Your agents, right where your meetings are

You already run agents. Maybe it's Claude Code for a coding task, maybe it's a knowledge agent you point at project docs. Those agents are good at drafting, summarizing, and pulling structure out of text, but they don't know what happened in this afternoon's call, because that transcript is sitting in a notetaker's cloud dashboard, not anywhere your agent can reach it.

Cloud notetakers try to close that gap from their end. Most now ship their own built-in assistant, and several expose an MCP server or an API you can point your own agent at, so it isn't that your meetings are unreachable. It's that they're somewhere else: you finish the call, switch to another app, wait for the transcript to land in the vendor's cloud, and connect an agent to it from outside. The work happens away from the meeting, which is exactly when you have the least appetite for it.

Earkeep starts from the other end: your meeting history is a local resource, and any agent you choose to run can be pointed at it. Run one inside Earkeep and you go from "the call just ended" to a drafted artifact without leaving the app or uploading anything. Prefer your agent where it already lives? Earkeep's own MCP server hands the same meetings to it from outside, so the choice is yours rather than the vendor's.

It also means you don't pay twice for the same intelligence. The agent you point at a meeting is your agent, on the plan you already have with it, and Earkeep neither resells model access nor charges a per-meeting AI fee on top. That's what lets it be a one-time €39 license rather than a subscription: what makes this category recurring is somebody else's inference bill, and here there isn't one to pass on to you.

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, the same ten whose marks sit at the top of this page.

Because what Earkeep integrates with is the protocol rather than ten separate agents, any other agent that speaks ACP works the same way, including ones that don't exist yet.

In-app agents via ACP. Earkeep speaks the Agent Client Protocol, so an agent runs inside Earkeep itself, with your local transcripts as its context and its whole session kept beside the meeting it was about. Which agents, in the section above.

A built-in MCP server for everything else. Earkeep also runs an MCP (Model Context Protocol) server, so tools that aren't ACP agents, like Claude Desktop or Cursor, can connect and read your meeting history the same way they'd connect to any other MCP data source. You don't have to run your agent inside Earkeep to use your meetings with it.

Or no protocol at all. Your transcripts are plain files you own, sitting in a folder you picked. Anything that can open a text file can read them, so a tool that speaks neither ACP nor MCP still doesn't need an integration built for it.

A summary is the least interesting thing an agent can do with a meeting. What runs inside Earkeep is a working agent with your own tools attached, so the output can be the artifact the meeting was supposed to produce, made while the call is still fresh:

• The follow-up email, or the notes themselves, written from what was actually said rather than from what you remember an hour later.
• A spec or PRD draft out of a discovery call, with the requirements someone stated out loud already in it. For product managers →
• Backlog items with acceptance criteria from a refinement session, worded the way the team agreed rather than the way you paraphrased it afterwards. For product owners →
• A working prototype of what the room just described. A coding agent that can read the critique can also build the screen it asked for. For UX designers →
• Findings with the quotes still attached, grouped across a week of interviews instead of one session at a time. For UX researchers →
• A status report, or the list of what was promised, gathered across every call where the project came up. For project managers →
• What you agreed with someone over a quarter of 1:1s, so the record is the record and not your memory of it. For people managers →
• The decision trail from a run of incident calls and design reviews, so being the person who remembers stops being part of the job. For engineering leads →
• A report that joins what was said to your own data, if the agent you run has connectors to the systems holding it. A meeting becomes one more source it can read, beside the database and the repo.

• Private by design: the audio-never-leaves-the-device mechanism this page's agent access sits on top of, and why transcripts are plain files anything can read.
• Always-on capture: how spans turn into the transcript history your agents read.
• Feeding meeting context to AI agents: the practical guide, with prompts and workflows.