Quick Start
AgentX is two pieces: a server you run (agents, memory, tools) and a desktop client you point at it. Getting up and running is three moves:
- Run a server — pick your path below.
- Connect the client — installer + your server’s URL.
- Go further — expose it, tune it, or build on it.
1 · Run a server — pick your path
| Path | Best for | Guide |
|---|---|---|
| Self-host — Docker image + web dashboard, nothing else on the host | Running AgentX to use it | Self-Hosting |
From source — hot-reloading full stack (task dev) | Hacking on AgentX itself | Installation |
| Local clusters — several isolated prod-like instances, one dashboard | Power users & staging | Clusters & Gateway |
The self-host path is three commands, and the bundled deployment manager takes it from there — watch the first boot progress live, stream logs, start/stop, and see resource usage from a dashboard instead of a terminal:
tar xzf agentx-deploy.tar.gz && cd agentx-deploy
cp .env.example .env # fill 3 values — the file tells you which
docker compose up -d # then open http://127.0.0.1:123202 · Connect the client
Grab an installer from the latest release (Windows / Linux), enter your server’s URL on first run, and set the root password from the built-in setup screen. Done — you’re chatting.
3 · Go further
| You want to… | Go to |
|---|---|
| Reach your server from the internet (token gateway + tunnel) | Going public |
| Understand logins, sessions, and the gateway token | Authentication |
| Tweak environment variables & config files | Configuration |
| Drive everything from the ops dashboard / CLI | Deployment Manager |
| Script against the HTTP API | API Endpoints |
| See how it all fits together | Architecture Overview |
| Explore features (chat, memory, MCP tools, translation, prompts) | Chat · Memory · MCP · Translation · Prompts |