@kenkaiiii/gg-ai 5.89.0 → 5.91.0

package/dist/index.cjs

@@ -35,6 +35,7 @@ __export(index_exports, {
   EventStream: () => EventStream,
   FAST_MODE_SUFFIX: () => FAST_MODE_SUFFIX,
   GGAIError: () => GGAIError,
+  LOCAL_NO_API_KEY: () => LOCAL_NO_API_KEY,
   PRECONNECT_INTERVAL_MS: () => PRECONNECT_INTERVAL_MS,
   ProviderError: () => ProviderError,
   REDACTION_MARKER: () => REDACTED,
@@ -57,6 +58,7 @@ __export(index_exports, {
   isHardBillingMessage: () => isHardBillingMessage,
   isUsageLimitError: () => isUsageLimitError,
   isValidToolCallName: () => isValidToolCallName,
+  localBearerKey: () => localBearerKey,
   localWireModelId: () => localWireModelId,
   opensTurn: () => opensTurn,
   palsuAssistantMessage: () => palsuAssistantMessage,
@@ -333,6 +335,8 @@ function formatError(err) {
         guidance: fastRejection.guidance
       };
     }
+    const miniMax = err.provider === "minimax" ? formatMiniMaxError(err, cleanMessage) : void 0;
+    if (miniMax) return miniMax;
     if (isUsageLimitError(err)) {
       const resetClause = err.resetsAt ? ` It resets at ${formatResetTime(err.resetsAt)}.` : "";
       return {
@@ -439,6 +443,38 @@ function inferSource(err) {
   }
   return "ggcoder";
 }
+function formatMiniMaxError(err, cleanMessage) {
+  const code = /\((\d{4,5})\)/.exec(err.message)?.[1];
+  const base = {
+    source: "provider",
+    message: cleanMessage,
+    provider: err.provider,
+    statusCode: err.statusCode,
+    ...err.requestId ? { requestId: err.requestId } : {}
+  };
+  if (code === "2049" || code === "1004") {
+    return {
+      ...base,
+      headline: "MiniMax didn't accept this API key.",
+      guidance: "A MiniMax key only works in the region that issued it (Global: platform.minimax.io, Mainland China: platform.minimaxi.com). Connect MiniMax again with the key so GG Coder can detect its region, and check the key is active."
+    };
+  }
+  if (code === "2061") {
+    return {
+      ...base,
+      headline: "Your MiniMax plan doesn't include this model.",
+      guidance: "Switch to MiniMax M3.1 Flash in the model selector \u2014 it's the model M Plan and Token Plan cover. MiniMax M3 needs a pay-as-you-go key."
+    };
+  }
+  if (code === "1008") {
+    return {
+      ...base,
+      headline: "MiniMax has no balance for this request.",
+      guidance: "Check this key's plan is active, or its pay-as-you-go balance, at platform.minimax.io (platform.minimaxi.com for Mainland China keys). Plan keys only cover MiniMax M3.1 Flash. Not a GG Coder issue."
+    };
+  }
+  return void 0;
+}
 function providerGuidance(provider, message, statusCode) {
   const name = provider ? providerDisplayName(provider) : "the provider";
   const status = provider ? PROVIDER_STATUS_URL[provider] : void 0;
@@ -1656,10 +1692,22 @@ function toAnthropicMessages(messages, cacheControl, options) {
           (name) => toolReferences.has(name) && !referencedTools.has(name)
         ) : [];
         if (refs.length === 0) {
+          const content = toAnthropicToolResultContent(result.content);
+          if (result.isError && typeof content !== "string" && content.some((part) => part.type !== "text")) {
+            const text = content.filter((part) => part.type === "text" && part.text !== "");
+            trailing.push({ type: "text", text: `Attachments from tool call ${toolUseId}:` });
+            for (const part of content) if (part.type !== "text") trailing.push(part);
+            return {
+              type: "tool_result",
+              tool_use_id: toolUseId,
+              content: text.length > 0 ? text : "Tool call failed; its attachments follow.",
+              is_error: true
+            };
+          }
           return {
             type: "tool_result",
             tool_use_id: toolUseId,
-            content: toAnthropicToolResultContent(result.content),
+            content,
             is_error: result.isError
           };
         }
@@ -1678,7 +1726,8 @@ ${part.text}` } : part
           type: "tool_result",
           tool_use_id: toolUseId,
           content: refs.map((name) => ({ type: "tool_reference", tool_name: name })),
-          is_error: result.isError
+          // tool_reference isn't text, so is_error would 400; the error text trails.
+          is_error: void 0
         };
       });
       out.push({
@@ -1701,8 +1750,8 @@ ${part.text}` } : part
           markAnthropicCacheBreakpoint(out, tail, cacheControl);
         } else {
           if (anchor !== void 0) markAnthropicCacheBreakpoint(out, anchor, cacheControl);
-          const longRun = isLongToolRun(messages, trajectoryStartIdx);
-          markAnthropicCacheBreakpoint(out, tail, longRun ? cacheControl : { type: "ephemeral" });
+          const blocking = calledBlockingTool(messages, trajectoryStartIdx);
+          markAnthropicCacheBreakpoint(out, tail, blocking ? cacheControl : { type: "ephemeral" });
         }
       } else {
         markAnthropicCacheBreakpoint(out, tail, cacheControl);
@@ -1738,19 +1787,13 @@ var BLOCKING_TOOL_NAMES = /* @__PURE__ */ new Set([
   "subagent",
   "ask_user"
 ]);
-var LONG_RUN_ROUNDS = 6;
-function isLongToolRun(messages, trajectoryStartIdx) {
-  let rounds = 0;
+function calledBlockingTool(messages, trajectoryStartIdx) {
   for (let i = Math.max(0, trajectoryStartIdx + 1); i < messages.length; i++) {
     const m = messages[i];
     if (m?.role !== "assistant" || typeof m.content === "string") continue;
-    let called = false;
     for (const part of m.content) {
-      if (part.type !== "tool_call") continue;
-      called = true;
-      if (BLOCKING_TOOL_NAMES.has(part.name)) return true;
+      if (part.type === "tool_call" && BLOCKING_TOOL_NAMES.has(part.name)) return true;
     }
-    if (called && ++rounds >= LONG_RUN_ROUNDS) return true;
   }
   return false;
 }
@@ -1809,6 +1852,13 @@ function isAdaptiveThinkingModel(model) {
   return /opus-5|opus-4[-.]8|opus-4[-.]7|opus-4[-.]6|sonnet-5|haiku-5|fable-5|mythos-5/.test(model);
 }
 function toAnthropicThinking(level, maxTokens, model) {
+  if (model.startsWith("MiniMax-M3.1")) {
+    return {
+      thinking: { type: "adaptive" },
+      maxTokens,
+      outputConfig: { effort: level === "ultra" ? "max" : level }
+    };
+  }
   if (isAdaptiveThinkingModel(model)) {
     let effort = level;
     if (effort === "xhigh" && !/fable-5|mythos-5|opus-5|opus-4-8|opus-4-7|sonnet-5[-.]5|haiku-5[-.]5/.test(model)) {
@@ -2330,8 +2380,7 @@ async function* runStream(options) {
     ...system ? { system } : {},
     ...thinking ? { thinking } : {},
     ...outputConfig ? { output_config: outputConfig } : {},
-    ...options.temperature != null && !thinking ? { temperature: options.temperature } : {},
-    ...options.topP != null ? { top_p: options.topP } : {},
+    // API version 2026-10-21 (SDK 1.x) rejects temperature/top_p/top_k with a 400.
     ...options.stop ? { stop_sequences: options.stop } : {},
     ...(deferredPlan?.live ?? options.tools)?.length || options.serverTools?.length || options.webSearch ? (() => {
       const reservedServerNames = /* @__PURE__ */ new Set();
@@ -2432,7 +2481,7 @@ async function* runStream(options) {
       yield* synthesizeEventsFromMessage(message);
       return messageToResponse(message);
     } catch (err) {
-      throw toError(err, { fastMode: useFastMode });
+      throw toError(err, { fastMode: useFastMode, provider: options.provider });
     }
   }
   const contentParts = [];
@@ -2548,7 +2597,7 @@ async function* runStream(options) {
                 const rawPartial = accum.argsJson;
                 const snippet = rawPartial.length > 200 ? `${rawPartial.slice(0, 200)}\u2026` : rawPartial;
                 throw new ProviderError(
-                  "anthropic",
+                  options.provider,
                   `Tool "${accum.toolName}" input JSON was truncated in the stream (${rawPartial.length} bytes): ${snippet}; ${parseErr.message}`,
                   { cause: parseErr }
                 );
@@ -2651,15 +2700,15 @@ async function* runStream(options) {
       }
     }
   } catch (err) {
-    throw toError(err, { fastMode: useFastMode });
+    throw toError(err, { fastMode: useFastMode, provider: options.provider });
   }
   if (!receivedAnyEvent) {
-    throw new ProviderError("anthropic", "Stream ended without producing any events.", {
+    throw new ProviderError(options.provider, "Stream ended without producing any events.", {
       statusCode: 504
     });
   }
   if (stopReason === null) {
-    throw new ProviderError("anthropic", "Stream ended before completion (no stop_reason).", {
+    throw new ProviderError(options.provider, "Stream ended before completion (no stop_reason).", {
       statusCode: 504,
       cause: { partialContent: contentParts, outputTokens }
     });
@@ -2801,6 +2850,7 @@ function readUnifiedRateLimit(headers) {
   return { rejected: status === "rejected", ...resetsAt ? { resetsAt } : {} };
 }
 function toError(err, context) {
+  const provider = context?.provider ?? "anthropic";
   if (err instanceof ProviderError) return err;
   if (err instanceof import_sdk.default.APIError) {
     const errorBody = err.error;
@@ -2834,22 +2884,22 @@ function toError(err, context) {
     }
     if (isHardBillingMessage(message)) {
       const usageMessage = /usage limit reached/i.test(message) ? message : `usage limit reached: ${message}`;
-      return new ProviderError("anthropic", usageMessage, {
+      return new ProviderError(provider, usageMessage, {
         statusCode: err.status,
         ...requestId ? { requestId } : {},
         cause: err
       });
     }
-    return new ProviderError("anthropic", message, {
+    return new ProviderError(provider, message, {
       statusCode: err.status,
       ...requestId ? { requestId } : {},
       cause: err
     });
   }
   if (err instanceof Error) {
-    return new ProviderError("anthropic", err.message, { cause: err });
+    return new ProviderError(provider, err.message, { cause: err });
   }
-  return new ProviderError("anthropic", String(err));
+  return new ProviderError(provider, String(err));
 }
 
 // src/providers/openai.ts
@@ -2937,6 +2987,12 @@ function getEnvironment() {
   return globalThis.process?.env;
 }
 
+// src/local-api-key.ts
+var LOCAL_NO_API_KEY = "local";
+function localBearerKey(apiKey) {
+  return apiKey && apiKey !== LOCAL_NO_API_KEY ? apiKey : void 0;
+}
+
 // src/providers/openai.ts
 function toKimiK3Effort(level) {
   switch (level) {
@@ -2981,6 +3037,8 @@ function createClient2(options) {
     const cached = openaiClientCache.get(cacheKey);
     if (cached) return cached;
   }
+  const omitAuth = options.provider === "local" && localBearerKey(options.apiKey) === void 0;
+  const defaultHeaders = omitAuth ? { ...options.defaultHeaders, Authorization: null } : options.defaultHeaders;
   const client = new import_openai.default({
     apiKey: options.apiKey,
     ...options.baseUrl ? { baseURL: options.baseUrl } : {},
@@ -2990,7 +3048,7 @@ function createClient2(options) {
     // events). The SDK default of 2 silent retries stacked under it: one 5xx
     // cost up to 3 hidden requests plus SDK sleeps before the loop even saw it.
     maxRetries: 0,
-    ...options.defaultHeaders ? { defaultHeaders: options.defaultHeaders } : {}
+    ...defaultHeaders ? { defaultHeaders } : {}
   });
   if (!options.fetch) {
     if (openaiClientCache.size >= 8) {
@@ -5747,19 +5805,25 @@ providerRegistry.register("local", {
   }),
   // Every supported local server exposes OpenAI-compatible `/v1/embeddings`
   // (the model must be an embedding model). None has an input-type parameter.
-  embed: (request) => embedOpenAICompatible(request, {
-    provider: "local",
-    baseUrl: requireLocalBaseUrl(request.baseUrl),
-    wireModel: localWireModelId(request.model)
-  }),
+  embed: (request) => embedOpenAICompatible(
+    { ...request, apiKey: localBearerKey(request.apiKey) },
+    {
+      provider: "local",
+      baseUrl: requireLocalBaseUrl(request.baseUrl),
+      wireModel: localWireModelId(request.model)
+    }
+  ),
   embedBatchSize: LOCAL_EMBED_BATCH_SIZE,
   // llama.cpp (`llama-server --rerank`) and vLLM serve `/v1/rerank`; Ollama and
   // LM Studio don't, and answer with a 404 ProviderError.
-  rerank: (request) => rerankCohereCompatible(request, {
-    provider: "local",
-    baseUrl: requireLocalBaseUrl(request.baseUrl),
-    wireModel: localWireModelId(request.model)
-  })
+  rerank: (request) => rerankCohereCompatible(
+    { ...request, apiKey: localBearerKey(request.apiKey) },
+    {
+      provider: "local",
+      baseUrl: requireLocalBaseUrl(request.baseUrl),
+      wireModel: localWireModelId(request.model)
+    }
+  )
 });
 function requireLocalBaseUrl(baseUrl) {
   if (!baseUrl) {
@@ -6133,6 +6197,13 @@ function formatChatError(error) {
       "Check your connection before trying again."
     );
   }
+  if (base.provider === "local" && (base.source === "auth" || base.statusCode === 401)) {
+    return present(
+      "auth",
+      "The local server rejected its key",
+      "Check the endpoint's key under Local models in AI Providers settings."
+    );
+  }
   if (base.source === "auth" || base.statusCode === 401) {
     return present(
       "auth",
@@ -6506,6 +6577,7 @@ function palsuRerankScore(query, document) {
   EventStream,
   FAST_MODE_SUFFIX,
   GGAIError,
+  LOCAL_NO_API_KEY,
   PRECONNECT_INTERVAL_MS,
   ProviderError,
   REDACTION_MARKER,
@@ -6528,6 +6600,7 @@ function palsuRerankScore(query, document) {
   isHardBillingMessage,
   isUsageLimitError,
   isValidToolCallName,
+  localBearerKey,
   localWireModelId,
   opensTurn,
   palsuAssistantMessage,