Repository navigation
✨ Inbox: drop audio and video files and transcribe them - #18476
Conversation
Audio/video (mp3, m4a, aac, wav, ogg, opus, webm, mp4, mov) is checked by container signature, streamed to the upload store under the recording size cap and the uploads quota, played from disk through the generic same-origin stream (AudioOpener) and auto-transcribed when transcription is on. Originals follow the recordings retention; the transcript, item and fact stay. Video keeps only its audio. Upload requests get a 20-min read window. Signed-off-by: Andy Anderson <andy@clubanderson.com>
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of
|
|
[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 |
|
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. |
Fifth and last PR of drag-and-drop ingest: audio and video files. Builds on the recording pipeline (#18408) and its local-file transcription (#18432).
What it does
ftyp, RIFF/WAVE, Ogg, EBML); a mismatch, or media without an extension, is refused. Audio/video links stay refused.connector-state/uploads/<id>/original(0700/0600), never held in memory, capped by the recording limitinbox.recording.max_bytes(200 MiB) and the uploads quota. The browser applies the same cap per file (settings.uploads.max_media_bytes).uploaditem (upload_kind: audio|video) with a pending Knowledge fact (attr_upload_content: pending) that the machine transcript later fills, like a recording's. The Inbox shows its own "added without text" note.AudioSourceandAudioOpener, so ▶ Play streams the file from disk through the existing same-origin/api/inbox/upload/<id>/audio/streamhandler (Range supported). Video keeps only its audio track: no video playback.inbox.transcriptionenabled and consented, the transcription is queued as soon as the file is added (audited asinbox_transcribe, actor = who added it). It reads the file locally throughAudioOpener(✨ Inbox: auto-transcribe hive recordings #18432), so with whisper the audio never leaves the hive. Otherwise 🎧 Transcribe shows the usual reason.inbox.uploads.media_retention_days, which now defaults toinbox.recording.retention_days(5 days); the transcript, item and fact stay.Fix for the earlier upload PRs (#18411, #18418, #18425, #18426)
Upload requests now get a 20-minute read window (
http.ResponseController.SetReadDeadline) instead of the dashboard server's 30-secondReadTimeout. Without it, a 25 MB document from the earlier PRs could be cut off mid-upload on a slow link, and a 200 MB recording could never arrive. Request processing is bounded at 25 minutes; text extraction keeps its own 30 s limit and link fetches their 60 s limit.Tests
OpenLocalAudioand the same-origin link; recording-cap and quota refusals that leave nothing behind; retention following the recording default with the item kept; transcription queued on add; HTTP upload of an mp3 played back through the stream handler, and a fake.mp3refused with 415.No new dependencies.
🤖 Generated with Claude Code