokens : Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);\n\n\tif (maxTokens <= thinkingBudget) {\n\t\tthinkingBudget = clampThinkingBudgetToAnswerRoom(thinkingBudget, maxTokens);\n\t}\n\n\treturn { maxTokens, thinkingBudget };\n}\n"]}

package/dist/api/strict-tool-limits.d.ts

@@ -0,0 +1,13 @@
+/** Returns true when a provider error message says the request's strict tools exceed the provider's limits. */
+export declare function isStrictToolLimitErrorMessage(message: string): boolean;
+/**
+ * Records that a session's strict tools exceeded Claude's limits; does nothing without a session. Keeps at most
+ * `MAX_SESSIONS_OVER_STRICT_TOOL_LIMITS` sessions.
+ *
+ * Entries are normally freed by `cleanupSessionResources(sessionId)` when a session ends. The cap bounds memory for
+ * library users that never call it; a dropped session just pays one more rejection.
+ */
+export declare function markStrictToolLimitsExceeded(sessionId: string | undefined): void;
+/** Returns true when an earlier request of this session had its strict tools rejected for exceeding Claude's limits. */
+export declare function hasExceededStrictToolLimits(sessionId: string | undefined): boolean;
+//# sourceMappingURL=strict-tool-limits.d.ts.map

package/dist/api/strict-tool-limits.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"strict-tool-limits.d.ts","sourceRoot":"","sources":["../../src/api/strict-tool-limits.ts"],"names":[],"mappings":"AAYA,+GAA+G;AAC/G,wBAAgB,6BAA6B,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAEtE;AAWD;;;;;;GAMG;AACH,wBAAgB,4BAA4B,CAAC,SAAS,EAAE,MAAM,GAAG,SAAS,GAAG,IAAI,CAOhF;AAED,wHAAwH;AACxH,wBAAgB,2BAA2B,CAAC,SAAS,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,CAElF"}

package/dist/api/strict-tool-limits.js

@@ -0,0 +1,50 @@
+import { registerSessionResourceCleanup } from "../session-resources.js";
+/**
+ * Claude rejects (on the Anthropic API and on Bedrock) a request whose strict tools exceed its limits with one of these
+ * messages. Observed limits: 20 strict tools, and 16 optional or union-typed parameters across all strict tools.
+ */
+const STRICT_TOOL_LIMIT_ERROR_MESSAGES = [
+    "The compiled grammar is too large",
+    "Too many strict tools (",
+    "Schemas contains too many parameters with union types (",
+];
+/** Returns true when a provider error message says the request's strict tools exceed the provider's limits. */
+export function isStrictToolLimitErrorMessage(message) {
+    return STRICT_TOOL_LIMIT_ERROR_MESSAGES.some((limitMessage) => message.includes(limitMessage));
+}
+/**
+ * Sessions whose strict tools Claude rejected for exceeding its limits. Their later requests send `strict: "prefer"`
+ * tools non-strict right away: the tool set rarely changes between turns, so trying strict again would only repeat
+ * the rejection, and switching back to strict later would rewrite the tool definitions and miss the prompt cache.
+ * Shared by the Anthropic and Bedrock providers, which run the same Claude limits.
+ */
+const sessionsOverStrictToolLimits = new Set();
+const MAX_SESSIONS_OVER_STRICT_TOOL_LIMITS = 10_000;
+/**
+ * Records that a session's strict tools exceeded Claude's limits; does nothing without a session. Keeps at most
+ * `MAX_SESSIONS_OVER_STRICT_TOOL_LIMITS` sessions.
+ *
+ * Entries are normally freed by `cleanupSessionResources(sessionId)` when a session ends. The cap bounds memory for
+ * library users that never call it; a dropped session just pays one more rejection.
+ */
+export function markStrictToolLimitsExceeded(sessionId) {
+    if (sessionId === undefined)
+        return;
+    sessionsOverStrictToolLimits.add(sessionId);
+    if (sessionsOverStrictToolLimits.size > MAX_SESSIONS_OVER_STRICT_TOOL_LIMITS) {
+        const oldest = sessionsOverStrictToolLimits.values().next().value;
+        if (oldest !== undefined)
+            sessionsOverStrictToolLimits.delete(oldest);
+    }
+}
+/** Returns true when an earlier request of this session had its strict tools rejected for exceeding Claude's limits. */
+export function hasExceededStrictToolLimits(sessionId) {
+    return sessionId !== undefined && sessionsOverStrictToolLimits.has(sessionId);
+}
+registerSessionResourceCleanup((sessionId) => {
+    if (sessionId)
+        sessionsOverStrictToolLimits.delete(sessionId);
+    else
+        sessionsOverStrictToolLimits.clear();
+});
+//# sourceMappingURL=strict-tool-limits.js.map

package/dist/api/strict-tool-limits.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"strict-tool-limits.js","sourceRoot":"","sources":["../../src/api/strict-tool-limits.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,8BAA8B,EAAE,MAAM,yBAAyB,CAAC;AAEzE;;;GAGG;AACH,MAAM,gCAAgC,GAAG;IACxC,mCAAmC;IACnC,yBAAyB;IACzB,yDAAyD;CACzD,CAAC;AAEF,+GAA+G;AAC/G,MAAM,UAAU,6BAA6B,CAAC,OAAe;IAC5D,OAAO,gCAAgC,CAAC,IAAI,CAAC,CAAC,YAAY,EAAE,EAAE,CAAC,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAC,CAAC,CAAC;AAChG,CAAC;AAED;;;;;GAKG;AACH,MAAM,4BAA4B,GAAG,IAAI,GAAG,EAAU,CAAC;AACvD,MAAM,oCAAoC,GAAG,MAAM,CAAC;AAEpD;;;;;;GAMG;AACH,MAAM,UAAU,4BAA4B,CAAC,SAA6B;IACzE,IAAI,SAAS,KAAK,SAAS;QAAE,OAAO;IACpC,4BAA4B,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;IAC5C,IAAI,4BAA4B,CAAC,IAAI,GAAG,oCAAoC,EAAE,CAAC;QAC9E,MAAM,MAAM,GAAG,4BAA4B,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC;QAClE,IAAI,MAAM,KAAK,SAAS;YAAE,4BAA4B,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IACvE,CAAC;AACF,CAAC;AAED,wHAAwH;AACxH,MAAM,UAAU,2BAA2B,CAAC,SAA6B;IACxE,OAAO,SAAS,KAAK,SAAS,IAAI,4BAA4B,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;AAC/E,CAAC;AAED,8BAA8B,CAAC,CAAC,SAAS,EAAE,EAAE;IAC5C,IAAI,SAAS;QAAE,4BAA4B,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;;QACzD,4BAA4B,CAAC,KAAK,EAAE,CAAC;AAC3C,CAAC,CAAC,CAAC","sourcesContent":["import { registerSessionResourceCleanup } from \"../session-resources.ts\";\n\n/**\n * Claude rejects (on the Anthropic API and on Bedrock) a request whose strict tools exceed its limits with one of these\n * messages. Observed limits: 20 strict tools, and 16 optional or union-typed parameters across all strict tools.\n */\nconst STRICT_TOOL_LIMIT_ERROR_MESSAGES = [\n\t\"The compiled grammar is too large\",\n\t\"Too many strict tools (\",\n\t\"Schemas contains too many parameters with union types (\",\n];\n\n/** Returns true when a provider error message says the request's strict tools exceed the provider's limits. */\nexport function isStrictToolLimitErrorMessage(message: string): boolean {\n\treturn STRICT_TOOL_LIMIT_ERROR_MESSAGES.some((limitMessage) => message.includes(limitMessage));\n}\n\n/**\n * Sessions whose strict tools Claude rejected for exceeding its limits. Their later requests send `strict: \"prefer\"`\n * tools non-strict right away: the tool set rarely changes between turns, so trying strict again would only repeat\n * the rejection, and switching back to strict later would rewrite the tool definitions and miss the prompt cache.\n * Shared by the Anthropic and Bedrock providers, which run the same Claude limits.\n */\nconst sessionsOverStrictToolLimits = new Set<string>();\nconst MAX_SESSIONS_OVER_STRICT_TOOL_LIMITS = 10_000;\n\n/**\n * Records that a session's strict tools exceeded Claude's limits; does nothing without a session. Keeps at most\n * `MAX_SESSIONS_OVER_STRICT_TOOL_LIMITS` sessions.\n *\n * Entries are normally freed by `cleanupSessionResources(sessionId)` when a session ends. The cap bounds memory for\n * library users that never call it; a dropped session just pays one more rejection.\n */\nexport function markStrictToolLimitsExceeded(sessionId: string | undefined): void {\n\tif (sessionId === undefined) return;\n\tsessionsOverStrictToolLimits.add(sessionId);\n\tif (sessionsOverStrictToolLimits.size > MAX_SESSIONS_OVER_STRICT_TOOL_LIMITS) {\n\t\tconst oldest = sessionsOverStrictToolLimits.values().next().value;\n\t\tif (oldest !== undefined) sessionsOverStrictToolLimits.delete(oldest);\n\t}\n}\n\n/** Returns true when an earlier request of this session had its strict tools rejected for exceeding Claude's limits. */\nexport function hasExceededStrictToolLimits(sessionId: string | undefined): boolean {\n\treturn sessionId !== undefined && sessionsOverStrictToolLimits.has(sessionId);\n}\n\nregisterSessionResourceCleanup((sessionId) => {\n\tif (sessionId) sessionsOverStrictToolLimits.delete(sessionId);\n\telse sessionsOverStrictToolLimits.clear();\n});\n"]}

package/dist/auth/oauth/load.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"load.d.ts","sourceRoot":"","sources":["../../../src/auth/oauth/load.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAa7C,KAAK,gBAAgB,GAAG;IACvB,SAAS,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IAChD,WAAW,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IAClD,aAAa,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IACpD,aAAa,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IACpD,UAAU,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IACjD,UAAU,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IACjD,IAAI,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IAC3C,GAAG,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IAC1C,MAAM,EAAE,CAAC,OAAO,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,KAAK,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;CACvF,CAAC;AAIF,4EAA4E;AAC5E,wBAAgB,+BAA+B,CAAC,OAAO,EAAE,gBAAgB,GAAG,IAAI,CAE/E;AAED,eAAO,MAAM,kBAAkB,QAAa,OAAO,CAAC,SAAS,CAG5D,CAAC;AAEF,eAAO,MAAM,oBAAoB,QAAa,OAAO,CAAC,SAAS,CAG9D,CAAC;AAEF,eAAO,MAAM,sBAAsB,QAAa,OAAO,CAAC,SAAS,CAGhE,CAAC;AAEF,eAAO,MAAM,sBAAsB,QAAa,OAAO,CAAC,SAAS,CAGhE,CAAC;AAEF,eAAO,MAAM,mBAAmB,QAAa,OAAO,CAAC,SAAS,CAG7D,CAAC;AAEF,eAAO,MAAM,mBAAmB,QAAa,OAAO,CAAC,SAAS,CAG7D,CAAC;AAEF,eAAO,MAAM,aAAa,QAAa,OAAO,CAAC,SAAS,CAGvD,CAAC;AAEF,eAAO,MAAM,YAAY,QAAa,OAAO,CAAC,SAAS,CAGtD,CAAC;AAEF,eAAO,MAAM,eAAe,YAAmB;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,KAAG,OAAO,CAAC,SAAS,CAOnG,CAAC"}
+{"version":3,"file":"load.d.ts","sourceRoot":"","sources":["../../../src/auth/oauth/load.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAgB7C,KAAK,gBAAgB,GAAG;IACvB,SAAS,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IAChD,WAAW,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IAClD,aAAa,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IACpD,aAAa,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IACpD,UAAU,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IACjD,UAAU,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IACjD,IAAI,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IAC3C,GAAG,EAAE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IAC1C,MAAM,EAAE,CAAC,OAAO,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,KAAK,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;CACvF,CAAC;AAIF,4EAA4E;AAC5E,wBAAgB,+BAA+B,CAAC,OAAO,EAAE,gBAAgB,GAAG,IAAI,CAE/E;AAED,eAAO,MAAM,kBAAkB,QAAa,OAAO,CAAC,SAAS,CAG5D,CAAC;AAEF,eAAO,MAAM,oBAAoB,QAAa,OAAO,CAAC,SAAS,CAG9D,CAAC;AAEF,eAAO,MAAM,sBAAsB,QAAa,OAAO,CAAC,SAAS,CAGhE,CAAC;AAEF,eAAO,MAAM,sBAAsB,QAAa,OAAO,CAAC,SAAS,CAGhE,CAAC;AAEF,eAAO,MAAM,mBAAmB,QAAa,OAAO,CAAC,SAAS,CAG7D,CAAC;AAEF,eAAO,MAAM,mBAAmB,QAAa,OAAO,CAAC,SAAS,CAG7D,CAAC;AAEF,eAAO,MAAM,aAAa,QAAa,OAAO,CAAC,SAAS,CAGvD,CAAC;AAEF,eAAO,MAAM,YAAY,QAAa,OAAO,CAAC,SAAS,CAGtD,CAAC;AAEF,eAAO,MAAM,eAAe,YAAmB;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,KAAG,OAAO,CAAC,SAAS,CAOnG,CAAC"}

package/dist/auth/oauth/load.js

@@ -10,11 +10,14 @@ var __rewriteRelativeImportExtension = (this && this.__rewriteRelativeImportExte
  * Loads an OAuth flow module through a variable specifier so bundlers cannot
  * follow the import into Node-only flow code (`node:http` callback servers,
  * `node:crypto` PKCE). The `.ts`/`.js` rewrite keeps the trick working from
- * both source and built output.
+ * both source and built output. The specifier is resolved against this
+ * module's URL, not left relative: a bundler inlining this file (Chord's
+ * worker base) keeps the variable import, which would otherwise resolve
+ * against the bundle's location.
  */
 const importOAuthModule = (specifier) => {
     const runtimeSpecifier = import.meta.url.endsWith(".js") ? specifier.replace(/\.ts$/, ".js") : specifier;
-    return import(__rewriteRelativeImportExtension(runtimeSpecifier));
+    return import(__rewriteRelativeImportExtension(new URL(runtimeSpecifier, import.meta.url).href));
 };
 let bundledLoaders;
 /** Registers statically bundled OAuth flows for standalone Bun binaries. */

package/dist/auth/oauth/load.js.map

@@ -1 +1 @@
-{"version":3,"file":"load.js","sourceRoot":"","sources":["../../../src/auth/oauth/load.ts"],"names":[],"mappings":";;;;;;;;AAEA;;;;;GAKG;AACH,MAAM,iBAAiB,GAAG,CAAC,SAAiB,EAAoB,EAAE;IACjE,MAAM,gBAAgB,GAAG,OAAO,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,OAAO,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACzG,OAAO,MAAM,kCAAC,gBAAgB,EAAC,CAAC;AACjC,CAAC,CAAC;AAcF,IAAI,cAA4C,CAAC;AAEjD,4EAA4E;AAC5E,MAAM,UAAU,+BAA+B,CAAC,OAAyB;IACxE,cAAc,GAAG,OAAO,CAAC;AAC1B,CAAC;AAED,MAAM,CAAC,MAAM,kBAAkB,GAAG,KAAK,IAAwB,EAAE;IAChE,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,SAAS,EAAE,CAAC;IACtD,OAAQ,CAAC,MAAM,iBAAiB,CAAC,gBAAgB,CAAC,CAAmC,CAAC,cAAc,CAAC;AACtG,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,oBAAoB,GAAG,KAAK,IAAwB,EAAE;IAClE,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,WAAW,EAAE,CAAC;IACxD,OAAQ,CAAC,MAAM,iBAAiB,CAAC,mBAAmB,CAAC,CAAqC,CAAC,gBAAgB,CAAC;AAC7G,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,sBAAsB,GAAG,KAAK,IAAwB,EAAE;IACpE,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,aAAa,EAAE,CAAC;IAC1D,OAAQ,CAAC,MAAM,iBAAiB,CAAC,qBAAqB,CAAC,CAAuC,CAAC,kBAAkB,CAAC;AACnH,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,sBAAsB,GAAG,KAAK,IAAwB,EAAE;IACpE,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,aAAa,EAAE,CAAC;IAC1D,OAAQ,CAAC,MAAM,iBAAiB,CAAC,qBAAqB,CAAC,CAAuC,CAAC,kBAAkB,CAAC;AACnH,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,mBAAmB,GAAG,KAAK,IAAwB,EAAE;IACjE,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,UAAU,EAAE,CAAC;IACvD,OAAQ,CAAC,MAAM,iBAAiB,CAAC,iBAAiB,CAAC,CAAoC,CAAC,eAAe,CAAC;AACzG,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,mBAAmB,GAAG,KAAK,IAAwB,EAAE;IACjE,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,UAAU,EAAE,CAAC;IACvD,OAAQ,CAAC,MAAM,iBAAiB,CAAC,kBAAkB,CAAC,CAAoC,CAAC,eAAe,CAAC;AAC1G,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,aAAa,GAAG,KAAK,IAAwB,EAAE;IAC3D,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,IAAI,EAAE,CAAC;IACjD,OAAQ,CAAC,MAAM,iBAAiB,CAAC,WAAW,CAAC,CAA8B,CAAC,SAAS,CAAC;AACvF,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,YAAY,GAAG,KAAK,IAAwB,EAAE;IAC1D,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,GAAG,EAAE,CAAC;IAChD,OAAQ,CAAC,MAAM,iBAAiB,CAAC,UAAU,CAAC,CAA6B,CAAC,QAAQ,CAAC;AACpF,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,eAAe,GAAG,KAAK,EAAE,OAA0C,EAAsB,EAAE;IACvG,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAC1D,OACC,CAAC,MAAM,iBAAiB,CAAC,aAAa,CAAC,CAGvC,CAAC,iBAAiB,CAAC,OAAO,CAAC,CAAC;AAC9B,CAAC,CAAC","sourcesContent":["import type { OAuthAuth } from \"../types.ts\";\n\n/**\n * Loads an OAuth flow module through a variable specifier so bundlers cannot\n * follow the import into Node-only flow code (`node:http` callback servers,\n * `node:crypto` PKCE). The `.ts`/`.js` rewrite keeps the trick working from\n * both source and built output.\n */\nconst importOAuthModule = (specifier: string): Promise<unknown> => {\n\tconst runtimeSpecifier = import.meta.url.endsWith(\".js\") ? specifier.replace(/\\.ts$/, \".js\") : specifier;\n\treturn import(runtimeSpecifier);\n};\n\ntype OAuthFlowLoaders = {\n\tanthropic: () => OAuthAuth | Promise<OAuthAuth>;\n\topenaiCodex: () => OAuthAuth | Promise<OAuthAuth>;\n\topenaiChatGPT: () => OAuthAuth | Promise<OAuthAuth>;\n\tgithubCopilot: () => OAuthAuth | Promise<OAuthAuth>;\n\topenrouter: () => OAuthAuth | Promise<OAuthAuth>;\n\tkimiCoding: () => OAuthAuth | Promise<OAuthAuth>;\n\tmeta: () => OAuthAuth | Promise<OAuthAuth>;\n\txai: () => OAuthAuth | Promise<OAuthAuth>;\n\tradius: (options: { name: string; gateway: string }) => OAuthAuth | Promise<OAuthAuth>;\n};\n\nlet bundledLoaders: OAuthFlowLoaders | undefined;\n\n/** Registers statically bundled OAuth flows for standalone Bun binaries. */\nexport function registerBundledOAuthFlowLoaders(loaders: OAuthFlowLoaders): void {\n\tbundledLoaders = loaders;\n}\n\nexport const loadAnthropicOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.anthropic();\n\treturn ((await importOAuthModule(\"./anthropic.ts\")) as { anthropicOAuth: OAuthAuth }).anthropicOAuth;\n};\n\nexport const loadOpenAICodexOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.openaiCodex();\n\treturn ((await importOAuthModule(\"./openai-codex.ts\")) as { openaiCodexOAuth: OAuthAuth }).openaiCodexOAuth;\n};\n\nexport const loadOpenAIChatGPTOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.openaiChatGPT();\n\treturn ((await importOAuthModule(\"./openai-chatgpt.ts\")) as { openaiChatGPTOAuth: OAuthAuth }).openaiChatGPTOAuth;\n};\n\nexport const loadGitHubCopilotOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.githubCopilot();\n\treturn ((await importOAuthModule(\"./github-copilot.ts\")) as { githubCopilotOAuth: OAuthAuth }).githubCopilotOAuth;\n};\n\nexport const loadOpenRouterOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.openrouter();\n\treturn ((await importOAuthModule(\"./openrouter.ts\")) as { openRouterOAuth: OAuthAuth }).openRouterOAuth;\n};\n\nexport const loadKimiCodingOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.kimiCoding();\n\treturn ((await importOAuthModule(\"./kimi-coding.ts\")) as { kimiCodingOAuth: OAuthAuth }).kimiCodingOAuth;\n};\n\nexport const loadMetaOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.meta();\n\treturn ((await importOAuthModule(\"./meta.ts\")) as { metaOAuth: OAuthAuth }).metaOAuth;\n};\n\nexport const loadXaiOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.xai();\n\treturn ((await importOAuthModule(\"./xai.ts\")) as { xaiOAuth: OAuthAuth }).xaiOAuth;\n};\n\nexport const loadRadiusOAuth = async (options: { name: string; gateway: string }): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.radius(options);\n\treturn (\n\t\t(await importOAuthModule(\"./radius.ts\")) as {\n\t\t\tcreateRadiusOAuth: (input: { name: string; gateway: string }) => OAuthAuth;\n\t\t}\n\t).createRadiusOAuth(options);\n};\n"]}
+{"version":3,"file":"load.js","sourceRoot":"","sources":["../../../src/auth/oauth/load.ts"],"names":[],"mappings":";;;;;;;;AAEA;;;;;;;;GAQG;AACH,MAAM,iBAAiB,GAAG,CAAC,SAAiB,EAAoB,EAAE;IACjE,MAAM,gBAAgB,GAAG,OAAO,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,OAAO,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACzG,OAAO,MAAM,kCAAC,IAAI,GAAG,CAAC,gBAAgB,EAAE,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,EAAC,CAAC;AAChE,CAAC,CAAC;AAcF,IAAI,cAA4C,CAAC;AAEjD,4EAA4E;AAC5E,MAAM,UAAU,+BAA+B,CAAC,OAAyB;IACxE,cAAc,GAAG,OAAO,CAAC;AAC1B,CAAC;AAED,MAAM,CAAC,MAAM,kBAAkB,GAAG,KAAK,IAAwB,EAAE;IAChE,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,SAAS,EAAE,CAAC;IACtD,OAAQ,CAAC,MAAM,iBAAiB,CAAC,gBAAgB,CAAC,CAAmC,CAAC,cAAc,CAAC;AACtG,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,oBAAoB,GAAG,KAAK,IAAwB,EAAE;IAClE,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,WAAW,EAAE,CAAC;IACxD,OAAQ,CAAC,MAAM,iBAAiB,CAAC,mBAAmB,CAAC,CAAqC,CAAC,gBAAgB,CAAC;AAC7G,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,sBAAsB,GAAG,KAAK,IAAwB,EAAE;IACpE,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,aAAa,EAAE,CAAC;IAC1D,OAAQ,CAAC,MAAM,iBAAiB,CAAC,qBAAqB,CAAC,CAAuC,CAAC,kBAAkB,CAAC;AACnH,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,sBAAsB,GAAG,KAAK,IAAwB,EAAE;IACpE,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,aAAa,EAAE,CAAC;IAC1D,OAAQ,CAAC,MAAM,iBAAiB,CAAC,qBAAqB,CAAC,CAAuC,CAAC,kBAAkB,CAAC;AACnH,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,mBAAmB,GAAG,KAAK,IAAwB,EAAE;IACjE,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,UAAU,EAAE,CAAC;IACvD,OAAQ,CAAC,MAAM,iBAAiB,CAAC,iBAAiB,CAAC,CAAoC,CAAC,eAAe,CAAC;AACzG,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,mBAAmB,GAAG,KAAK,IAAwB,EAAE;IACjE,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,UAAU,EAAE,CAAC;IACvD,OAAQ,CAAC,MAAM,iBAAiB,CAAC,kBAAkB,CAAC,CAAoC,CAAC,eAAe,CAAC;AAC1G,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,aAAa,GAAG,KAAK,IAAwB,EAAE;IAC3D,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,IAAI,EAAE,CAAC;IACjD,OAAQ,CAAC,MAAM,iBAAiB,CAAC,WAAW,CAAC,CAA8B,CAAC,SAAS,CAAC;AACvF,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,YAAY,GAAG,KAAK,IAAwB,EAAE;IAC1D,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,GAAG,EAAE,CAAC;IAChD,OAAQ,CAAC,MAAM,iBAAiB,CAAC,UAAU,CAAC,CAA6B,CAAC,QAAQ,CAAC;AACpF,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,eAAe,GAAG,KAAK,EAAE,OAA0C,EAAsB,EAAE;IACvG,IAAI,cAAc;QAAE,OAAO,cAAc,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAC1D,OACC,CAAC,MAAM,iBAAiB,CAAC,aAAa,CAAC,CAGvC,CAAC,iBAAiB,CAAC,OAAO,CAAC,CAAC;AAC9B,CAAC,CAAC","sourcesContent":["import type { OAuthAuth } from \"../types.ts\";\n\n/**\n * Loads an OAuth flow module through a variable specifier so bundlers cannot\n * follow the import into Node-only flow code (`node:http` callback servers,\n * `node:crypto` PKCE). The `.ts`/`.js` rewrite keeps the trick working from\n * both source and built output. The specifier is resolved against this\n * module's URL, not left relative: a bundler inlining this file (Chord's\n * worker base) keeps the variable import, which would otherwise resolve\n * against the bundle's location.\n */\nconst importOAuthModule = (specifier: string): Promise<unknown> => {\n\tconst runtimeSpecifier = import.meta.url.endsWith(\".js\") ? specifier.replace(/\\.ts$/, \".js\") : specifier;\n\treturn import(new URL(runtimeSpecifier, import.meta.url).href);\n};\n\ntype OAuthFlowLoaders = {\n\tanthropic: () => OAuthAuth | Promise<OAuthAuth>;\n\topenaiCodex: () => OAuthAuth | Promise<OAuthAuth>;\n\topenaiChatGPT: () => OAuthAuth | Promise<OAuthAuth>;\n\tgithubCopilot: () => OAuthAuth | Promise<OAuthAuth>;\n\topenrouter: () => OAuthAuth | Promise<OAuthAuth>;\n\tkimiCoding: () => OAuthAuth | Promise<OAuthAuth>;\n\tmeta: () => OAuthAuth | Promise<OAuthAuth>;\n\txai: () => OAuthAuth | Promise<OAuthAuth>;\n\tradius: (options: { name: string; gateway: string }) => OAuthAuth | Promise<OAuthAuth>;\n};\n\nlet bundledLoaders: OAuthFlowLoaders | undefined;\n\n/** Registers statically bundled OAuth flows for standalone Bun binaries. */\nexport function registerBundledOAuthFlowLoaders(loaders: OAuthFlowLoaders): void {\n\tbundledLoaders = loaders;\n}\n\nexport const loadAnthropicOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.anthropic();\n\treturn ((await importOAuthModule(\"./anthropic.ts\")) as { anthropicOAuth: OAuthAuth }).anthropicOAuth;\n};\n\nexport const loadOpenAICodexOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.openaiCodex();\n\treturn ((await importOAuthModule(\"./openai-codex.ts\")) as { openaiCodexOAuth: OAuthAuth }).openaiCodexOAuth;\n};\n\nexport const loadOpenAIChatGPTOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.openaiChatGPT();\n\treturn ((await importOAuthModule(\"./openai-chatgpt.ts\")) as { openaiChatGPTOAuth: OAuthAuth }).openaiChatGPTOAuth;\n};\n\nexport const loadGitHubCopilotOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.githubCopilot();\n\treturn ((await importOAuthModule(\"./github-copilot.ts\")) as { githubCopilotOAuth: OAuthAuth }).githubCopilotOAuth;\n};\n\nexport const loadOpenRouterOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.openrouter();\n\treturn ((await importOAuthModule(\"./openrouter.ts\")) as { openRouterOAuth: OAuthAuth }).openRouterOAuth;\n};\n\nexport const loadKimiCodingOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.kimiCoding();\n\treturn ((await importOAuthModule(\"./kimi-coding.ts\")) as { kimiCodingOAuth: OAuthAuth }).kimiCodingOAuth;\n};\n\nexport const loadMetaOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.meta();\n\treturn ((await importOAuthModule(\"./meta.ts\")) as { metaOAuth: OAuthAuth }).metaOAuth;\n};\n\nexport const loadXaiOAuth = async (): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.xai();\n\treturn ((await importOAuthModule(\"./xai.ts\")) as { xaiOAuth: OAuthAuth }).xaiOAuth;\n};\n\nexport const loadRadiusOAuth = async (options: { name: string; gateway: string }): Promise<OAuthAuth> => {\n\tif (bundledLoaders) return bundledLoaders.radius(options);\n\treturn (\n\t\t(await importOAuthModule(\"./radius.ts\")) as {\n\t\t\tcreateRadiusOAuth: (input: { name: string; gateway: string }) => OAuthAuth;\n\t\t}\n\t).createRadiusOAuth(options);\n};\n"]}

package/dist/cache-warming.d.ts

@@ -0,0 +1,53 @@
+import type { Api, Model, SimpleStreamOptions } from "./types.ts";
+/** Whether a refresh is sent ("warm") or warming ends until the next real request ("stop"). */
+export type CacheWarmingAction = "warm" | "stop";
+/** "streaming" while the agent run that sent the request is still active; "idle" after it settled. */
+export type CacheWarmingPhase = "streaming" | "idle";
+/** Inputs and outcome of one warm-or-stop decision. */
+export interface CacheWarmingDecision {
+    phase: CacheWarmingPhase;
+    /** Price of this refresh: a cache read of the prompt plus one output token. */
+    warmCost: number;
+    /** Extra price of the next real request if the cache entry is lost. */
+    missCost: number;
+    /** Estimated chance that a real request arrives before the entry expires. */
+    continuationProbability: number;
+    /** `continuationProbability * missCost - warmCost`. */
+    expectedSavings: number;
+    /** False when the prompt size or the model's prices are unknown. */
+    economicsAvailable: boolean;
+    /** "warm" when `expectedSavings` is at least `CACHE_WARMING_MINIMUM_EXPECTED_SAVINGS`. */
+    action: CacheWarmingAction;
+}
+/** Streaming warming never continues past this long after the real request that started it. */
+export declare const CACHE_WARMING_MAX_AGE_MS: number;
+/** Idle warming uses a shorter horizon because continuation estimates become less reliable with age. */
+export declare const CACHE_WARMING_MAX_IDLE_AGE_MS: number;
+/** A refresh is sent only when it is expected to save at least this many dollars. */
+export declare const CACHE_WARMING_MINIMUM_EXPECTED_SAVINGS = 0.05;
+/**
+ * Chance that a real request arrives before the cache entry expires while the agent sits idle. Measured from our own
+ * usage; per-session estimates were not better than this constant.
+ */
+export declare const CACHE_WARMING_IDLE_CONTINUATION_PROBABILITY = 0.15;
+/** Refresh at 90% of the TTL while preserving at least ten seconds of margin. */
+export declare function getCacheWarmingDelayMs(ttlMs: number): number | undefined;
+/**
+ * Lifetime of the prompt cache entry a request writes, from the model's `promptCache` tier for the retention the
+ * request used. Undefined when the model has no lifetime for that tier or caching is off.
+ */
+export declare function getPromptCacheTtlMs(model: Model<Api>, options: SimpleStreamOptions | undefined): number | undefined;
+/**
+ * Whether replaying the request with a one-token output cap leaves its cache entry untouched. Anthropic's budget-based
+ * thinking (Claude models without adaptive thinking) derives `budget_tokens` from `max_tokens`; the replay would get a
+ * different budget, which Anthropic keys the message cache on, and the model could still think for thousands of tokens.
+ */
+export declare function isCacheWarmingReplayable(model: Model<Api>, options: SimpleStreamOptions | undefined): boolean;
+/**
+ * The options of a refresh: the request's own, capped at one output token, without retries, and without replacing the
+ * conversation's provider thread state (the refresh repeats the last request).
+ */
+export declare function getCacheWarmingRequestOptions<T extends SimpleStreamOptions>(options: T, signal: AbortSignal): T;
+/** Decide whether refreshing the cache entry of a `promptTokens`-token request pays for itself in `phase`. */
+export declare function evaluateCacheWarming(model: Model<Api>, promptTokens: number, phase: CacheWarmingPhase): CacheWarmingDecision;
+//# sourceMappingURL=cache-warming.d.ts.map

package/dist/cache-warming.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"cache-warming.d.ts","sourceRoot":"","sources":["../src/cache-warming.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,GAAG,EAAE,KAAK,EAAE,mBAAmB,EAAS,MAAM,YAAY,CAAC;AAGzE,+FAA+F;AAC/F,MAAM,MAAM,kBAAkB,GAAG,MAAM,GAAG,MAAM,CAAC;AAEjD,sGAAsG;AACtG,MAAM,MAAM,iBAAiB,GAAG,WAAW,GAAG,MAAM,CAAC;AAErD,uDAAuD;AACvD,MAAM,WAAW,oBAAoB;IACpC,KAAK,EAAE,iBAAiB,CAAC;IACzB,+EAA+E;IAC/E,QAAQ,EAAE,MAAM,CAAC;IACjB,uEAAuE;IACvE,QAAQ,EAAE,MAAM,CAAC;IACjB,6EAA6E;IAC7E,uBAAuB,EAAE,MAAM,CAAC;IAChC,uDAAuD;IACvD,eAAe,EAAE,MAAM,CAAC;IACxB,oEAAoE;IACpE,kBAAkB,EAAE,OAAO,CAAC;IAC5B,0FAA0F;IAC1F,MAAM,EAAE,kBAAkB,CAAC;CAC3B;AAED,+FAA+F;AAC/F,eAAO,MAAM,wBAAwB,QAAc,CAAC;AAEpD,wGAAwG;AACxG,eAAO,MAAM,6BAA6B,QAAc,CAAC;AAEzD,qFAAqF;AACrF,eAAO,MAAM,sCAAsC,OAAO,CAAC;AAE3D;;;GAGG;AACH,eAAO,MAAM,2CAA2C,OAAO,CAAC;AAEhE,iFAAiF;AACjF,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAGxE;AAED;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,OAAO,EAAE,mBAAmB,GAAG,SAAS,GAAG,MAAM,GAAG,SAAS,CAOnH;AAED;;;;GAIG;AACH,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,OAAO,EAAE,mBAAmB,GAAG,SAAS,GAAG,OAAO,CAG7G;AAED;;;GAGG;AACH,wBAAgB,6BAA6B,CAAC,CAAC,SAAS,mBAAmB,EAAE,OAAO,EAAE,CAAC,EAAE,MAAM,EAAE,WAAW,GAAG,CAAC,CAE/G;AAED,8GAA8G;AAC9G,wBAAgB,oBAAoB,CACnC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,YAAY,EAAE,MAAM,EACpB,KAAK,EAAE,iBAAiB,GACtB,oBAAoB,CAoBtB"}

package/dist/cache-warming.js

@@ -0,0 +1,82 @@
+// Prompt cache warming policy shared by pi's hosts: when a refresh is due, whether a request can be replayed, and
+// whether a refresh pays for itself. Hosts own the scheduling; this module owns the numbers.
+import { calculateCost } from "./models.js";
+import { getProviderEnvValue } from "./utils/provider-env.js";
+/** Streaming warming never continues past this long after the real request that started it. */
+export const CACHE_WARMING_MAX_AGE_MS = 60 * 60_000;
+/** Idle warming uses a shorter horizon because continuation estimates become less reliable with age. */
+export const CACHE_WARMING_MAX_IDLE_AGE_MS = 30 * 60_000;
+/** A refresh is sent only when it is expected to save at least this many dollars. */
+export const CACHE_WARMING_MINIMUM_EXPECTED_SAVINGS = 0.05;
+/**
+ * Chance that a real request arrives before the cache entry expires while the agent sits idle. Measured from our own
+ * usage; per-session estimates were not better than this constant.
+ */
+export const CACHE_WARMING_IDLE_CONTINUATION_PROBABILITY = 0.15;
+/** Refresh at 90% of the TTL while preserving at least ten seconds of margin. */
+export function getCacheWarmingDelayMs(ttlMs) {
+    if (ttlMs <= 10_000)
+        return undefined;
+    return Math.max(1, Math.floor(Math.min(ttlMs * 0.9, ttlMs - 10_000)));
+}
+/**
+ * Lifetime of the prompt cache entry a request writes, from the model's `promptCache` tier for the retention the
+ * request used. Undefined when the model has no lifetime for that tier or caching is off.
+ */
+export function getPromptCacheTtlMs(model, options) {
+    const retention = options?.cacheRetention ??
+        (getProviderEnvValue("PI_CACHE_RETENTION", options?.env) === "long" ? "long" : "short");
+    if (retention === "none")
+        return undefined;
+    const seconds = model.promptCache?.[retention];
+    return seconds === undefined ? undefined : seconds * 1000;
+}
+/**
+ * Whether replaying the request with a one-token output cap leaves its cache entry untouched. Anthropic's budget-based
+ * thinking (Claude models without adaptive thinking) derives `budget_tokens` from `max_tokens`; the replay would get a
+ * different budget, which Anthropic keys the message cache on, and the model could still think for thousands of tokens.
+ */
+export function isCacheWarmingReplayable(model, options) {
+    if (!options?.reasoning || model.api !== "anthropic-messages")
+        return true;
+    return model.compat?.forceAdaptiveThinking === true;
+}
+/**
+ * The options of a refresh: the request's own, capped at one output token, without retries, and without replacing the
+ * conversation's provider thread state (the refresh repeats the last request).
+ */
+export function getCacheWarmingRequestOptions(options, signal) {
+    return { ...options, maxTokens: 1, disableContinuation: true, maxRetries: 0, signal };
+}
+/** Decide whether refreshing the cache entry of a `promptTokens`-token request pays for itself in `phase`. */
+export function evaluateCacheWarming(model, promptTokens, phase) {
+    const cacheHitCost = price(model, { cacheRead: promptTokens });
+    const cacheMissCost = price(model, model.cost.cacheWrite > 0 ? { cacheWrite: promptTokens } : { input: promptTokens });
+    const warmCost = price(model, { cacheRead: promptTokens, output: 1 });
+    const missCost = Math.max(0, cacheMissCost - cacheHitCost);
+    const continuationProbability = phase === "idle" ? CACHE_WARMING_IDLE_CONTINUATION_PROBABILITY : 1;
+    const economicsAvailable = promptTokens > 0 && (cacheHitCost > 0 || cacheMissCost > 0);
+    const expectedSavings = continuationProbability * missCost - warmCost;
+    return {
+        phase,
+        warmCost,
+        missCost,
+        continuationProbability,
+        expectedSavings,
+        economicsAvailable,
+        action: expectedSavings >= CACHE_WARMING_MINIMUM_EXPECTED_SAVINGS ? "warm" : "stop",
+    };
+}
+function price(model, tokens) {
+    const usage = {
+        input: 0,
+        output: 0,
+        cacheRead: 0,
+        cacheWrite: 0,
+        totalTokens: 0,
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+        ...tokens,
+    };
+    return calculateCost(model, usage).total;
+}
+//# sourceMappingURL=cache-warming.js.map

package/dist/cache-warming.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cache-warming.js","sourceRoot":"","sources":["../src/cache-warming.ts"],"names":[],"mappings":"AAAA,kHAAkH;AAClH,6FAA6F;AAC7F,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAE5C,OAAO,EAAE,mBAAmB,EAAE,MAAM,yBAAyB,CAAC;AAyB9D,+FAA+F;AAC/F,MAAM,CAAC,MAAM,wBAAwB,GAAG,EAAE,GAAG,MAAM,CAAC;AAEpD,wGAAwG;AACxG,MAAM,CAAC,MAAM,6BAA6B,GAAG,EAAE,GAAG,MAAM,CAAC;AAEzD,qFAAqF;AACrF,MAAM,CAAC,MAAM,sCAAsC,GAAG,IAAI,CAAC;AAE3D;;;GAGG;AACH,MAAM,CAAC,MAAM,2CAA2C,GAAG,IAAI,CAAC;AAEhE,iFAAiF;AACjF,MAAM,UAAU,sBAAsB,CAAC,KAAa;IACnD,IAAI,KAAK,IAAI,MAAM;QAAE,OAAO,SAAS,CAAC;IACtC,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,GAAG,GAAG,EAAE,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;AACvE,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,mBAAmB,CAAC,KAAiB,EAAE,OAAwC;IAC9F,MAAM,SAAS,GACd,OAAO,EAAE,cAAc;QACvB,CAAC,mBAAmB,CAAC,oBAAoB,EAAE,OAAO,EAAE,GAAG,CAAC,KAAK,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;IACzF,IAAI,SAAS,KAAK,MAAM;QAAE,OAAO,SAAS,CAAC;IAC3C,MAAM,OAAO,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC,SAAS,CAAC,CAAC;IAC/C,OAAO,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,GAAG,IAAI,CAAC;AAC3D,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,wBAAwB,CAAC,KAAiB,EAAE,OAAwC;IACnG,IAAI,CAAC,OAAO,EAAE,SAAS,IAAI,KAAK,CAAC,GAAG,KAAK,oBAAoB;QAAE,OAAO,IAAI,CAAC;IAC3E,OAAQ,KAAqC,CAAC,MAAM,EAAE,qBAAqB,KAAK,IAAI,CAAC;AACtF,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,6BAA6B,CAAgC,OAAU,EAAE,MAAmB;IAC3G,OAAO,EAAE,GAAG,OAAO,EAAE,SAAS,EAAE,CAAC,EAAE,mBAAmB,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC;AACvF,CAAC;AAED,8GAA8G;AAC9G,MAAM,UAAU,oBAAoB,CACnC,KAAiB,EACjB,YAAoB,EACpB,KAAwB;IAExB,MAAM,YAAY,GAAG,KAAK,CAAC,KAAK,EAAE,EAAE,SAAS,EAAE,YAAY,EAAE,CAAC,CAAC;IAC/D,MAAM,aAAa,GAAG,KAAK,CAC1B,KAAK,EACL,KAAK,CAAC,IAAI,CAAC,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,YAAY,EAAE,CAClF,CAAC;IACF,MAAM,QAAQ,GAAG,KAAK,CAAC,KAAK,EAAE,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC;IACtE,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,aAAa,GAAG,YAAY,CAAC,CAAC;IAC3D,MAAM,uBAAuB,GAAG,KAAK,KAAK,MAAM,CAAC,CAAC,CAAC,2CAA2C,CAAC,CAAC,CAAC,CAAC,CAAC;IACnG,MAAM,kBAAkB,GAAG,YAAY,GAAG,CAAC,IAAI,CAAC,YAAY,GAAG,CAAC,IAAI,aAAa,GAAG,CAAC,CAAC,CAAC;IACvF,MAAM,eAAe,GAAG,uBAAuB,GAAG,QAAQ,GAAG,QAAQ,CAAC;IACtE,OAAO;QACN,KAAK;QACL,QAAQ;QACR,QAAQ;QACR,uBAAuB;QACvB,eAAe;QACf,kBAAkB;QAClB,MAAM,EAAE,eAAe,IAAI,sCAAsC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;KACnF,CAAC;AACH,CAAC;AAED,SAAS,KAAK,CACb,KAAiB,EACjB,MAA6E;IAE7E,MAAM,KAAK,GAAU;QACpB,KAAK,EAAE,CAAC;QACR,MAAM,EAAE,CAAC;QACT,SAAS,EAAE,CAAC;QACZ,UAAU,EAAE,CAAC;QACb,WAAW,EAAE,CAAC;QACd,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;QACpE,GAAG,MAAM;KACT,CAAC;IACF,OAAO,aAAa,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC,KAAK,CAAC;AAC1C,CAAC","sourcesContent":["// Prompt cache warming policy shared by pi's hosts: when a refresh is due, whether a request can be replayed, and\n// whether a refresh pays for itself. Hosts own the scheduling; this module owns the numbers.\nimport { calculateCost } from \"./models.ts\";\nimport type { Api, Model, SimpleStreamOptions, Usage } from \"./types.ts\";\nimport { getProviderEnvValue } from \"./utils/provider-env.ts\";\n\n/** Whether a refresh is sent (\"warm\") or warming ends until the next real request (\"stop\"). */\nexport type CacheWarmingAction = \"warm\" | \"stop\";\n\n/** \"streaming\" while the agent run that sent the request is still active; \"idle\" after it settled. */\nexport type CacheWarmingPhase = \"streaming\" | \"idle\";\n\n/** Inputs and outcome of one warm-or-stop decision. */\nexport interface CacheWarmingDecision {\n\tphase: CacheWarmingPhase;\n\t/** Price of this refresh: a cache read of the prompt plus one output token. */\n\twarmCost: number;\n\t/** Extra price of the next real request if the cache entry is lost. */\n\tmissCost: number;\n\t/** Estimated chance that a real request arrives before the entry expires. */\n\tcontinuationProbability: number;\n\t/** `continuationProbability * missCost - warmCost`. */\n\texpectedSavings: number;\n\t/** False when the prompt size or the model's prices are unknown. */\n\teconomicsAvailable: boolean;\n\t/** \"warm\" when `expectedSavings` is at least `CACHE_WARMING_MINIMUM_EXPECTED_SAVINGS`. */\n\taction: CacheWarmingAction;\n}\n\n/** Streaming warming never continues past this long after the real request that started it. */\nexport const CACHE_WARMING_MAX_AGE_MS = 60 * 60_000;\n\n/** Idle warming uses a shorter horizon because continuation estimates become less reliable with age. */\nexport const CACHE_WARMING_MAX_IDLE_AGE_MS = 30 * 60_000;\n\n/** A refresh is sent only when it is expected to save at least this many dollars. */\nexport const CACHE_WARMING_MINIMUM_EXPECTED_SAVINGS = 0.05;\n\n/**\n * Chance that a real request arrives before the cache entry expires while the agent sits idle. Measured from our own\n * usage; per-session estimates were not better than this constant.\n */\nexport const CACHE_WARMING_IDLE_CONTINUATION_PROBABILITY = 0.15;\n\n/** Refresh at 90% of the TTL while preserving at least ten seconds of margin. */\nexport function getCacheWarmingDelayMs(ttlMs: number): number | undefined {\n\tif (ttlMs <= 10_000) return undefined;\n\treturn Math.max(1, Math.floor(Math.min(ttlMs * 0.9, ttlMs - 10_000)));\n}\n\n/**\n * Lifetime of the prompt cache entry a request writes, from the model's `promptCache` tier for the retention the\n * request used. Undefined when the model has no lifetime for that tier or caching is off.\n */\nexport function getPromptCacheTtlMs(model: Model<Api>, options: SimpleStreamOptions | undefined): number | undefined {\n\tconst retention =\n\t\toptions?.cacheRetention ??\n\t\t(getProviderEnvValue(\"PI_CACHE_RETENTION\", options?.env) === \"long\" ? \"long\" : \"short\");\n\tif (retention === \"none\") return undefined;\n\tconst seconds = model.promptCache?.[retention];\n\treturn seconds === undefined ? undefined : seconds * 1000;\n}\n\n/**\n * Whether replaying the request with a one-token output cap leaves its cache entry untouched. Anthropic's budget-based\n * thinking (Claude models without adaptive thinking) derives `budget_tokens` from `max_tokens`; the replay would get a\n * different budget, which Anthropic keys the message cache on, and the model could still think for thousands of tokens.\n */\nexport function isCacheWarmingReplayable(model: Model<Api>, options: SimpleStreamOptions | undefined): boolean {\n\tif (!options?.reasoning || model.api !== \"anthropic-messages\") return true;\n\treturn (model as Model<\"anthropic-messages\">).compat?.forceAdaptiveThinking === true;\n}\n\n/**\n * The options of a refresh: the request's own, capped at one output token, without retries, and without replacing the\n * conversation's provider thread state (the refresh repeats the last request).\n */\nexport function getCacheWarmingRequestOptions<T extends SimpleStreamOptions>(options: T, signal: AbortSignal): T {\n\treturn { ...options, maxTokens: 1, disableContinuation: true, maxRetries: 0, signal };\n}\n\n/** Decide whether refreshing the cache entry of a `promptTokens`-token request pays for itself in `phase`. */\nexport function evaluateCacheWarming(\n\tmodel: Model<Api>,\n\tpromptTokens: number,\n\tphase: CacheWarmingPhase,\n): CacheWarmingDecision {\n\tconst cacheHitCost = price(model, { cacheRead: promptTokens });\n\tconst cacheMissCost = price(\n\t\tmodel,\n\t\tmodel.cost.cacheWrite > 0 ? { cacheWrite: promptTokens } : { input: promptTokens },\n\t);\n\tconst warmCost = price(model, { cacheRead: promptTokens, output: 1 });\n\tconst missCost = Math.max(0, cacheMissCost - cacheHitCost);\n\tconst continuationProbability = phase === \"idle\" ? CACHE_WARMING_IDLE_CONTINUATION_PROBABILITY : 1;\n\tconst economicsAvailable = promptTokens > 0 && (cacheHitCost > 0 || cacheMissCost > 0);\n\tconst expectedSavings = continuationProbability * missCost - warmCost;\n\treturn {\n\t\tphase,\n\t\twarmCost,\n\t\tmissCost,\n\t\tcontinuationProbability,\n\t\texpectedSavings,\n\t\teconomicsAvailable,\n\t\taction: expectedSavings >= CACHE_WARMING_MINIMUM_EXPECTED_SAVINGS ? \"warm\" : \"stop\",\n\t};\n}\n\nfunction price(\n\tmodel: Model<Api>,\n\ttokens: Partial<Pick<Usage, \"input\" | \"output\" | \"cacheRead\" | \"cacheWrite\">>,\n): number {\n\tconst usage: Usage = {\n\t\tinput: 0,\n\t\toutput: 0,\n\t\tcacheRead: 0,\n\t\tcacheWrite: 0,\n\t\ttotalTokens: 0,\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t...tokens,\n\t};\n\treturn calculateCost(model, usage).total;\n}\n"]}

package/dist/env-api-keys.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"env-api-keys.d.ts","sourceRoot":"","sources":["../src/env-api-keys.ts"],"names":[],"mappings":"AAyBA,OAAO,KAAK,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAG7D,eAAO,MAAM,wBAAwB,yBAAyB,CAAC;AAC/D,eAAO,MAAM,yBAAyB,0BAA0B,CAAC;AACjE,eAAO,MAAM,qBAAqB,sBAAsB,CAAC;AACzD,eAAO,MAAM,gCAAgC,iCAAiC,CAAC;AAC/E,eAAO,MAAM,6BAA6B,8BAA8B,CAAC;AACzE,eAAO,MAAM,gCAAgC,iCAAiC,CAAC;AAC/E,eAAO,MAAM,iCAAiC,kCAAkC,CAAC;AACjF,eAAO,MAAM,0BAA0B,2BAA2B,CAAC;AA6FnE;;;;;;GAMG;AACH,wBAAgB,WAAW,CAAC,QAAQ,EAAE,aAAa,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,EAAE,GAAG,SAAS,CAAC;AAC9F,wBAAgB,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,EAAE,GAAG,SAAS,CAAC;AASvF;;;;GAIG;AACH,wBAAgB,YAAY,CAAC,QAAQ,EAAE,aAAa,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,GAAG,SAAS,CAAC;AAC7F,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,GAAG,SAAS,CAAC"}
+{"version":3,"file":"env-api-keys.d.ts","sourceRoot":"","sources":["../src/env-api-keys.ts"],"names":[],"mappings":"AAyBA,OAAO,KAAK,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAG7D,eAAO,MAAM,wBAAwB,yBAAyB,CAAC;AAC/D,eAAO,MAAM,yBAAyB,0BAA0B,CAAC;AACjE,eAAO,MAAM,qBAAqB,sBAAsB,CAAC;AACzD,eAAO,MAAM,gCAAgC,iCAAiC,CAAC;AAC/E,eAAO,MAAM,6BAA6B,8BAA8B,CAAC;AACzE,eAAO,MAAM,gCAAgC,iCAAiC,CAAC;AAC/E,eAAO,MAAM,iCAAiC,kCAAkC,CAAC;AACjF,eAAO,MAAM,0BAA0B,2BAA2B,CAAC;AA8FnE;;;;;;GAMG;AACH,wBAAgB,WAAW,CAAC,QAAQ,EAAE,aAAa,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,EAAE,GAAG,SAAS,CAAC;AAC9F,wBAAgB,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,EAAE,GAAG,SAAS,CAAC;AASvF;;;;GAIG;AACH,wBAAgB,YAAY,CAAC,QAAQ,EAAE,aAAa,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,GAAG,SAAS,CAAC;AAC7F,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,GAAG,SAAS,CAAC"}