@absolutejs/ai 0.2.5 → 0.2.7

package/dist/ai/index.js

@@ -40,6 +40,7 @@ var inspectAIInput = async (provider, params) => {
   return {
     limits,
     inputTokens,
+    inputTokenScope: provider.inputCapacity.countScope?.(params) ?? "request",
     outputTokens,
     availableInputTokens,
     fits: outputTokens <= maxOutputTokens && inputTokens <= availableInputTokens
@@ -511,7 +512,7 @@ var withResilience = (provider, providerName = "unknown") => {
       noteSuccess(providerName);
     } catch (err) {
       const providerError = ProviderError.from(err, providerName);
-      const canRetry = !yielded && providerError.retryable && attemptNo < config.maxRetries && !params.signal?.aborted;
+      const canRetry = !yielded && providerError.retryable && attemptNo < (params.maxRetries === undefined ? config.maxRetries : Math.max(0, Math.min(config.maxRetries, params.maxRetries))) && !params.signal?.aborted;
       if (canRetry) {
         await sleep(backoffDelay(attemptNo), params.signal);
         yield* attempt(params, attemptNo + 1);
@@ -2886,6 +2887,35 @@ var anthropic = (config2) => {
       signal: params.signal ? AbortSignal.any([params.signal, AbortSignal.timeout(15000)]) : AbortSignal.timeout(15000)
     }));
   };
+  const countingRequest = (params) => {
+    const built = buildRequestBody3(params, configuredMax, promptCaching);
+    const body = config2.transformRequestBody ? config2.transformRequestBody(built, params) : built;
+    const counted = {};
+    for (const key of [
+      "model",
+      "messages",
+      "system",
+      "tools",
+      "tool_choice",
+      "thinking"
+    ])
+      if (body[key] !== undefined)
+        counted[key] = body[key];
+    const tools = Array.isArray(body.tools) ? body.tools : [];
+    const hosted = tools.filter((tool) => isRecord4(tool) && typeof tool.type === "string" && /^(web_search_|web_fetch_|code_execution_|bash_code_execution_|text_editor_code_execution_|tool_search_tool_)/u.test(tool.type));
+    const scope = hosted.length > 0 || body.mcp_servers !== undefined ? "caller-input" : "request";
+    if (hosted.length > 0) {
+      const clientTools = tools.filter((tool) => !hosted.includes(tool));
+      if (clientTools.length)
+        counted.tools = clientTools;
+      else
+        delete counted.tools;
+      const choice = counted.tool_choice;
+      if (!clientTools.length || isRecord4(choice) && choice.type === "tool" && hosted.some((tool) => isRecord4(tool) && tool.name === choice.name))
+        delete counted.tool_choice;
+    }
+    return { counted, scope };
+  };
   const inputCapacity = {
     outputTokens: (params) => {
       const built = buildRequestBody3(params, configuredMax, promptCaching);
@@ -2901,20 +2931,9 @@ var anthropic = (config2) => {
         maxOutputTokens: positiveTokenLimit(data.max_tokens)
       };
     }, config2.tokenSource || typeof config2.headers === "function" ? 0 : 300000),
+    countScope: (params) => countingRequest(params).scope,
     countTokens: async (params) => {
-      const built = buildRequestBody3(params, configuredMax, promptCaching);
-      const body = config2.transformRequestBody ? config2.transformRequestBody(built, params) : built;
-      const counted = {};
-      for (const key of [
-        "model",
-        "messages",
-        "system",
-        "tools",
-        "tool_choice",
-        "thinking"
-      ])
-        if (body[key] !== undefined)
-          counted[key] = body[key];
+      const { counted } = countingRequest(params);
       return inputTokenCount((await capacityRequest(params, "/v1/messages/count_tokens", counted)).input_tokens);
     }
   };
@@ -5631,6 +5650,7 @@ var generateAI = async (options) => {
   const stream = streamWithAIContext(options.provider, {
     cacheSystemPrompt: options.cacheSystemPrompt,
     maxTokens: options.maxTokens,
+    maxRetries: options.maxRetries,
     messages: options.messages,
     model: options.model,
     promptCaching: options.promptCaching,
@@ -5835,6 +5855,7 @@ var generateObjectAI = async (options) => {
       contextPolicy: options.contextPolicy,
       cacheSystemPrompt: options.cacheSystemPrompt,
       maxTokens: options.maxTokens,
+      maxRetries: options.maxRetries,
       messages,
       model: options.model,
       promptCaching: options.promptCaching,
@@ -8765,6 +8786,7 @@ var createAIProviderRouter = (routes) => {
       outputTokens: (params) => capacity(params).outputTokens?.(params),
       isContextError: (error) => [...new Set(providers.values())].some((provider) => provider.inputCapacity?.isContextError?.(error) === true),
       getLimits: async (params) => capacity(params).getLimits(params),
+      countScope: (params) => capacity(params).countScope?.(params) ?? "request",
       countTokens: async (params) => capacity(params).countTokens(params)
     },
     stream: (params) => resolve(params).stream(params)
@@ -8880,5 +8902,5 @@ export {
   xai
 };
 
-//# debugId=3B9702E9B461551864756E2164756E21
+//# debugId=F8FFC7094E06DA7B64756E2164756E21
 //# sourceMappingURL=index.js.map