Repository navigation
✨ Inbox: auto-transcribe hive recordings - #18432
Conversation
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of
|
Mutation testing2 mutant(s) in changed files are not reached by any test (NOT COVERED). A maintainer may apply the
Verdict: |
|
Hive status — updated in place; edits do not notify. Full history is in the Activity Monitor. @clubanderson this PR is held at Failing checks:
Next: the hive does not push to your branch. Push a new head (or reply here); CI and the review swarm re-run automatically and the PR returns to the normal path. stage 080-awaiting-merge-admission → 071-needs-author: optional-check-failing:overlayfs-exec-guard |
|
hive/scanner: branch is 4 commits behind v5 (baseline check: MERGE_BASE) and test (rest 4/4)/coverage are red. Please merge v5 and re-check; not pushing to a non-hive branch. 🐝 Hive Agent: — hive: agent=scanner backend=claude model=claude-sonnet-5-5 claude=2.1.325 |
05e0514 to
7c63ee3
Compare
There was a problem hiding this comment.
Decision needed: @clubanderson should recordings auto-transcribe when provider: auto would fall back to Deepgram (cloud), or only when the resolved provider is local whisper?
Read the diff against the tree at the PR head (Transcribe, precheck, resolveProvider, TranscribeSweep callers). The auto-transcribe plumbing looks correct to me: autoTranscribeRecording (recordingsource.go:~333) goes through Service.Transcribe (transcription.go:298), so the Enabled and ConsentAcknowledged gates, the busy/already-done checks and precheck still apply, and ErrRecordingExpired is now mapped to a skip rather than a failure (transcription.go:415). One point is worth a maintainer's eye.
Docs overstate "audio stays on the hive" (low–medium, privacy wording) — src/docs/inbox-recording.md (new "Transcription" section) says provider: auto "picks the built-in whisper first, so the audio stays on the hive". resolveProvider (transcription_providers.go:83-92) falls back to Deepgram when no earlier provider in autoProviderOrder is available. So on a hive where whisper is not available, a meeting recording is sent to a third-party API automatically at stop time. The recorder consent text is also silent on this: it says only "this hive transcribes the recording automatically" (inbox-recorder.js:112). The hive-wide consent presumably covers cloud providers, but recordings of people are the most sensitive audio this feature handles. Suggest either stating the fallback in the docs and consent copy, or having recordings only auto-transcribe when the resolved provider is local.
Suggestion, not a defect: TestRecordingAutoTranscribedOnFinalize uses a fake transcriber, so it cannot show which provider auto resolves to; a test pinning that resolution for a recording would back the docs claim.
— hive: agent=reviewer-2 backend=claude model=claude-sonnet-5-5 claude=2.1.325
7c63ee3 to
f4a6606
Compare
|
[APPROVALNOTIFIER] This PR is APPROVED Approval requirements bypassed by manually added approval. This pull-request has been approved by: The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
A finished recording is queued for transcription with the hive's provider as soon as it is finalized. AudioOpener (now in its own file, general to any source whose audio is on this hive's disk, such as uploads) lets the transcriber open the local file directly, so the spoke never calls its own HTTP API. Recording does not imply transcription consent: nothing runs until inbox.transcription is enabled with consent, after which recordings count as opted in regardless of mode and the sweep (also run from the recording ticker) catches up on recordings made while it was off. Signed-off-by: Andy Anderson <andy@clubanderson.com>
f4a6606 to
b192799
Compare
|
New changes are detected. LGTM label has been removed. |
Third of the "record a meeting directly in hive" PRs (backend #18408, recorder #18422): a recording is transcribed automatically when it ends, and a general
AudioOpenerlocal-file transcription path.What it adds
AudioOpener(pkg/inbox/audioopener.go): a general interface for any Inbox source whose audio lives on this hive's disk, such as dashboard recordings or uploaded audio and video files.OpenAudio(ctx, Item) (LocalAudio, error)returns the open file with its MIME type, size and modification time. Playback already serves such files withhttp.ServeContent. Transcription now opens them directly too (localTranscribeAudio), so the spoke never makes an HTTP call to itself. The provider gets the file's MIME type and size, retries reopen the file, andinbox.transcription.max_bytesis checked before upload. The hive still never decodes audio; the sandboxed whisper subprocess does (its ffmpeg already has the matroska/mov demuxers and the opus/aac decoders).automeans whisper first), with the same progress bar and ETA as any other Inbox transcription. It is audited asinbox_transcribewith the recorder as actor (orinbox-recording-sweepfor idle auto-finalized ones).inbox.transcriptionis enabled withconsent_acknowledged: true. From then on, everyhive-recordingitem counts as opted in whateverinbox.transcription.modesays. Recordings made while transcription was off are picked up by the transcription sweep, which now also runs from the 10-minute recording ticker. Expired audio is never sent.src/docs/inbox-recording.md.Tests
Go:
transcribe_unavailable=disabledmax_bytesskip🤖 Generated with Claude Code