Agents that work the record,
on the model you choose.
Vexa’s agents read your meetings and write to your workspace. Anything a meeting, email or web page starts, a person approves.
- Your meetings
- Agent drafts
- A person approves
- Your workspace
After the meeting
When a meeting ends, an agent drafts a write-up, and a person approves it.
Ask the record
Ask an agent about a client or a deal: it answers from your workspace’s pages.
- Sep 26 · Moved the review to the risk committee.Transcript
Proposals, not writes
What an agent may do depends on where its input came from.
- Started by you, in chat
- It writes directly, and git is the undo.
- Started by a transcript, an email or a web page
- Read-only: it can only propose, and a person approves.
The model you choose
Vexa can run on your own model: transcription runs on its bundled unit or any OpenAI-compatible endpoint you host, and its agent reaches a model you host through an OpenAI-compatible gateway. The agent itself runs Anthropic’s Claude Code, with your firm’s own key or pointed at that gateway.
In its own container
Each agent runs in an isolated container, removed when idle.
Your own agent instead
Give your agent Vexa’s Model Context Protocol (MCP) server, and it can send a bot and read your meetings.