Welcome to Jian
Understand the system, build your agents, and operate your own gateway.
Jian is a self-hosted agent gateway with a desktop app and a web panel. It gives agents persistent identities, conversations, memories, skills, tools, and scheduled work. A model supplies intelligence; the gateway supplies the runtime and keeps the state.
Jian is free and open source, licensed under Apache 2.0. There is no Jian subscription or paid feature tier. Connected model services, optional integrations, and server hosting can have their own costs.
Choose your starting point
| You want to… | Start here |
|---|---|
| Install Jian and talk to an agent | Your first agent |
| Understand what runs where | Architecture |
| Use the gateway built into the app | Desktop and local gateway |
| Connect the app to an existing server | Remote gateways |
| Let an agent operate another machine | Jian Remote and Computer Use |
| Run an always-on installation | Self-hosting |
| Build an integration | HTTP API |
The pieces, in one minute
- Gateway: runs agents, stores their state, authenticates requests, connects providers and integrations, and coordinates work.
- Desktop app: starts a local gateway automatically and can also connect to server gateways. Switching gateways switches the installation you are looking at.
- Web panel: a browser interface served by
jian-web, which forwards API traffic to the gateway. It does not execute the agent itself. - Jian Remote: a worker installed on a computer. It receives tasks from a gateway and performs shell, file, and desktop operations on that machine.
- Agent/profile: one persistent identity within a gateway. Its conversations, memories, and skills belong to that profile.
Connecting the app to a server and connecting a machine to that server solve different problems: the first gives you an interface; the second gives agents a place to act.
Build an agent that improves with use
Start with identity and instructions, then connect models. Use memories for durable facts and skills for reusable procedures. Automatic learning reviews useful experiences and can write both.
Conversations and context explains history, compaction, model selection, and long-running objectives. Collaboration distinguishes asking another agent from delegating to an internal worker.
Put it to work
Use tools, GitHub projects, MCP servers, and files and media for concrete tasks. Use schedules for a time and Listening for a GitHub state change. Channels brings server agents to WhatsApp, Telegram, and API clients.
Operate it deliberately
Read security and protection before granting machine or service access. For a server, keep configuration, backups and upgrades, and troubleshooting close at hand.
These guides describe the 1.1 release line, checked against the 1.1.4 source. Paths, limits, and protocol examples below refer to that implementation. Your gateway exposes its own versioned HTTP contract at /openapi.json; consult it when integrating a different release.