What kcap is

kcap stores the sessions your coding agents run, so that you, your teammates and the agents themselves can read them later. Each stored session feeds the next one: the decisions it made, the plan it left unfinished, the lesson an evaluation drew from it

The three things it does

Stores the work. A small hook in each coding agent sends the session to your workspace as it happens: your prompts, the agent’s replies, every tool call and its result, the commits it made. You can watch a session live, read it afterwards, and share it with a teammate who can then read it too

Hands the work back to agents. Your agent gets tools to search past sessions, resume an unfinished plan, see who else is working in the repository, file its session on a ticket, and save something the team should remember. The next session starts with what the last one learned

Lets you run agents from anywhere. A daemon on any machine you own turns it into a place to run agents from your browser or the desktop app. A second agent can review your work before you merge, and a flow can run several agents against one task until they agree it is done

How you use it

There are four separate ways in. They all talk to the same workspace, so use one or all of them

Way in	What it is
Your coding agent	Where most of kcap happens. You ask in plain language; the agent uses kcap’s tools to answer. Nothing to open
The web app	https://<your-workspace>.kcap.ai in a browser: sessions, agents, work items, knowledge, evaluations, insights
The desktop app	A Mac app for setting up, running hosted agents and following sessions and pull requests without a browser
The kcap command	A terminal utility: setup, status, recaps, imports, and everything a script or a CI runner needs

How sessions are organised

A session started inside a git repository is filed under that repository, found from the directory’s git remote; one started outside a repository is stored without one and shows under All Repositories. Repositories are how the web app groups sessions, how your agent searches them, and how sharing defaults and exclusions are expressed. On Team and Enterprise plans, a project groups several repositories that belong to one effort, with its own members, so a session can be shared with “everyone on this project” and knowledge, evaluations and insights can be read across the group

Supported coding agents

Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot CLI, Kiro, OpenCode, Pi, Antigravity, Muse Code, Grok Build and Mistral Vibe. Every one is stored and importable; most can also run as hosted agents. The per-agent details are in Your coding agents

Further reading

• Install kcap if you are setting up your own machine
• Join a kcap workspace if a teammate invited you
• What you can do if you want to see it work before reading anything else