@gotcos/glasses-server 6.8.0 → 6.9.0

package/CHANGELOG.md

@@ -1,5 +1,25 @@
 # Changelog
 
+## 6.9.0
+
+Live recoverable prompt transcription for COS Glasses builds 200+.
+
+- **Words appear while speaking.** After each audio chunk is durably acknowledged,
+  its sanitized fast/local transcript is published on the existing authenticated,
+  replayable display stream as `prompt_transcript`; the phone/G2 client can fill
+  the Listening body without adding another recorder, polling loop, or ASR job.
+- **Recovery remains authoritative.** The event is optional presentation state.
+  Stored WAV chunks, final HQ transcription, glossary cleanup, editing, retry,
+  and send behavior remain unchanged and continue even if no display client is
+  connected.
+- **Stale retries cannot repaint.** The server rechecks the exact draft, chunk
+  index, and audio bytes after warm transcription. Replaced audio never emits
+  its obsolete words, while client-side draft scoping, ordering, and replay
+  deduplication handle reconnects safely.
+- **Public boundary retained.** This release adds no private COS paths, personal
+  data, LaunchAgent controls, remote restart authority, or machine-management
+  endpoints.
+
 ## 6.8.0
 
 Public-safe meeting finalization for COS Glasses build 199.

package/README.md

@@ -64,7 +64,9 @@ The built-in IP allowlist blocks public-internet traffic regardless.
 - Send phone photos with queued prompts, and review assistant-selected generated,
   research, or explicitly used email images in Messages and on the G2 lens
 - Recover long voice prompts after phone, network, or server interruptions. Audio
-  chunks are saved before transcription and retained locally for 72 hours.
+  chunks are saved before transcription and retained locally for 72 hours. On
+  compatible app builds, their warm transcript also appears live while speaking;
+  final HQ transcription remains authoritative.
 - Live voice capture + transcription during meetings
 - Local whisper.cpp transcription (free) with OpenAI fallback (optional)
 - Tasks / calendar / people context **if** you run the

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gotcos/glasses-server",
-  "version": "6.8.0",
+  "version": "6.9.0",
   "description": "COS Glasses — self-hosted AI heads-up-display server for Even G2 smart glasses, powered by your local Claude Code or Codex CLI",
   "type": "module",
   "bin": {

package/server/lib/display-bus.ts

@@ -8,7 +8,7 @@ const bus = new EventEmitter()
 bus.setMaxListeners(20) // Multiple glasses clients
 
 export interface DisplayEvent {
-  type: 'chunk' | 'done' | 'error' | 'tool_status' | 'start' | 'session_restore' | 'transcript_chunk' | 'recording_start' | 'recording_stop' | 'coaching_nudge'
+  type: 'chunk' | 'done' | 'error' | 'tool_status' | 'start' | 'session_restore' | 'transcript_chunk' | 'prompt_transcript' | 'recording_start' | 'recording_stop' | 'coaching_nudge'
   data: Record<string, unknown>
 }
 

package/server/routes/prompt-drafts.ts

@@ -38,6 +38,7 @@ import { createBreaker } from '../lib/claude-circuit.js'
 import { logTokenAudit } from '../lib/token-audit.js'
 import { atomicWriteFileSync } from '../lib/atomic-fs.js'
 import { dataPath } from '../lib/data-dir.js'
+import { emitDisplay } from '../lib/display-bus.js'
 
 export const promptDraftsRouter = Router()
 
@@ -234,7 +235,17 @@ promptDraftsRouter.post('/prompt-drafts/:draftId/chunks', async (req, res) => {
     const nextTotal = Object.values(before.chunkBytes).reduce((sum, bytes) => sum + bytes, 0) - existingBytes + audio.length
     if (nextTotal > MAX_DRAFT_BYTES) return res.status(413).json({ error: 'prompt draft too large' })
     const meta = await savePromptDraftChunk(req.params.draftId, chunkIndex, audio)
-    warmTail = warmTail.then(() => transcribeChunk(req.params.draftId, chunkIndex, audio, 'fast', 'warm').then(() => undefined)).catch(err => {
+    warmTail = warmTail.then(async () => {
+      const text = await transcribeChunk(req.params.draftId, chunkIndex, audio, 'fast', 'warm')
+      // The durability ACK above remains immediate. Publish the optional warm
+      // transcript only after rechecking that this exact audio still owns the
+      // chunk index; a retry/replacement must never paint stale words.
+      if (!isCurrentChunk(req.params.draftId, chunkIndex, audio)) return
+      emitDisplay({
+        type: 'prompt_transcript',
+        data: { draftId: req.params.draftId, chunkIndex, text },
+      })
+    }).catch(err => {
       console.warn(`[prompt-draft] warm transcription failed ${req.params.draftId}/${chunkIndex}: ${err.message}`)
     })
     res.json({ draftId: meta.draftId, chunkIndex, acked: true, receivedChunkIndexes: meta.receivedChunkIndexes, chunkBytes: meta.chunkBytes[String(chunkIndex)] ?? audio.length, transcriptPending: true, expiresAt: meta.expiresAt })