FastClaw / Get started
FastClaw Overview
FastClaw is a lightweight AI Agent OS written in Go that puts Agent Factory, sessions, tools, sandboxes, skills, channels and upstream APIs into the same runtime.
What is FastClaw?
FastClaw's README defines it as a lightweight AI Agent runtime: single binary, arbitrary LLM, multi-agent, sandbox, cloud-friendly. Rather than just providing a chat UI, it packages the infrastructure needed to create, manage, run, and integrate Agents.
- Agent Factory: Each Agent has its own SOUL.md, memories, skills, tools, sessions and run configurations.
- Runtime kernel: Responsible for LLM invocation, tool execution, sandbox isolation, session management, project preview and usage statistics.
- Integration layer: Provides access to Dashboard, OpenAI-compatible API, Channels, Cron, plug-ins and MCP tools.
what problem does it solve
If you want to embed AI Agent capabilities into your own products, you usually have to handle accounts, API keys, session isolation, tool permissions, model configurations, long-term memory, file space, channel callbacks, and usage limits yourself. FastClaw turns these common layers into a deployable Agent OS.
Connect Agent to SaaS
Use FastClaw as the Agent backend behind your product by isolating /v1/chat/completions and app_user.
Be a Coding Agent
Project Runtime can launch template applications, open previews, read logs, and locate file tools in the project directory.
Be a channel robot
The same Agent can be bound to Telegram, Discord, Slack, WeChat, Feishu, LINE and other channels.
Build an internal Agent platform
Dashboard, API key, provider, skill, plugin, cron and quota are combined into an operation and maintenance panel.
Recommended mental model
Think of FastClaw as Agent OS: the database is the control plane, the Agent runtime is the execution plane, skills/tools/plugins/channels is the extension plane, and the /v1 API is the stable interface for upstream application calls.
| Level | Function |
|---|---|
| Control plane | Dashboard, user, API key, Agent, Provider, Channel, Cron, usage configuration |
| Execution surface | Agent loop, LLM provider, tool registration, sandbox execution, Project Runtime |
| Extension surface | Skills, MCP, JSON-RPC plug-in, Channels, tool provider registry |
| Integration surface | OpenAI-compatible API, user isolation, Quota, Usage, Webhook |