Open-source · self-hostable · AI optional
BriefRelay is a modular meeting (and media) capture stack: ingest audio/video, produce a diarised transcript, turn it into actions + decisions, and ship it to the places your team actually works, with governance, storage choice, and an actual record-only mode.
No dark patterns. No "we recorded your meeting to help you" surprises. You decide what gets captured, where it lives, and which modules are allowed to run.
BriefRelay started as a late-night habit: record something useful, then spend too long trying to make sense of it later. I kept stitching together tools like Whisper/WhisperX with small scripts that watched folders and pushed files from one step to the next. It worked, but only in the way a homemade rig "works".
The thing I wanted was simple: a pipeline that stays flexible. Drop an .mp4 in and let it become a diarised transcript. Or drop a transcript in directly and have the next stage treat it exactly the same, with no special cases and no "wrong door". That non-mutually-exclusive flow is basically the founding idea.
There’s also a selfish reason: I’m not great at distinguishing voices. When you’re processing podcasts with recurring hosts, getting consistent speaker labels across episodes via voiceprints is a quality-of-life upgrade, and I haven’t seen it done well outside of deep integrations (where the platform already knows who joined the call). That doesn’t help when multiple people speak through one participant, or when you’re starting from raw audio.
BriefRelay is me trying to turn that scrappy set of scripts into a tool that’s actually pleasant to run: open-source, self-hostable by default, and honest about what it does. It’s early stages. The aim right now is getting the architecture right, then earning the polish.
Everything is modular: turn off what you don’t want. Lock down what you do.
Composable stages. Drop files in. Get structured output out.
Bring your own capture: upload, watch-folder, calendar rules, bot-join, or local screen/audio capture.
Media and metadata become a single “meeting object” in storage, with a stable ID and audit trail.
Transcription, diarisation, voiceprints, summaries, embeddings — each can be enabled/disabled per policy.
Send recaps, create tasks, post to channels, and keep a single source of truth for review and governance.
If you already have a transcript, you shouldn’t have to “re-run capture”. Drop it into the same meeting folder and the next stage picks it up. If you only have audio/video, BriefRelay can produce the transcript first — and everything downstream stays identical.
A quick, opinionated view of what most tools optimise for vs what BriefRelay is trying to do.
| Product | Self-host | AI-off mode | Local capture (no integration) | Pluggable storage | Voiceprint continuity |
|---|---|---|---|---|---|
| BriefRelay (target) | Yes | Yes | Yes | Yes | Yes |
| Otter.ai | No | No | Sometimes | No | Limited |
| Fireflies.ai | No | No | Sometimes | No | Limited |
| Fathom | No | No | Sometimes | No | No |
| tl;dv | No | No | Sometimes | No | No |
| Tactiq (transcript-first) | No | Partial | Yes | No | No |
| Teams + Copilot / Recap | No | No | No | No | Integration-based |
Notes: “Sometimes” varies by plan and capture mode. “Integration-based” means the platform knows the participant identity because it owns the meeting context, not because it can robustly resolve voices from raw audio.
BriefRelay is a concept in active development. Expect rough edges, missing bits, and sharp corners. If you need a polished SaaS today, this probably isn’t it - yet.