PRODUCT · LOCAL-FIRST MEETING INTELLIGENCEActive development · macOS

ClosedRoom

Turn meetings into operational memory. Record, transcribe and understand conversations on your Mac, then carry decisions, actions, risks and context across the project.

Local-first by defaultSpeaker-aware transcriptsCross-meeting project memory

See the workflow ↓

ClosedRoom Today workspace with meetings and reusable project context — Today · meetings become reusable context

ClosedRoom meeting intelligence workspace with structured analysis — Meeting intelligence · transcript, speakers and structured outputs

Record → Transcribe → Understand → Remember

THE PROBLEM

Meetings create knowledge.
Most tools leave it fragmented.

The useful output is rarely the transcript itself. Teams need to recover what changed, what was decided, who owns what, which risks remain open and how the project evolved across several conversations.

Transcripts are not operational memory

Long transcripts preserve words, but decisions, commitments, risks and open questions still have to be recovered manually.

Meetings stay isolated

One meeting can explain one call. It does not automatically preserve the project state across a sequence of conversations.

Cloud often becomes the default boundary

Sensitive audio and transcripts can leave the machine before the user receives any useful intelligence or project context.

ClosedRoom is designed around meeting → intelligence → memory, not around accumulating another archive of transcripts.

WHY CLOSEDROOM IS DIFFERENT

Private by default. Useful beyond the transcript.

The product focuses on three outcomes that matter in real meeting workflows.

Private by default

Sensitive meeting data stays on the Mac in the default path. External providers are explicit choices rather than hidden dependencies.

Intelligence, not just transcripts

ClosedRoom turns conversations into speaker-aware transcripts, actions, decisions, risks, questions and editable notes.

Memory across meetings

Outputs become reusable project context so commitments, risks and decisions do not disappear inside isolated meeting files.

HOW IT WORKS

Record → Transcribe → Understand → Remember

The user mental model stays simple even though capture, inference, enrichment and persistence remain separate engineering stages underneath.

Record

Capture microphone and system audio locally and persist recoverable meeting artifacts before expensive inference begins.

Transcribe

Run the local ASR path as a persisted, observable job. Speaker diarization remains a separate enrichment instead of a prerequisite.

Understand

Turn the transcript into speaker-aware summaries, actions, decisions, risks, open questions and editable meeting notes.

Remember

Reuse meeting outputs inside Today and Project views so current status and historical context survive beyond a single call.

ClosedRoom workflow from recording through transcription and understanding to reusable project memory — The product keeps the user journey simple while capture, inference and persistence remain explicit engineering boundaries.

THE PRODUCT

From live capture to cross-meeting project memory.

The product surfaces the whole operational path instead of treating transcription as the final destination.

ClosedRoom recording setup — Recording · focused setup before capture

ClosedRoom active local recording — Capture · the meeting is persisted before inference

ClosedRoom meeting intelligence workspace — Meeting intelligence · transcript, speakers and structured outputs

ClosedRoom deep-dive action items — Actions · operational detail extracted from the conversation

ClosedRoom cross-meeting project analysis — Project memory · status, decisions, risks and updates across meetings

ClosedRoom project detail workspace — Project detail · reusable context instead of isolated transcripts

ARCHITECTURE

ClosedRoom owns the meeting product. Korgis owns reusable Local AI runtime infrastructure.

The macOS app and loopback FastAPI boundary coordinate recording, jobs, persistence, ASR, diarization and meeting state. Korgis provides the reusable LLM/VLM execution boundary underneath the product workflow.

ClosedRoom local-first architecture showing the macOS product boundary, local ASR and diarization, Korgis runtime, persistence and optional cloud providers outside the default trust boundary

CLOSEDROOMOwns the user problem

Capture, meetings, transcripts, speaker state, structured notes, user edits, persistence and project memory.

KORGISOwns reusable Local AI runtime concerns

Model loading, backend selection, runtime lifecycle, text and vision inference, logs and diagnostics.

NATIVE HELPERSOwn platform-specific execution

macOS capture and local diarization stay behind explicit process boundaries instead of leaking into product logic.

EVIDENCE & LIMITS

Local-first is an explicit system boundary, not a blanket quality claim.

ClosedRoom already implements the core meeting-intelligence workflow and makes provider boundaries visible. Model quality, hardware behavior and broader privacy guarantees remain separate evidence questions.

DEFAULTLocal Mac boundary

Recording, local ASR, diarization, meeting analysis state, edits and project memory stay on the Mac in the default path.

Local filesystem · SQLite · local inference · Korgis

EXPLICIT OPT-INCloud provider boundary

Speechmatics and Gemini can be selected intentionally. They are outside the default trust boundary and are never silent fallbacks.

Provider choice changes the data boundary visibly.

Designed to degrade usefullyRecording finalized firstTranscript survives optional enrichment failuresSpeaker identity may abstainUser edits remain explicit

EVIDENCE BOUNDARY

The product workflow and trust boundary are implemented; model and hardware claims stay scoped.

ClosedRoom persists meeting artifacts locally, supports local ASR and diarization paths, produces structured meeting intelligence, preserves user edits and carries outputs into project memory. Optional Speechmatics and Gemini paths are explicit opt-ins and can move selected meeting data outside the local boundary.

Supported today

Local meeting capture with recoverable persisted artifacts before expensive inference
Local transcription paths plus local FluidAudio diarization on supported Macs
Structured summaries, actions, decisions, risks, questions and editable notes
Cross-meeting Today and Project memory built from persisted meeting outputs
Explicit provider selection with no silent cloud fallback in the default workflow
Optional enrichment can degrade or abstain without invalidating a usable transcript

Not claimed here

Public downloadable GitHub Release for the macOS application
Independent end-to-end privacy certification
Representative latency, memory and thermal benchmark across supported Macs
Guarantees of perfect transcription, speaker identity or meeting understanding

CONNECTED SYSTEM

Where this project fits

BUILD · INFRASTRUCTUREKorgisReusable Local AI runtime and execution boundary

PRODUCT · MEETING INTELLIGENCEClosedRoomSensitive workflow proving ground and operational memory product

MEASURE · EVIDENCEPerformance LabRepresentative runtime and hardware viability evidence

CLOSEDROOM · MACOS

Turn meetings into
operational memory.

Explore the source-built macOS project, its Local AI boundaries and the engineering decisions behind the full meeting-to-memory workflow.

Explore ClosedRoom ↗Explore Korgis ↗