@tiens.nguyen/gonext-local-worker 1.0.13 → 1.0.15

package/gonext-local-worker.mjs

@@ -145,25 +145,91 @@ async function runChatJob(job) {
     apiKey: payload.apiKey || "ollama",
   });
 
+  let buf = "";
+  let flushTimer = null;
+  let fullText = "";
+
+  const flushChunks = async () => {
+    flushTimer = null;
+    const t = buf;
+    buf = "";
+    if (!t) return;
+    const res = await workerFetch(`/api/worker/jobs/${jobId}/chunk`, {
+      method: "POST",
+      body: JSON.stringify({ text: t }),
+    });
+    if (!res.ok && res.status !== 204) {
+      console.error(`[gonext-worker] chunk POST failed ${res.status} for ${jobId}`);
+    }
+  };
+
+  const enqueueText = (s) => {
+    if (!s) return;
+    fullText += s;
+    buf += s;
+    if (!flushTimer) {
+      flushTimer = setTimeout(() => void flushChunks(), 12);
+    }
+  };
+
   try {
-    const completion = await client.chat.completions.create({
+    const stream = await client.chat.completions.create({
       model: payload.modelId,
       messages: toOpenAIMessages(payload.messages),
+      stream: true,
       temperature: 0,
     });
-    const text = completion.choices[0]?.message?.content ?? "";
+
+    let tokenCount = 0;
+    let isStartThinking = false;
+    let isEndThinking = false;
+
+    for await (const chunk of stream) {
+      const delta = chunk.choices[0]?.delta;
+      const content = delta?.content ?? "";
+      const reasoningContent = delta?.reasoning_content;
+      tokenCount += 1;
+
+      if (reasoningContent) {
+        if (!isStartThinking) {
+          isStartThinking = true;
+          enqueueText("<think>");
+        }
+        enqueueText(reasoningContent);
+      } else {
+        if (isStartThinking && !isEndThinking) {
+          isEndThinking = true;
+          enqueueText("</think>");
+        }
+        if (content) {
+          enqueueText(content);
+        }
+      }
+    }
+
+    if (flushTimer) {
+      clearTimeout(flushTimer);
+      flushTimer = null;
+    }
+    await flushChunks();
+
     const totalTimeSeconds = (Date.now() - start) / 1000;
     await workerFetch(`/api/worker/jobs/${jobId}`, {
       method: "PATCH",
       body: JSON.stringify({
         jobStatus: "completed",
-        resultText: text,
-        tokenCount: 1,
+        resultText: fullText,
+        tokenCount: Math.max(1, tokenCount),
         totalTimeSeconds,
       }),
     });
     console.log(`[gonext-worker] completed ${jobId} (${totalTimeSeconds.toFixed(1)}s)`);
   } catch (e) {
+    if (flushTimer) {
+      clearTimeout(flushTimer);
+      flushTimer = null;
+    }
+    await flushChunks().catch(() => {});
     const message = e instanceof Error ? e.message : String(e);
     await workerFetch(`/api/worker/jobs/${jobId}`, {
       method: "PATCH",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.13",
+  "version": "1.0.15",
   "description": "Polls GoNext cloud API for async local LLM jobs and runs them against Ollama/OpenAI-compatible servers on this Mac",
   "type": "module",
   "license": "MIT",