Introduction
Kessho is a desktop application that keeps a structured memory for your AI agents: a folder of Markdown files that belongs to you. A simple, flat memory, where notes just pile up, grows out of control; Kessho's is organised into zones, projects and domains, with summaries and archives, so that its growth stays under control. Your AI agents read it and add to it; the application shows it to you, keeps it tidy and lets you edit it. This guide walks through the application, screen by screen.
The screenshots in this guide show a fictitious demonstration memory.
The ideas to know
The memory. It is a plain folder on your disk. It is portable: you can copy it, back it up, open it with other tools. It works with Obsidian, without depending on it.
Notes. Each note is a Markdown file. Kessho files notes in zones, which say what a note is for: a session, a piece of knowledge, a procedure, a principle, a goal, a person… See Zones.
Projects and domains. A note belongs to a project; a domain is a cross-cutting scope that brings several projects together, and a project can belong to several domains (see Zones). Each carries a summary you can hand to an agent so it picks the work back up: these are the context keys (see Home).
Agents. Your AI assistant talks to the memory through an MCP server, kessho-mcp, shipped with the application. Kessho declares it to the assistants installed on your machine (see Connect your agents). What the agent writes shows up in the application; what you change in the application, the agent reads.
Where to start
Getting started
- First launch: the five-step wizard.
- Choose your memory: where the folder lives, and how to change it.
- Connect your agents: declare Kessho to your assistants.
Screen by screen
- Home: the state of the memory at a glance, and the context keys.
- Zones: browse, filter and file notes.
- Atom and editor: read and write a note.
- Graph: see how notes relate to one another.
- Sync: back up the memory with git.
- Settings: appearance, memory, connectors.
- Help: find help inside the application.
And with an agent?
This guide covers what you do with the mouse and keyboard. For what you ask your agent to do — resume a session, archive, search — see With an agent. The tools themselves are described in the MCP reference.