01
Context disappears
Decisions, risks and commitments get buried inside long transcripts, notes and chat history.
ClosedRoomMac App
GitHub ↗Local-first meeting intelligence for macOS
Remember the work.
Not just the transcript.
ClosedRoom records, transcribes, identifies speakers, analyzes, and remembers meetings locally , so decisions, actions, risks, and project context do not disappear after the call.
Local-first by defaultSpeaker-aware transcriptsProject memory, not isolated notes
1/2 · Today Workspace
2/2 · Meeting Analysis
Meeting → actions → decisions → project memory
Why ClosedRoom exists
Meetings create knowledge.
Most tools leave it fragmented.
Raw transcripts are useful, but they are rarely the final thing people need. The value lives in what was decided, who committed to what, what changed, and what still needs attention.
02
Meetings stay isolated
A transcript explains one call. It does not automatically show how a project changed across several meetings.
03
Cloud becomes the default boundary
Many meeting assistants require sensitive audio and transcript data to leave the machine before intelligence becomes useful.
ClosedRoom turns meetings into a local operational memory instead of another pile of transcripts.
The product idea
Capture first.
Build intelligence after.
ClosedRoom preserves the meeting locally, then adds ASR, speaker diarization, visual evidence, and structured analysis as separate layers.
1
ConfigureChoose diarization, visual intelligence and provider options.2
RecordSave microphone, system audio and recoverable artifacts locally.3
TranscribeRun local MLX / Nemotron ASR and persist timestamped text.4
EnrichAdd speaker clusters, visual evidence and audio intelligence.5
AnalyzeExtract summaries, actions, decisions, risks and project updates.6
RememberReuse the output across Today, Meeting and Project workspaces.What matters
Useful intelligence with
explicit boundaries.
01
Local-first control
Audio, transcripts, prompts, analysis outputs, diarization and visual artifacts stay on the Mac by default.
“The local machine is the default trust boundary.”02
Intelligence over transcription
ClosedRoom is designed to surface actions, decisions, risks, minutes and project updates , not just text.
03
Human-reviewable attribution
Speaker clusters remain stable. Visual evidence can suggest names only when confidence rules are satisfied.
04
Fail-soft enrichment
Optional diarization or visual intelligence can fail without invalidating a usable transcript.
Product proof
From a live meeting to
something you can act on.
RECORD → TRANSCRIBE
Preserve the meeting before inference starts.
ClosedRoom records microphone and system audio locally, then runs transcription asynchronously. The meeting exists as a recoverable artifact before expensive processing begins.
Slide 1 / 2
Configure the recording
Capture locally
TRANSCRIPT → INTELLIGENCE
See what matters without rereading everything.
The Meeting workspace connects audio, transcript, speaker context and structured outputs , including actions, decisions, risks and deeper analysis.
Slide 1 / 2
Meeting intelligence
Actions in detail
MEETING → PROJECT MEMORY
Carry context across multiple conversations.
Project views reuse meeting-level intelligence so the product can show current status, open actions, decisions, risks and recent changes without starting from zero after every call.
Slide 1 / 2
Cross-meeting project view
Operational project detail
End-to-end flow
The complete pipeline,
without hiding the boundaries.
Architecture
Local-first is a
system design choice.
The UI talks to one local FastAPI boundary. Recording, jobs, ASR, diarization, local model runtimes, persistence and diagnostics are coordinated behind that boundary.
Product surface
React/TypeScript runs in the browser or native WKWebView shell.
Local orchestration
FastAPI + AppServices coordinate recording, transcription, analysis, runtime and workspace state.
Replaceable execution
MLX/Nemotron, FluidAudio, Qwen3-VL and local-llm-server remain isolated behind explicit service boundaries.
Privacy boundary
Local by default.
Cloud only when chosen.
Default local path
Audio, transcripts, diarization artifacts, visual observations, local prompts, analysis results and project state remain on the Mac.
MLX / Nemotron · FluidAudio · Qwen3-VL · local-llm-serverExplicit opt-in path
Speechmatics and Gemini sit outside the default trust boundary and are used only when the user intentionally selects those providers.
No hidden cloud dependency in the default workflow.ClosedRoom
Turn meetings into
operational memory.
A local-first macOS workspace for recording, transcription, speaker context, structured analysis, and project memory.