How to build agent memory from meetings
To build agent memory from meetings, save every meeting and dictation as local files, keep speaker identity consistent, and give your agent a structured way to search the archive. Transcripted creates that file-based memory layer.
Steps
- 1
Capture meetings and dictations
Use meetings for the shared record and dictation for your private recap, next steps, and synthesis.
- 2
Save more than a summary
Keep the clean note, full transcript, JSON sidecar, and corpus index so the agent has both readable and structured context.
- 3
Preserve speaker identity
Keep names tied to speakers across meetings. Agent memory gets stronger when it can connect the same person over time.
- 4
Point the agent at the archive
Start with files or a starter prompt. Use read-only MCP when your client supports a stronger query surface.
- 5
Ask cross-meeting questions
Use the archive to answer questions about decisions, blockers, commitments, people, objections, and follow-ups.
Agent memory needs durable inputs
A meeting summary is a useful artifact. It is not enough to become memory by itself.
Agents need repeated context, stable files, speaker identity, and a way to search across the whole corpus.
What belongs in the memory layer
Think of the archive as three layers: what humans read, what machines query, and what connects the whole history.
Transcripted writes all three layers from meetings, dictations, and audio files.
- Markdown for readable notes.
- JSON sidecars for structured utterances and metadata.
- transcripted.json for corpus-level discovery.
- Persistent speaker identity for people-aware recall.
Questions agent memory should answer
The value shows up when the agent can answer questions that cross meeting boundaries.
This is where a local archive becomes more than a folder of transcripts.
- What did we decide about pricing?
- What has this customer asked for over time?
- Who owns the next follow-up?
- Which objections keep repeating in sales calls?
Keep going
Common questions
What is agent memory for meetings?
It is a durable archive of meetings, dictations, transcripts, metadata, and speaker identity that an agent can use as context over time.
Do I need MCP for agent memory?
No. Files and prompts can work first. MCP is useful when you want a supported agent to search and read the local archive through a cleaner interface.
Why include dictation?
Dictation captures your private interpretation after the meeting. That helps the agent connect what was said with what you think should happen next.
Build memory from the conversations you already have.
Meetings, dictations, and audio notes can become local context your agent can reuse.