Docs
Run KyoubeAI, and build on it.
How to install and run KyoubeAI on your own server, and how it works underneath, from its security model to the apps agents build on it.
These pages are the documentation from the KyoubeAI repository, as of version 1.11.0.
Get started
- Install KyoubeAIInstall KyoubeAI with one script and Docker: what you need, what install.sh does, claiming the instance, the first-run wizard and troubleshooting.
- Agent harnessesInstall the harnesses your agents work through from the Terminal: Claude Code, Codex, Hermes Agent, pi, Gemini CLI, OpenCode and Kimi Code.
- Using KyoubeAIWhat you find once KyoubeAI runs: the Studio layout, agent working rules, the phone app, the Terminal, a database per company, apps and project files.
- KyoubeAI on your phoneInstall KyoubeAI as an app on phones and computers and turn on push notifications, sent by your own instance with no store app in between.
- Update KyoubeAIMove a KyoubeAI install to a new release with update.sh, which backs up first, and go back to the previous release with one command if needed.
- Back up and restoreBack up both KyoubeAI databases, the Kyoube roles and the home volume with one script, and restore them on the same machine or a new one.
Run it
- Operating KyoubeAIDay-to-day running: volumes, nightly and off-site backups, restores, logs, health checks, rotating the board API key and resource limits.
- Access governance for agentsThe two gates on an agent's calls to Kyoube tools, and the defaults to set: a read-only profile, approval for dangerous calls and limits on bulk writes.
- User groupsDecide which people may use which agents and apps, and at what data level, with groups that owners and admins manage in Company Settings.
- Agent working rulesThe rules that keep agents working the way a small team expects: each agent finishes its own task, and approvals go to the person who started it.
- LicensingKyoubeAI is free for up to five users on an instance. How the limit works, applying a licence key, expiry, and freeing a seat.
- Security policyHow to report a vulnerability, and KyoubeAI's security model: trust zones, the Terminal, data, apps, files, telemetry and secret rotation.
Build on it
- Building KyoubeAI appsHow to write a KyoubeAI app: one HTML document and a manifest, the window.kyoube SDK for company data, and the sandbox the app runs in.
- ConnectionsLet apps and agents call outside services with a key the company stores once: setting a connection up, who may call it, the limits and the log.
- App tasksLet a Kyoube App ask one of the company's agents to do a piece of work: templates, what the agent receives, who may start a task, and the limits.
- Typed decisionsLet agents and apps ask a fast judge model closed questions and branch on the answers: providers, the guardrail, what leaves the server, AI columns.
- RoadmapWhat was left out of KyoubeAI 1.0, roughly in the order people ask for it: multi-file apps, realtime table updates, and more.