FastClaw / Run the model
Architecture
The core architecture of FastClaw is Gateway + Store + Agent Manager + Tool Registry + Channel Manager + Project Runtime.
core layer
gateway.New will open store, workspace, channel, cron, webhook, plugin, taskqueue, usage/quota and user spaces; Agent Manager then binds provider, memory, skills, tool registry, sandbox, MCP and other capabilities to each Agent.
| Components | Responsibilities |
|---|---|
| Gateway | Process-level orchestration, store, channels, cron, plugins, usage, runtime |
| Store | user, API key, Agent, session, project, files, config, cron, usage |
| Agent Manager | Build Agent loop according to user/agent, bind provider, memory, tools, skills |
| Tool Registry | Built-in tools, provider-backed tools, MCP tools, plugin tools |
| Project Runtime | web app template, dev server, preview, logs, file location |
data directory
The directory model for README is simple: the database is the source of truth for most configuration and runtime state, and skills remain in file system form. The default is SQLite for a single machine, and Postgres can be used for multiple copies.
~/.fastclaw/
fastclaw.db
skills/
agents/
<agentId>/agent/skills/Cloud model assumptions
In the source code, IsCloudMode is always true, and multi-user capability is the default path. Even if you only use it locally, you must understand it according to the multi-user, multi-Agent, and multi-session isolation model.
- Identity.CanAccessAgent will verify the agent ACL and return 404 when the authority is exceeded to avoid leaking the existence.
- API server, Dashboard and Channels all follow the same set of store and identity boundaries.
- Redis Optional; without Redis, use DB-backed channel lease.