Welcome to our Documentation 🦋
Kaja is a customisable, multi-purpose AI agentic1 harness2. Give it a task and it keeps looping with an LLM – calling tools, switching personas, remembering what matters – until the job is done.
There are two ways to run it:
| Cloud | Local | |
|---|---|---|
| Agent loop runs | On api.kaja.io | On your machine |
| Needs an account | Yes (device login) | No |
| Needs your own LLM | No | Yes |
| Sessions & memory stored | Postgres, server-side | SQLite, in your home dir |
| Shell, your own MCP servers and plugins | No | Yes |
What’s in the box
- Front doors ― Ways to talk to the same agent:
- Terminal UI ― Console-based chat client.
- Telegram ― Minimal chat client on the go.
- Widget ― Customised website clients.
- Web app ― Your cloud account: abilities, keys, widgets and usage.
- Abilities ― From the marketplace:
- Personas ― A named character with its own instructions, and optionally its own model.
- Skills ― Instructions the agent loads when a request matches it.
- Tools ― APIs the agent can call, like weather or search.
- MCP servers ― Ready-made tool sets, like a browser or library docs.
- Memory & datasets ― Long-term notes and structured questionnaires.
- Configuration ― The TOML files behind local mode.
Data safety
In local mode, conversation history and memory never leave your computer. Configs are plain text files and everything else lives in a single SQLite file. Run your own models locally and no internet connection is needed at all.
In cloud mode, sessions and memory are stored server-side against your account. See the Privacy Policy for what that means in practice.
Next: