Debrief
Use cases / Developers
Talk to your
coding agent.
You explain a bug faster than you type it. MeetingDebrief turns what you say into prompts in any terminal or editor, and keeps a searchable, agent-readable record of the meetings where the decisions were made.
The problem
The context lives in your head and in last week's meeting.
Good prompts are long. Typing three paragraphs of context into a terminal is the part of AI-assisted coding nobody enjoys, so the prompts get short and the results get vague.
Meanwhile the reasons behind the code sit in calls: why the API paginates by cursor, who owns the migration, what was agreed in the incident review. None of that reaches your editor.
How it works for you
Talk to your coding agent
Hold the dictation shortcut in Claude Code, Cursor, or any terminal and describe the change. Let go and the prompt is typed in. Paste a stack trace first; your clipboard is restored after dictating, so the log is still there to paste underneath.
Record the meetings around the code
Planning, design reviews, and incident calls are recorded from your microphone and system audio, transcribed on your machine, and summarized into decisions and action items with owners.
Hand your meetings to your tools
The built-in MCP server gives Claude Desktop, Claude Code, and Cursor read-only access to your transcripts and summaries. Ask your agent what was decided about the API before it writes a line.
What you get
Dictation into any text field
Editors, terminals, chat boxes, PR descriptions. Hold to talk, tap to toggle, or both. Esc throws a recording away.
Vocabulary for your stack
Teach it Kubernetes, Terraform, and your internal service names once. Dictation and meeting transcripts both spell them the way you mean them.
Built-in MCP server
A ready-to-paste config in Settings wires up Claude Desktop, Claude Code, or Cursor. Read-only, runs on demand as a local sidecar, opens no network ports.
Local HTTP API
Automate recordings, search, and vocabulary from your own scripts. Swagger docs included. Bound to localhost unless you choose otherwise.
Markdown on disk
Session Mirror writes every transcript and summary to Markdown files, so grep, git, and Obsidian can read them too.
Offline by design
Parakeet and Whisper run locally with Metal on Apple Silicon and CUDA on NVIDIA. After setup, nothing needs a network connection.
Prompts at the speed of speech.
A 15-second dictation is typed out within a fraction of a second of letting go. Every dictation is kept in a searchable history, so the prompt you wrote an hour ago is one search away.
Your meetings, as context.
Ask a meeting a question and get an answer with timestamp citations. Or let your coding agent ask it over MCP, without you copying anything.
A day with MeetingDebrief
-
09:30
Standup gets recorded. By the time your coffee is refilled, the summary lists who is blocked on what.
-
10:15
A test fails. You copy the log, hold the shortcut, and say "take a look at this error, the migration renamed the column". Claude Code gets the prompt; the log is still on your clipboard.
-
14:00
Design review for the new API. Action items come out with owners and dates.
-
16:45
Your agent asks MeetingDebrief over MCP what the review decided about pagination, then writes the endpoint to match.
-
17:30
You dictate the PR description instead of typing it. It is the most thorough one you have written this month.
FAQ
Does dictation work in the terminal?
Which AI tools can read my meetings?
Will it spell framework and product names correctly?
Does anything leave my machine?
Other use cases