Under the hood
This section is for the curious and for contributors — how Kiki actually works inside. You don't need any of it to use Kiki or to build apps, but if you want to understand the machine (or hack on it), start here.
Kiki is built across several open-source repositories:
| Part | What it is | License |
|---|---|---|
| kiki-os | The OS images (Fedora bootc). | GPL-3.0 |
| kiki-wm | The desktop compositor and shell. | GPL-3.0 |
| kiki-agent | agentd, the agent daemon. | MIT |
| kiki-sdk | The Rust SDK for building apps. | MIT |
| kiki-app | The mobile companion. | MIT |
| kiki-cloud | The optional backend. | Proprietary |
See Licensing for the full policy.
In this section
- Architecture — the cross-cutting decisions: bootc, content-addressed everything, real-time sync.
- The agent loop — perceive, reason, act: the loop every agent runs.
- State & migration — how a session is checkpointed and moved between machines.
- Computer use — how the agent reads the screen through the accessibility tree, not screenshots.
Contributing
The OS, desktop, agent, and SDK are open source. Issues and pull requests are welcome on the GitHub repositories.