Privacy-first conversation analysis
Drop in an interview, a meeting, a lecture — Debrief transcribes it on your own machine, works out who said what from the actual audio, and turns the transcript into analysis you can question. Nothing is uploaded unless you choose a cloud AI.
Local-first. Your recordings, transcripts, and speaker data stay on this device.
Drag in audio or video — MP3, WAV, MP4, MOV, and more. Whisper runs locally in a bundled sidecar, with FFmpeg included, so it just works offline.
Speaker diarisation from the actual audio — pyannote listens to voices, not text, so speakers are separated reliably. Name them once and the transcript follows.
Summaries, themes, tone, and a conversation-quality score across clarity, depth, balance, and pace — with concrete strengths and things to work on.
Question a single transcript, a project, or your whole library. Retrieval-backed chat finds the moments that matter and answers with the evidence.
Point Debrief at Ollama and everything — audio, transcript, analysis, chat — stays on your machine. Transcription and speaker identification are local no matter what.
Anthropic Claude, OpenAI, Google Gemini, Groq, OpenRouter, or any OpenAI-compatible endpoint. Your key, your choice, one settings tab each.
Speech and speaker models ship inside the installer. No HuggingFace token, no sign-up, no external server to stand up first.
Research interviews, staff meetings, counselling sessions — the conversations worth analysing are exactly the ones that should not sit on someone else's server.
Qualitative interviews and focus groups, transcribed and diarised without the audio ever leaving your machine — then organised into projects you can search, tag, and interrogate across participants.
Debrief lectures, tutorials, and meetings: see how balanced the discussion was, who spoke and for how long, what themes kept coming up — and get the summary before the next session.