@theokit/sdk 4.54.0 → 4.55.0

package/dist/{chunk-UNDROG5N.cjs → chunk-HLPSLDGL.cjs}

@@ -7,7 +7,7 @@ var chunk6JSIEDKL_cjs = require('./chunk-6JSIEDKL.cjs');
 var chunkFRITRLZQ_cjs = require('./chunk-FRITRLZQ.cjs');
 var chunkJHPGF3FP_cjs = require('./chunk-JHPGF3FP.cjs');
 var chunkXSX2UU6Y_cjs = require('./chunk-XSX2UU6Y.cjs');
-var chunkRM7Y65IG_cjs = require('./chunk-RM7Y65IG.cjs');
+var chunkNIRE5CNO_cjs = require('./chunk-NIRE5CNO.cjs');
 var chunkK3FW2XZD_cjs = require('./chunk-K3FW2XZD.cjs');
 var chunkNUKRL3I6_cjs = require('./chunk-NUKRL3I6.cjs');
 var fs = require('fs');
@@ -18,7 +18,7 @@ var url = require('url');
 var module$1 = require('module');
 
 var _documentCurrentScript = typeof document !== 'undefined' ? document.currentScript : null;
-var discoveryState = chunkRM7Y65IG_cjs.globalSingleton("theokit-sdk.providers.discovered", () => ({
+var discoveryState = chunkNIRE5CNO_cjs.globalSingleton("theokit-sdk.providers.discovered", () => ({
   done: false
 }));
 function pluginsRoot() {
@@ -97,7 +97,7 @@ async function loadOne(dir, entryName) {
       if (plugin !== null && typeof plugin === "object" && plugin.kind === "model-provider") {
         const profile = plugin.profile;
         if (profile !== void 0 && typeof profile === "object") {
-          chunkRM7Y65IG_cjs.registerProvider(profile);
+          chunkNIRE5CNO_cjs.registerProvider(profile);
         }
       }
       return;
@@ -757,7 +757,7 @@ async function resolveBedrockToken(region) {
     return cachedToken.value;
   }
   try {
-    const r = module$1.createRequire((typeof document === 'undefined' ? require('u' + 'rl').pathToFileURL(__filename).href : (_documentCurrentScript && _documentCurrentScript.tagName.toUpperCase() === 'SCRIPT' && _documentCurrentScript.src || new URL('chunk-UNDROG5N.cjs', document.baseURI).href)));
+    const r = module$1.createRequire((typeof document === 'undefined' ? require('u' + 'rl').pathToFileURL(__filename).href : (_documentCurrentScript && _documentCurrentScript.tagName.toUpperCase() === 'SCRIPT' && _documentCurrentScript.src || new URL('chunk-HLPSLDGL.cjs', document.baseURI).href)));
     const gen = r("@aws/bedrock-token-generator");
     if (typeof gen.getToken !== "function") return void 0;
     const providers = r("@aws-sdk/credential-providers");
@@ -791,10 +791,10 @@ var BedrockAnthropicClient = class {
     );
   }
   async *stream(request, signal) {
-    const region = chunkRM7Y65IG_cjs.inferRegionFromModelId(request.model) ?? process.env.AWS_REGION ?? "us-east-1";
+    const region = chunkNIRE5CNO_cjs.inferRegionFromModelId(request.model) ?? process.env.AWS_REGION ?? "us-east-1";
     const token = await this.resolveTokenOrThrow(region);
-    const bareModel = chunkRM7Y65IG_cjs.stripBedrockPrefix(request.model);
-    const baseUrl = this.options.baseUrl ?? chunkRM7Y65IG_cjs.resolveBedrockBaseUrl(request.model);
+    const bareModel = chunkNIRE5CNO_cjs.stripBedrockPrefix(request.model);
+    const baseUrl = this.options.baseUrl ?? chunkNIRE5CNO_cjs.resolveBedrockBaseUrl(request.model);
     const endpoint = `/model/${encodeURIComponent(bareModel)}/invoke`;
     const url = `${baseUrl}${endpoint}`;
     const response = await postAnthropicRequest({
@@ -2076,7 +2076,7 @@ function abortError(signal) {
 }
 
 // src/internal/llm/responses.ts
-function messageToInputItems(message) {
+function messageToInputItems(message, reasoningByCallId) {
   const items = [];
   if (message.role === "user") {
     const content = [];
@@ -2103,6 +2103,14 @@ function messageToInputItems(message) {
       if (part.type === "text") {
         content.push({ type: "output_text", text: part.text });
       } else if (part.type === "tool_use") {
+        for (const reasoning of reasoningByCallId?.get(part.id) ?? []) {
+          items.push({
+            type: "reasoning",
+            id: reasoning.id,
+            encrypted_content: reasoning.encrypted_content,
+            summary: []
+          });
+        }
         items.push({
           type: "function_call",
           call_id: part.id,
@@ -2118,10 +2126,10 @@ function messageToInputItems(message) {
   if (text.length > 0) items.push({ role: "system", content: text });
   return items;
 }
-function buildResponsesBody(request) {
+function buildResponsesBody(request, options = {}) {
   const input = [];
   for (const message of request.messages) {
-    for (const item of messageToInputItems(message)) input.push(item);
+    for (const item of messageToInputItems(message, options.reasoningByCallId)) input.push(item);
   }
   const slash = request.model.lastIndexOf("/");
   const model = slash >= 0 ? request.model.slice(slash + 1) : request.model;
@@ -2138,9 +2146,21 @@ function buildResponsesBody(request) {
     strict: false
   }));
   if (tools.length > 0) body.tools = tools;
-  if (request.reasoning !== void 0) body.reasoning = { effort: request.reasoning.effort };
+  applyPromptCachingFields(body, request, options);
   return body;
 }
+function applyPromptCachingFields(body, request, options) {
+  const carryReasoning = options.encryptedReasoning === true;
+  if (request.reasoning !== void 0) {
+    body.reasoning = {
+      effort: request.reasoning.effort,
+      ...carryReasoning ? { context: "all_turns" } : {}
+    };
+  }
+  if (carryReasoning) body.include = ["reasoning.encrypted_content"];
+  if (request.promptCacheKey !== void 0) body.prompt_cache_key = request.promptCacheKey;
+}
+var MAX_REMEMBERED_REASONING_CALLS = 256;
 var ResponsesApiClient = class {
   constructor(options) {
     this.options = options;
@@ -2152,6 +2172,26 @@ var ResponsesApiClient = class {
   name;
   baseUrl;
   fetchImpl;
+  /**
+   * usetheokit/theokit-sdk#383 — reasoning ciphertext captured from earlier rounds, keyed by the
+   * `call_id` it preceded. Held on the CLIENT rather than threaded through `LlmRequest` because a
+   * round's reasoning is provider-shaped ciphertext with a provider-shaped ordering rule; the agent
+   * loop has no use for it and the same client instance serves every round of a turn. Insertion
+   * order is the eviction order (see {@link MAX_REMEMBERED_REASONING_CALLS}).
+   */
+  reasoningByCallId = /* @__PURE__ */ new Map();
+  /**
+   * usetheokit/theokit-sdk#383 — bind a batch of reasoning ciphertext to the tool call it produced,
+   * evicting the oldest binding once the cap is reached.
+   */
+  rememberReasoning(callId, items) {
+    this.reasoningByCallId.set(callId, [...items]);
+    while (this.reasoningByCallId.size > MAX_REMEMBERED_REASONING_CALLS) {
+      const oldest = this.reasoningByCallId.keys().next();
+      if (oldest.done === true) break;
+      this.reasoningByCallId.delete(oldest.value);
+    }
+  }
   // biome-ignore lint/complexity/noExcessiveCognitiveComplexity: the SSE dispatch (text / reasoning / tool-call add+delta+done / terminal+usage / error) is one cohesive state machine, mirroring OpenAIStreamAccumulator.consume.
   async *stream(request, signal) {
     const providerId = this.options.providerName ?? "openai";
@@ -2166,7 +2206,15 @@ var ResponsesApiClient = class {
       method: "POST",
       signal,
       headers,
-      body: JSON.stringify(buildResponsesBody(request))
+      body: JSON.stringify(
+        // Both options ride the SAME flag on purpose: a provider that never declared support gets
+        // the byte-identical pre-#383 body, even in the impossible case where it returned ciphertext
+        // nobody asked for.
+        buildResponsesBody(
+          request,
+          this.options.encryptedReasoning === true ? { encryptedReasoning: true, reasoningByCallId: this.reasoningByCallId } : {}
+        )
+      )
     });
     if (!response.ok) {
       const text2 = await response.text().catch(() => "");
@@ -2189,7 +2237,10 @@ var ResponsesApiClient = class {
     let inputTokens;
     let outputTokens;
     let reasoningTokens;
+    let cacheReadTokens;
+    let cacheWriteTokens;
     const pending = {};
+    let unclaimedReasoning = [];
     if (response.body !== null) {
       for await (const record of parseSseStream(response.body, signal)) {
         if (record.data === "[DONE]") break;
@@ -2209,6 +2260,12 @@ var ResponsesApiClient = class {
         } else if (t === "response.reasoning_summary_text.delta" || t === "response.reasoning_text.delta") {
           const d = event.delta ?? "";
           if (d.length > 0) yield { type: "reasoning_delta", text: d };
+        } else if (t === "response.output_item.done" && event.item?.type === "reasoning") {
+          const id = event.item.id;
+          const encrypted = event.item.encrypted_content;
+          if (id !== void 0 && encrypted !== void 0 && encrypted.length > 0) {
+            unclaimedReasoning.push({ id, encrypted_content: encrypted });
+          }
         } else if (t === "response.output_item.added" && event.item?.type === "function_call") {
           const id = event.item.id ?? event.item.call_id ?? "call-0";
           pending[id] = {
@@ -2234,6 +2291,10 @@ var ResponsesApiClient = class {
             input: parseToolArguments(rawArgs)
           };
           toolCalls.push(call);
+          if (unclaimedReasoning.length > 0) {
+            this.rememberReasoning(call.id, unclaimedReasoning);
+            unclaimedReasoning = [];
+          }
           delete pending[id];
         } else if (t === "response.completed" || t === "response.incomplete") {
           const usage = event.response?.usage;
@@ -2241,6 +2302,8 @@ var ResponsesApiClient = class {
             inputTokens = usage.input_tokens;
             outputTokens = usage.output_tokens;
             reasoningTokens = usage.output_tokens_details?.reasoning_tokens;
+            cacheReadTokens = usage.input_tokens_details?.cached_tokens;
+            cacheWriteTokens = usage.input_tokens_details?.cache_write_tokens;
           }
           stopReason = t === "response.incomplete" ? "max_tokens" : "end_turn";
         } else if (t === "response.failed" || t === "error") {
@@ -2263,7 +2326,9 @@ var ResponsesApiClient = class {
       toolCalls,
       inputTokens,
       outputTokens,
-      reasoningTokens
+      reasoningTokens,
+      cacheReadTokens,
+      cacheWriteTokens
     });
   }
 };
@@ -2439,7 +2504,7 @@ async function resolveVertexAccessToken() {
 async function getAuthClientOrThrow() {
   let mod;
   try {
-    const r = module$1.createRequire((typeof document === 'undefined' ? require('u' + 'rl').pathToFileURL(__filename).href : (_documentCurrentScript && _documentCurrentScript.tagName.toUpperCase() === 'SCRIPT' && _documentCurrentScript.src || new URL('chunk-UNDROG5N.cjs', document.baseURI).href)));
+    const r = module$1.createRequire((typeof document === 'undefined' ? require('u' + 'rl').pathToFileURL(__filename).href : (_documentCurrentScript && _documentCurrentScript.tagName.toUpperCase() === 'SCRIPT' && _documentCurrentScript.src || new URL('chunk-HLPSLDGL.cjs', document.baseURI).href)));
     mod = r("google-auth-library");
   } catch (err) {
     const isModuleMissing = err instanceof Error && (err.message.includes("Cannot find module") || err.code === "MODULE_NOT_FOUND");
@@ -2500,8 +2565,8 @@ var VertexAnthropicClient = class {
     const projectId = this.resolveProjectOrThrow();
     const location = this.options.location ?? resolveVertexLocation();
     const accessToken = await this.resolveTokenOrThrow();
-    const baseUrl = chunkRM7Y65IG_cjs.resolveVertexBaseUrl({ projectId, location, modelDialect: "anthropic" });
-    const bareModel = chunkRM7Y65IG_cjs.stripVertexPrefix(request.model);
+    const baseUrl = chunkNIRE5CNO_cjs.resolveVertexBaseUrl({ projectId, location, modelDialect: "anthropic" });
+    const bareModel = chunkNIRE5CNO_cjs.stripVertexPrefix(request.model);
     const endpoint = `/models/${encodeURIComponent(bareModel)}:rawPredict`;
     const url = `${baseUrl}${endpoint}`;
     const response = await postAnthropicRequest({
@@ -2550,7 +2615,7 @@ var VertexGeminiClient = class {
         { code: "auth_failed" }
       );
     }
-    const baseUrl = chunkRM7Y65IG_cjs.resolveVertexBaseUrl({ projectId, location, modelDialect: "gemini" });
+    const baseUrl = chunkNIRE5CNO_cjs.resolveVertexBaseUrl({ projectId, location, modelDialect: "gemini" });
     const url = `${baseUrl}/chat/completions`;
     const innerFetch = this.options.fetch ?? fetch;
     const rewriteFetch = (_input, init) => innerFetch(url, init);
@@ -2572,7 +2637,7 @@ var VertexRouterClient = class {
   options;
   name = "vertex";
   async *stream(request, signal) {
-    const dialect = chunkRM7Y65IG_cjs.inferModelDialect(request.model);
+    const dialect = chunkNIRE5CNO_cjs.inferModelDialect(request.model);
     if (dialect === "anthropic") {
       const inner2 = new VertexAnthropicClient(this.options);
       return yield* inner2.stream(request, signal);
@@ -2584,7 +2649,7 @@ var VertexRouterClient = class {
 
 // src/internal/llm/router.ts
 function resolveProviderChain(options) {
-  chunkRM7Y65IG_cjs.registerBuiltins();
+  chunkNIRE5CNO_cjs.registerBuiltins();
   return buildChain(options);
 }
 function buildChain(options) {
@@ -2609,7 +2674,7 @@ function buildChain(options) {
   return clients;
 }
 function buildClient(name, routerOptions, perCallBaseUrl) {
-  const baseProfile = chunkRM7Y65IG_cjs.getProviderProfile(name);
+  const baseProfile = chunkNIRE5CNO_cjs.getProviderProfile(name);
   if (baseProfile === void 0) return void 0;
   const profile = routerOptions.extractToolCallsFromContent === true && baseProfile.extractToolCallsFromContent !== true ? { ...baseProfile, extractToolCallsFromContent: true } : baseProfile;
   const resilience = routerOptions.onRateLimit !== void 0 ? { onRateLimit: routerOptions.onRateLimit } : {};
@@ -2704,7 +2769,7 @@ function warnUnknownProvidersInApiKeys(apiKeys) {
   if (apiKeys === void 0) return;
   for (const name of Object.keys(apiKeys)) {
     if (warnedProviders.has(name)) continue;
-    const profile = chunkRM7Y65IG_cjs.getProviderProfile(name);
+    const profile = chunkNIRE5CNO_cjs.getProviderProfile(name);
     if (profile === void 0) {
       warnedProviders.add(name);
       chunkNUKRL3I6_cjs.diag(
@@ -2792,6 +2857,9 @@ function selectTransport(profile, apiKey, baseUrl) {
       ...(baseUrl ?? profile.baseUrl) !== void 0 ? { baseUrl: baseUrl ?? profile.baseUrl } : {},
       ...mergedHeaders !== void 0 ? { extraHeaders: mergedHeaders } : {},
       ...t.fetch !== void 0 ? { fetch: t.fetch } : {},
+      // usetheokit/theokit-sdk#383 — forwarded only when the profile declares it, so a provider that
+      // has not been measured against `reasoning.context` keeps the request it had.
+      ...profile.encryptedReasoning === true ? { encryptedReasoning: true } : {},
       providerName: profile.name
     });
   }
@@ -2805,7 +2873,7 @@ function selectTransport(profile, apiKey, baseUrl) {
 function inferProviderFromApiKey(apiKey) {
   const provider = chunkFRITRLZQ_cjs.providerFromApiKeyPrefix(apiKey);
   if (provider === void 0) return void 0;
-  return chunkRM7Y65IG_cjs.getProviderProfile(provider) === void 0 ? void 0 : provider;
+  return chunkNIRE5CNO_cjs.getProviderProfile(provider) === void 0 ? void 0 : provider;
 }
 function detectPrimaryProvider() {
   if (process.env.ANTHROPIC_API_KEY !== void 0 && process.env.ANTHROPIC_API_KEY.length > 0) {
@@ -3000,9 +3068,9 @@ function resolveSummarizerRoute(opts) {
 }
 function buildDefaultSummarizer(opts) {
   return async (messages) => {
-    chunkRM7Y65IG_cjs.registerBuiltins();
+    chunkNIRE5CNO_cjs.registerBuiltins();
     const modelPrefix = opts.agentModel.includes("/") ? opts.agentModel.slice(0, opts.agentModel.indexOf("/")) : void 0;
-    const prefixProfile = modelPrefix !== void 0 ? chunkRM7Y65IG_cjs.getProviderProfile(modelPrefix) : void 0;
+    const prefixProfile = modelPrefix !== void 0 ? chunkNIRE5CNO_cjs.getProviderProfile(modelPrefix) : void 0;
     const prefixUsable = prefixProfile !== void 0 && (prefixProfile.authType !== "api_key" || prefixProfile.envVars.some((v) => (process.env[v] ?? "").length > 0));
     const route = resolveSummarizerRoute({
       keyProvider: inferProviderFromApiKey(opts.apiKey),
@@ -3106,5 +3174,5 @@ exports.shouldUseFixtureMode = shouldUseFixtureMode;
 exports.shouldUseRealLocalRuntime = shouldUseRealLocalRuntime;
 exports.tryFirstEvent = tryFirstEvent;
 exports.warnProviderPrecedenceOnce = warnProviderPrecedenceOnce;
-//# sourceMappingURL=chunk-UNDROG5N.cjs.map
-//# sourceMappingURL=chunk-UNDROG5N.cjs.map
+//# sourceMappingURL=chunk-HLPSLDGL.cjs.map
+//# sourceMappingURL=chunk-HLPSLDGL.cjs.map