feat(ai): show a live streaming token count in the chat

AUTONOMOUS_STATE.json

@@ -7,7 +7,7 @@
     "asset-curator",
     "qa-playtester"
   ],
-  "last_cycle": 226,
+  "last_cycle": 227,
   "completed_targets": [
     {
       "cycle": 75,
@@ -1330,6 +1330,12 @@
       "target": "Flow type-check over the session's additions (no change)",
       "metric_delta": "No code change. Ran `npm run flow` on the committed session state: 1039 errors total (the environment's current baseline - the previously recorded 870 was a different reading; the fork's CI does not gate app Flow). Crucially, verified the errors are NOT from this session's work: AiRequestUtils.js (where the per-agent summary/token data layer lives) has ZERO Flow errors, and no error mentions summarizeSubAgentActivity, localAiRequestProviderTelemetry or onProviderTelemetry. The CustomAIClient errors cited (lines 197/595/1052/1070) are in pre-existing code (e.g. normalizePersistedMessages' `entry.type === 'text'` from cycle 95). eslint/build remain the gating checks and pass.",
       "commit": "docs(autonomous): verify Flow over the session additions"
+    },
+    {
+      "cycle": 227,
+      "target": "Live streaming token counter + a slow-stream mock option (Tier 2)",
+      "metric_delta": "Added the Tier 2 'streaming token counter': PartialStreamText (fork-owned ChatMessages) now appends ' . ~N tokens' computed with estimateTokens(partialContent), so progress is legible while a local model generates. Also added MOCK_STREAM_DELAY_MS to the mock provider so streaming-only UI can be exercised (chunks spaced over time). Verified: unit (estimateTokens is tested) + eslint + production build exit 0 + divergence clean; the slow stream itself verified by curl (chunks + tool calls arrive over time). Honest caveat: the counter could not be caught in a Chrome DevTools snapshot mid-stream - the standalone form's transcript/working row was not captured while a turn ran - so it rests on build/unit evidence and the fact that PartialStreamText (its host) was live-verified earlier, not on a fresh live observation.",
+      "commit": "feat(ai): show a live streaming token count in the chat"
     }
   ],
   "backlog": [

newIDE/app/src/AiGeneration/AiRequestChat/ChatMessages.js

@@ -73,6 +73,7 @@ import {
   customGetAiRequestTokenTotal,
   customGetAiRequestPartialContent,
   customGetAiRequestProviderTelemetry,
+  estimateTokens,
 } from '../../AI/CustomAIClient';
 
 /**
@@ -93,9 +94,15 @@ const PartialStreamText = ({ aiRequestId }: {| aiRequestId: string |}) => {
   );
   if (!partialContent) return null;
   const tail = partialContent.slice(-160);
+  // Tier 2 streaming token counter: an approximate running count of the
+  // streamed content, so progress is legible while a local model generates.
+  const streamedTokens = estimateTokens(partialContent);
   return (
     <Text noMargin displayInlineAsSpan size="body-small" color="secondary">
       {tail}
+      {streamedTokens > 0
+        ? ` · ≈${streamedTokens.toLocaleString()} tokens`
+        : ''}
     </Text>
   );
 };

scripts/dev/mock-ai-provider.js

@@ -165,13 +165,23 @@ const server = http.createServer((req, res) => {
           Object.assign({ 'Content-Type': 'text/event-stream' }, CORS, headers)
         );
         const chunks = reply.match(/.{1,24}/g) || (reply ? [reply] : []);
-        for (const chunk of chunks) {
+        // MOCK_STREAM_DELAY_MS spaces out chunks so UI that only shows while a
+        // stream is live (e.g. the streaming token counter) can be observed.
+        const streamDelayMs = Number(process.env.MOCK_STREAM_DELAY_MS) || 0;
+        const writeChunk = index => {
+          if (index >= chunks.length) return sendTail();
           res.write(
             'data: ' +
-              JSON.stringify({ choices: [{ delta: { content: chunk } }] }) +
+              JSON.stringify({ choices: [{ delta: { content: chunks[index] } }] }) +
               '\n\n'
           );
-        }
+          if (streamDelayMs > 0) {
+            setTimeout(() => writeChunk(index + 1), streamDelayMs);
+          } else {
+            writeChunk(index + 1);
+          }
+        };
+        const sendTail = () => {
         if (toolCalls.length > 0) {
           res.write(
             'data: ' +
@@ -205,6 +215,9 @@ const server = http.createServer((req, res) => {
         );
         res.write('data: [DONE]\n\n');
         return res.end();
+        };
+        writeChunk(0);
+        return;
       }
 
       const message = { role: 'assistant', content: reply };