Keep recording if you're wrapping up notes or takeaways. Otherwise Minutes will stop and process this session.
Update available
A new version is ready to install
Release notesDev tools
Set up CLI
What's New
Update the CLI to match
MD
Viewing
Last Change Review
Before
Current
No meetings yet
Record your first conversation to get started.
Recall
Starting Recall...
Loading weekly summary…
Session ended.
Meeting
Select a meeting to view transcript and notes.
Create Draft
Pick the artifact you want Minutes to create from this conversation. It opens
immediately in source mode beside Recall.
Readiness Center
This view only reports what Minutes can verify right now. Optional items stay
optional, and unknown states are never hidden behind fake confidence.
Recovery Center
Retry preserved captures and watched files that failed to
process, or open them directly in a folder.
No recovery items right now.
Jobs Center
Background processing jobs for recent recordings. Open
transcripts early, retry failures, and keep moving.
No processing jobs right now.
Minutes
Conversation memory for humans and agents. Capture once, keep the transcript, decisions, and context useful everywhere.
Desktop app•Created by Mat Silverstein•Open-source conversation memory for AI assistants
Capture
Meetings, memos, and thoughts become structured markdown instead of disappearing
into app state.
Search
Timeline, transcript, action items, and context stay queryable across desktop,
CLI, and MCP surfaces.
Local First
Audio capture and storage stay on your machine unless you explicitly opt into
external model services.
Agent Native
Built to work well with Claude Desktop, Claude Code, Cowork, Codex, and other
tool-calling assistants.
Command-line tool
Built as a local-first desktop app with a tray workflow, CLI, and MCP surfaces that all point at the same markdown-backed record of your conversations.
Settings
IDENTITY
Your name
Used to attribute meeting content to you ("Mat said…"). Leave blank if you'd rather stay anonymous.
Your email addresses
Comma-separated. Every address you send from, so calendar attendees arriving under different emails fold onto the same person.
Name variants
Comma-separated. Alternate forms — nicknames, formal spellings — so everything folds onto your canonical name.
GENERAL
Show completion alerts
Play sound cues for start, stop, and complete
Open Minutes at login
Quick Thought shortcut
or Caps Lock /
fn
Hold to record a quick thought, or tap
to lock and tap again to stop.
DICTATION
Dictation shortcut
or Caps Lock /
fn
Hold to dictate, or tap to lock and tap
again to stop.
Dictation whisper model
Smaller models = faster dictation. "base" is recommended.
Log dictations to daily note
Silence timeout
End dictation session after this much silence (after first speech).
Recent dictations
Loading...
LIVE TRANSCRIPT
Backend
Standalone live transcript currently inherits your main transcription engine.
Keyboard shortcut
Toggle live transcript on/off from any app.
COMMAND PALETTE
Keyboard shortcut
Open the Minutes command palette from any app.
TRANSCRIPTION
Engine
Whisper and Parakeet control batch, recording, and recording-sidecar transcription. Standalone live transcript has its own backend setting above.
Whisper model
Parakeet backend
Checking Parakeet backend…
Runs locally with a Parakeet backend binary plus model assets. See minutes setup --parakeet.
Transcription language
Leave this on Auto-detect for mixed or unknown audio. Pick a language only
when Whisper is consistently guessing wrong.
Speaker diarization (identify who said what)
AUTO-SUMMARIZATION
Automatic pipeline summarization
By default, Claude summarizes your meetings
conversationally via MCP — no configuration needed. Turn this on only if you want summaries generated
automatically in the recording pipeline.
Agent command
Ollama URL
Base URL for your local Ollama server.
Ollama model
Name of a model you've pulled in Ollama (e.g. llama3.2, qwen2.5, gemma3).
Connection type
Pick the closest match. Minutes will fill the technical fields below.
Connection URL
Paste the base URL ending in /v1, /openai, or /compat. Minutes adds /chat/completions automatically.
Model
Use the model name your server or gateway expects.
API key
Paste a key once. Minutes stores it securely.
Advanced: env var fallback
Optional. Use this only if you prefer launching Minutes with a shell-managed secret.
PIPELINE HOOKS
Post-record command
Shell command run after each recording is processed. The transcript file path is appended as the final argument. Leave blank to disable.
CALL DETECTION
Detect calls automatically
Polling interval
How often to check for active calls (seconds).
Cooldown
Minutes to wait before re-triggering call detection.
Google Meet detection (experimental)
Opt-in browser integration for Meet in Chrome/Safari. Uses stronger
meeting-URL matching, may require macOS browser automation permission, and backs off after failures
instead of probing every second.
Microsoft Teams (web) detection (experimental)
Opt-in browser integration for Teams meetings in Arc, Chrome, and Safari.
Matches Teams meeting URLs only (chat/calendar pages are ignored). Requires macOS browser automation
permission on first use, and backs off after failures instead of probing every second.
Stop when call ends
When the detected call app exits mid-recording, show a countdown prompt
with Stop now and Keep recording. Only applies to recordings started from the call-detection banner.
Auto-stop countdown
Seconds to wait before stopping when the call ends.
PRIVACY
Hide windows from screen sharing
Meeting-adjacent desktop context
Off. App/window context stays disabled until you turn it on.
Focused window titles
Uses the best available platform APIs for focused-window titles. On macOS, missing Accessibility permission limits title capture.
Browser title context
Opt-in browser tab and window titles only. URLs and domains are not collected in v1.
Allowed apps
Optional comma-separated allowlist. Leave empty to allow all apps except denied ones.
Denied apps
Comma-separated app or bundle-id fragments to suppress from desktop context capture.
Capture screen context during recordings
Screenshots sent to vision LLM for context. Requires Screen Recording
permission.
Screenshot interval
VAULT SYNC
Connect an Obsidian or Logseq vault
Checking...
RECALL AGENT
AI agent for the Recall panel
Auto-approve all tool calls
Extra launch flags
Additional CLI arguments passed when starting the agent
STORAGE
Loading...
DIAGNOSTICS
System readiness and audio permission checks
Retry failed captures and unprocessed audio files
ADVANCED
Edit config.toml
Tuning knobs not shown in this panel — silence thresholds, diarization parameters, watcher paths, knowledge-base integration, and more — live in your config file.
Configuration reference
Full list of every config key, its default, and what it does.