Skip to content

fix: route managed Telegram voice transcription#825

Merged
artemgetmann merged 1 commit into
mainfrom
codex/final-managed-telegram-onboarding-fix
May 31, 2026
Merged

fix: route managed Telegram voice transcription#825
artemgetmann merged 1 commit into
mainfrom
codex/final-managed-telegram-onboarding-fix

Conversation

@artemgetmann
Copy link
Copy Markdown
Owner

Review Fast Path

  • User path fixed: managed Telegram voice note -> backend OpenAI transcription -> transcript reaches agent as text.
  • Proof: uv run --no-project --with-requirements services/jarvis-backend/requirements.txt pytest services/jarvis-backend/tests/test_app.py -q passed 34 tests; pnpm exec vitest run src/media-understanding/runner.managed-audio.test.ts src/consumer/jarvis-backend-client.test.ts --reporter verbose --testTimeout=10000 passed 8 tests; swift test --package-path apps/macos --filter TelegramSetupBootstrapTests passed 19 tests; git diff --check passed.
  • Shared-state footgun removed: packaged managed onboarding now seeds audio transcription config without bundling raw provider keys into the app.
  • Still hurts: no live Telegram voice-note proof on the isolated managed bot in this PR; Render deploy/live smoke remains the next gate.

Why This Matters

  • Before this, managed users could send a voice note and the agent saw a raw audio attachment, so it answered with local-transcription intent instead of the transcript.
  • Managed Personal users should not need local OpenAI STT keys for basic voice notes.

Scope Boundary

  • Adds one narrow backend managed utility: openai.audio.transcribe.
  • Routes managed media audio entries through that backend utility.
  • Seeds tools.media.audio during managed Telegram bootstrap.
  • Does not bundle provider keys into the macOS app.
  • Does not touch shared/default runtime, /Applications/Jarvis.app, or user Telegram.

Verification

  • uv run --no-project --with-requirements services/jarvis-backend/requirements.txt pytest services/jarvis-backend/tests/test_app.py -q -> 34 passed.
  • pnpm exec vitest run src/media-understanding/runner.managed-audio.test.ts src/consumer/jarvis-backend-client.test.ts --reporter verbose --testTimeout=10000 -> 8 passed.
  • swift test --package-path apps/macos --filter TelegramSetupBootstrapTests -> 19 passed.
  • git diff --check -> passed.
  • pnpm tsgo -> failed on pre-existing unrelated type drift across extensions/tests; no reported failures in changed managed-audio files.

AI Assistance

  • AI-assisted
  • Testing degree: targeted

- What: add a backend-managed OpenAI audio transcription utility and route managed media audio through it.

- Why: managed Telegram voice notes must become transcript text before the agent runs, without local user OpenAI keys.

- Risk: adds a bounded backend provider call for audio bytes; scoped to managed mode and existing managed utility auth.
@artemgetmann artemgetmann force-pushed the codex/final-managed-telegram-onboarding-fix branch from d9fb19c to 96f7af3 Compare May 31, 2026 05:57
@artemgetmann artemgetmann merged commit 18a8a3e into main May 31, 2026
4 of 9 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant