Meetings
Shared spoken notes from customer calls, one-on-ones, and team meetings.
This is where decisions, questions, and follow-ups show up.
Agent memory docs
Local spoken context your agent can actually use.
Transcripted turns meetings, dictation, and audio files into durable local Markdown: readable text for people, YAML frontmatter for tools, and capture folders agents can search.
Short answer
Transcripted is a local memory layer for spoken work. It records or imports audio on your Mac, transcribes it locally, and saves meetings and dictations as Markdown files with structured frontmatter that agents can search later.
What becomes memory?
Dictations
Private voice notes from quick thoughts, recaps, and post-meeting follow-ups.
This is where you save what you think happened and what to do next.
Together
Meetings and dictations give your agent both the shared record and your own notes.
Transcripted starts with meetings and gets better as you add the voice notes that connect the dots.
The artifact model
Each capture is useful to humans and machines at the same time.
Readable transcript
A clean transcript with timestamps and speaker names.
YAML frontmatter
Dates, duration, capture type, engines, speakers, and source details live at the top of the Markdown file.
Capture folders
Meetings and dictations are saved in predictable local folders that agents can search.
Persistent speaker identity
The same person can stay linked across meetings.
What the files look like
These examples are intentionally plain. Agents do better when the interface is boring, stable, and readable.
Meeting Markdown
---
capture_id: "0B9F..."
capture_type: meeting
date: 2026-04-22
time: 15:00:00
duration: "32:14"
transcription_engine: parakeet_local
diarization_engine: pyannote_offline
sources: [mic, system_audio]
title: "Customer onboarding call"
speakers:
- id: "0"
channel: system
name: "Maya Chen"
---
# Meeting Recording - Apr 22, 2026
## Full Transcript
[01:14] [System/Maya Chen]
The onboarding checklist is still where teams slow down.
[03:48] [Mic/You]
So the next step is a shorter first-run flow and a better export.Daily dictation Markdown
---
title: "Dictations for April 22, 2026"
date: 2026-04-22
capture_type: dictation_day
---
# Dictations for April 22, 2026
## 9:15 AM - Follow up on onboarding
Entry ID: `dictation-20260422-091500-000`
Captured: 2026-04-22T09:15:00.000Z
Source app: Notes
Delivery: pasted
Words: 12
Follow up on the onboarding export before changing pricing.Default folders
~/Library/Application Support/Transcripted/captures/meetings/
~/Library/Application Support/Transcripted/captures/dictations/
# Optional Claude Desktop MCP helper:
~/Library/Application Support/Transcripted/mcp/transcripted-mcpFour access modes
Start simple. Add structure only when it helps.
Files first
Point Claude, Cursor, Obsidian, or any file-reading tool at your local Transcripted folder.
Starter prompt
Transcripted can generate a simple prompt that tells your agent where the files live.
Read-only MCP
On supported agents, the local MCP server lets you search, recap, and read meetings or dictations.
CLI and automation
Developers can use the standalone CLI context commands or script against the same local Markdown files.
Starter prompt
A clear prompt helps an agent treat the folder as a corpus instead of a pile of random files.
You have access to my local Transcripted corpus.
Read my saved Transcripted Markdown files directly:
Meetings:
- ~/Library/Application Support/Transcripted/captures/meetings/
Dictations:
- ~/Library/Application Support/Transcripted/captures/dictations/
Use YAML frontmatter for dates, duration, speakers, source app, and capture type.
Use timestamped transcript lines for source-backed answers.
When answering, cite the meeting title and date when possible.
Prefer decisions, owners, open questions, and follow-ups over generic summaries.Questions this unlocks
Why this is different from cloud notes
Private by default
Audio stays on your Mac.
No meeting bot
Transcripted records from your Mac. It does not join as a bot.
Open source and free
MIT licensed, no subscription, no paid tier, and no account required.
Built for repeated context
Speaker identity makes old meetings more useful over time.
Fast on Apple Silicon
The local pipeline is designed for M-series Macs and keeps transcription fast enough to fit into normal workflows.
Yours to keep
Your transcripts, audio archive, and app-owned state stay on your Mac.
Next steps
FAQ
What is agent memory in Transcripted?
Agent memory in Transcripted is a local spoken-context corpus made from meetings, dictations, and audio files. The current app saves meetings and dictations as local Markdown with structured YAML frontmatter.
How do agents access Transcripted data?
Agents can read the local Markdown folders directly, use a starter prompt, connect through the read-only local MCP server on supported clients, or use the standalone CLI context commands.
Does Transcripted require cloud storage for agent memory?
No. Transcripted starts with local files on your Mac. You can choose to sync or share them, but cloud custody is not required for the core workflow.
Why does persistent speaker identity matter?
Persistent speaker identity lets agents reason about the same person across meetings instead of treating every transcript as an isolated conversation.