@code-yeongyu/senpi-ai 2026.7.31 → 2026.8.3

package/dist/api/openai-completions.js

@@ -1,5 +1,5 @@
 import OpenAI from "openai";
-import { calculateCost, clampThinkingLevel, supportsXhigh } from "../models.js";
+import { calculateCost, clampThinkingLevel, supportsMax, supportsXhigh } from "../models.js";
 import { formatProviderError, normalizeProviderError } from "../utils/error-body.js";
 import { AssistantMessageEventStream } from "../utils/event-stream.js";
 import { shortHash } from "../utils/hash.js";
@@ -197,7 +197,7 @@ export const stream = (model, context, options) => {
                 totalTokens: 0,
                 cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
             },
-            stopReason: "stop",
+            stopReason: "pending",
             timestamp: Date.now(),
         };
         try {
@@ -206,7 +206,7 @@ export const stream = (model, context, options) => {
             const grammarToolInputProperties = createGrammarToolInputProperties(context.tools, compat.supportsOpenAIGrammarTools);
             const cacheRetention = resolveCacheRetention(options?.cacheRetention ?? model.cacheRetention, options?.env);
             const cacheSessionId = cacheRetention === "none" ? undefined : options?.sessionId;
-            const client = createClient(model, context, clientAuth.apiKey, clientAuth.headers, cacheSessionId, compat);
+            const client = createClient(model, context, clientAuth.apiKey, clientAuth.headers, options?.fetch, cacheSessionId, compat);
             let params = buildParams(model, context, options, compat, cacheRetention, grammarToolInputProperties);
             const nextParams = await options?.onPayload?.(params, model);
             if (nextParams !== undefined) {
@@ -384,9 +384,7 @@ export const stream = (model, context, options) => {
                     }
                     // Note: the "input" fallback here should/must not be taken.  in case the LLM makes up
                     // a tool we don't knwo about, we at least have a place to stash our stuff.
-                    const customInputProperty = toolCall.custom
-                        ? (grammarToolInputProperties.get(name) ?? "input")
-                        : undefined;
+                    const customInputProperty = toolCall.custom && !toolCall.function ? (grammarToolInputProperties.get(name) ?? "input") : undefined;
                     const hasCustomInput = customInputProperty !== undefined;
                     block = {
                         type: "toolCall",
@@ -425,7 +423,7 @@ export const stream = (model, context, options) => {
                 if (!block.name && name) {
                     block.name = name;
                 }
-                if (toolCall.custom && !block.customInput) {
+                if (toolCall.custom && !toolCall.function && !block.customInput) {
                     const customInputProperty = grammarToolInputProperties.get(block.name) ?? "input";
                     block.arguments = { [customInputProperty]: "" };
                     block.customInput = {
@@ -483,6 +481,7 @@ export const stream = (model, context, options) => {
                     output.usage = parseChunkUsage(choiceUsage, model);
                 }
                 if (choice.finish_reason) {
+                    output.rawStopReason = choice.finish_reason;
                     const finishReasonResult = mapStopReason(choice.finish_reason);
                     output.stopReason = finishReasonResult.stopReason;
                     if (finishReasonResult.errorMessage) {
@@ -610,10 +609,13 @@ export const stream = (model, context, options) => {
             if (output.stopReason === "aborted") {
                 throw new Error("Request was aborted");
             }
+            if (!hasFinishReason && !compat.supportsFinishReason) {
+                output.stopReason = output.content.some((block) => block.type === "toolCall") ? "toolUse" : "stop";
+            }
             if (output.stopReason === "error") {
                 throw new Error(output.errorMessage || "Provider returned an error stop reason");
             }
-            if (!hasFinishReason) {
+            if ((compat.supportsFinishReason && !hasFinishReason) || output.stopReason === "pending") {
                 throw new Error("Stream ended without finish_reason");
             }
             stream.push({ type: "done", reason: output.stopReason, message: output });
@@ -652,7 +654,7 @@ export const streamSimple = (model, context, options) => {
     const clampedReasoning = options?.reasoning ? clampThinkingLevel(thinkingModel, options.reasoning) : undefined;
     const reasoningEffort = clampedReasoning === "off"
         ? undefined
-        : clampedReasoning === "max" && thinkingLevelMap?.max !== undefined
+        : clampedReasoning === "max" && supportsMax(thinkingModel)
             ? "max"
             : clampMaxForOpenAI(clampedReasoning, supportsXhigh(thinkingModel));
     const toolChoice = options?.toolChoice;
@@ -662,7 +664,7 @@ export const streamSimple = (model, context, options) => {
         toolChoice,
     });
 };
-function createClient(model, context, apiKey, optionsHeaders, sessionId, compat = getCompat(model)) {
+function createClient(model, context, apiKey, optionsHeaders, fetch, sessionId, compat = getCompat(model)) {
     const headers = { ...model.headers };
     if (model.provider === "github-copilot") {
         const hasImages = hasCopilotVisionInput(context.messages);
@@ -692,6 +694,7 @@ function createClient(model, context, apiKey, optionsHeaders, sessionId, compat
         apiKey,
         baseURL: model.baseUrl,
         dangerouslyAllowBrowser: true,
+        fetch,
         defaultHeaders: headers,
     });
 }
@@ -759,6 +762,12 @@ function buildParams(model, context, options, compat = getCompat(model), cacheRe
     }
     else if (compat.thinkingFormat === "qwen" && model.reasoning) {
         params.enable_thinking = !!options?.reasoningEffort;
+        if (options?.reasoningEffort && compat.supportsReasoningEffort) {
+            const effort = resolveReasoningEffort(thinkingLevelMap, options.reasoningEffort);
+            if (effort !== undefined) {
+                params.reasoning_effort = effort;
+            }
+        }
     }
     else if (compat.thinkingFormat === "qwen-chat-template" && model.reasoning) {
         params.chat_template_kwargs = {