Your agents, right where your meetings are

Your agents are good at drafting and pulling structure out of text, but this afternoon's call sits in a notetaker's cloud dashboard. Some notetakers expose an MCP server or an API, so the meetings are reachable, just somewhere else: you switch apps, wait for the transcript, and connect from outside.

Earkeep treats your meeting history as a local resource that any agent you choose can be pointed at, inside the app or over MCP. And the agent is your agent, on the plan you already have, with no per-meeting AI fee on top. That is what lets Earkeep be a one-time €39 license.

Ten agents run inside Earkeep today, with nothing to set up beyond picking one: Claude Code, Codex, Cursor Agent, Gemini CLI, Hermes, Kimi CLI, OpenClaw, opencode, Pi, and Qwen Code. Earkeep integrates with the protocol, so any other ACP agent works the same way.

In-app agents via ACP. Through the Agent Client Protocol, an agent runs inside Earkeep with your local transcripts as context, and its session is kept beside the meeting.

A built-in MCP server. Tools that aren't ACP agents, like Claude Desktop or Cursor, connect to Earkeep's MCP server as they would to any other data source.

Or no protocol at all. Transcripts are plain files you own, so anything that can open a text file can read them.

A summary is the least interesting thing an agent can do with a meeting. With your own tools attached, it can make the artifact the meeting was supposed to produce, while the call is still fresh:

• The follow-up email, or the notes themselves, from what was actually said.
• A spec or PRD draft out of a discovery call. For product managers →
• Backlog items with acceptance criteria, worded the way the team agreed. For product owners →
• A working prototype of what the room just described. For UX designers →
• Findings with the quotes still attached, across a week of interviews. For UX researchers →
• A status report, or the list of what was promised, across every call. For project managers →
• What you agreed over a quarter of 1:1s. For people managers →
• The decision trail from incident calls and design reviews. For engineering leads →
• A report that joins what was said to your own data, if your agent has connectors to it.

• Private by design: why audio never leaves the device, and why transcripts are plain files.
• Always-on capture: how spans become the history your agents read.
• Local RAG: querying a local corpus with a local model.