User manual

thClaws User Manual

34 chapters

A native-Rust AI agent workspace with CLI and desktop GUI. This manual covers everything from installation through building and deploying real projects — coding, automation, knowledge bases, and multi-agent teams.

Part I — Using thClaws

Chapter numbers are stable — they are the published URLs — so the groups below are a reading order, not a renumbering. Jump to whichever group matches what you are trying to do.

Getting started

# Chapter
1 What is thClaws?
2 Installation
3 Working directory & running modes
4 Desktop GUI tour

Everyday use

# Chapter
6 Providers, models & API keys
7 Sessions
10 Slash commands
11 Built-in tools
28 Browser automation

Keeping the agent in bounds

How you decide what the agent may do before it does it — and what leaves your machine.

# Chapter
5 Permissions
18 Plan mode
32 Thai PII masking

What the agent knows

# Chapter
8 Memory & project instructions (CLAUDE.md / AGENTS.md)
9 Knowledge bases (KMS)
20 Background research (/research)

Extending the agent

# Chapter
12 Skills
13 Hooks
14 MCP servers
16 Plugins
26 GUI Shells

Running several agents

# Chapter
35 Agents in a workspace
15 Subagents
17 Agent Teams
25 Workflows (/workflow run)
19 Scheduling
31 Loops and goals (/loop, /goal)

Reaching thClaws from elsewhere

# Chapter
21 LINE chat & web browser bridge
23 Telegram bot
24 Facebook Page Messenger bot
33 thClaws Remote
30 Job Artifacts (files in/out for orchestrators)

thClaws.cloud and managed copies

# Chapter
27 thClaws.cloud (Agent Templates + hosted + gateway)
34 Managed builds and org policy

Appendices

# Appendix
A Providers, models & prices (thClaws.cloud gateway)

Conventions used in this manual

  • ❯ is the REPL prompt; what follows on that line is what you type.
  • $ is a shell prompt outside thClaws.
  • [tool: Bash: …] / [tokens: Xin/Yout · Ts] lines show what thClaws prints back.
  • Code fences without a language are terminal output; fences with a language (rust, json, bash) are files you write or commands you run.
  • Bold inside a command label indicates a required input (e.g. name).
  • Every chapter is self-contained — skip around freely.