Getting Started

Get xbot running in under 5 minutes.

Note
System requirements: Any modern OS (Linux, macOS, Windows). No pre-installed dependencies needed — the installer downloads a static binary.

1. Install

# Linux / macOS
curl -fsSL https://raw.githubusercontent.com/ai-pivot/xbot/master/scripts/install.sh | bash

# Windows (PowerShell)
irm https://raw.githubusercontent.com/ai-pivot/xbot/master/scripts/install.ps1 | iex

The installer downloads the xbot-cli binary, generates a random admin token, and writes ~/.xbot/config.json. For Server mode it also installs a system service and downloads the Web UI.

Note
Behind a firewall (China)? Use the mirror-accelerated installer:
curl -fsSL https://ghfast.top/https://raw.githubusercontent.com/ai-pivot/xbot/master/scripts/install-cn.sh | bash

2. Choose a mode

The installer asks you to pick:

	Standalone	Server
Architecture	CLI runs the agent locally	Background server + CLI connects remotely
Best for	Solo developer	Teams, multi-channel
Channels	CLI only	Feishu · QQ · Web · CLI
LLM sharing	Each user configures	Admin configures once, everyone uses
Persistence	Stops when terminal closes	System service, auto-start

Most teams should choose Server mode. Pick Standalone for a quick solo test drive.

3. Configure your LLM

Run xbot-cli. The first launch opens a Setup wizard:

Choose an LLM provider (OpenAI / Anthropic / OpenAI-compatible)
Enter your API key
Set the API base URL (change this for DeepSeek, Qwen, Ollama, etc.)
Pick a model
Configure model tiers (Vanguard / Balance / Swift — used for different scenarios)

Re-run the wizard anytime with /setup or Ctrl+K → Setup.

Subscriptions, not a single global key. xbot uses a subscription system. You can create multiple subscriptions (e.g. work Claude, personal DeepSeek) and switch between them per session.

Tip
Using DeepSeek, Qwen, or Ollama? Set provider: "openai" and change the base_url in the Setup wizard. xbot works with any OpenAI-compatible API.

4. Start chatting

You’re ready. Type a message and press Enter. The agent can call tools, run commands, search the web, and delegate to sub-agents.

Try these right now

You: What files are in this directory?

Agent: *uses Shell tool to run ls*
Agent: The current directory contains...

You: Read the README.md and summarize it

Agent: *uses Read tool*
Agent: This project is about...

You: Create a new file called notes.txt with today's date

Agent: *uses FileCreate tool*
Agent: Done! Created notes.txt with today's date.

You: Search the web for the latest Go release

Agent: *uses WebSearch tool*
Agent: The latest Go release is...

The agent picks the right tool for the job — you just describe what you want.

Type / to see all slash commands. A few essentials:

Command	What it does
`/setup`	Reconfigure LLM, sandbox, theme
`/context`	Inspect token usage
`/clear`	Clear the conversation
`/new`	Start a new session
`/sessions`	List / switch sessions
`/settings`	Open settings panels
`/help`	Show all commands

Essential keyboard shortcuts

Shortcut	Action
`Ctrl+K`	Command palette (fuzzy search)
`Ctrl+N`	Cycle model
`Ctrl+P`	Switch subscription
`Ctrl+T`	Sub-agent progress panel
`Ctrl+L`	Switch model (per-session)

Where to go next

Installation guide — build from source, service management
Configuration reference — every config.json field
Channels — Feishu, QQ, Web setup

Features — tools, skills, sub-agents, MCP, plugins
Sandbox guide — Docker sandboxing
Architecture — how it all fits together