Thclaws
Open-source AI agent harness in native Rust — GUI, CLI, headless, and webapp from one binary. Multi-provider, MCP, skills, plugins, agent teams.
Last verified:
What is Thclaws?
thClaws is a native-Rust AI agent workspace that runs locally on your machine, enabling users to code, automate workflows, search knowledge bases, and coordinate teams of agents through natural language. It provides four interfaces from one binary: a Desktop GUI with Terminal, Chat, and Files tabs; a CLI REPL for headless servers; non-interactive one-shot mode for scripts and CI pipelines; and a webapp mode accessible over WebSocket for SSH-tunneled remote access.
Key features include multi-provider support (Anthropic, OpenAI, Gemini, Alibaba DashScope, OpenRouter, Ollama, and more), open standards like MCP for tools and AGENTS.md for project instructions, reusable skills that trigger automatically, MCP server integrations for GitHub/filesystems/databases/Slack, a plugin system bundling skills/commands/agents/MCP servers, persistent memory stored as markdown, per-project knowledge bases using grep-based search without embeddings, agent teams for multi-process orchestration, filesystem sandboxing for safety, offline-capable operation with local Ollama models, plan mode with audit-reviewable steps, scheduled cron jobs, long-running loops with /goal --auto, native document workflow for PDF/DOCX/PPTX/XLSX, hooks on 8 lifecycle events, and an official multi-arch Docker image.
thClaws is designed for developers, researchers, product managers, ops teams, legal professionals, marketing, and finance — anyone who needs AI assistance. Engineers use the Terminal tab for raw REPL access, while chat-oriented users leverage the Chat tab for natural-language interaction. It's ideal for teams wanting sovereign-by-design AI that keeps code, prompts, and conversations on their machine without cloud lock-in.
Thclaws pricing
Pricing model: Freemium
The client is free and open source forever under dual MIT/Apache-2.0 license. There is no paid tier for the core thClaws software. Your only costs are API usage fees for the LLM providers you choose (Anthropic, OpenAI, Gemini, etc.) when using cloud models. Running locally with Ollama is completely free with no API key required. Enterprise Edition, hosting, and support are commercial offerings available through Agentic Press (agentic-press.com) for organizations needing OIDC SSO, IT/security administration, and commercial support. Hosting via Agentic Press partners with SIS Cloud Service and Artech.Cloud for deploying agents and web apps.
Thclaws pros
- Open-source and free forever for the client
- Runs entirely locally with no cloud dependency when using Ollama
- Multi-provider support lets you switch models mid-session
- No provider lock-in — your prompts and skills are portable
- Four interfaces (GUI, CLI, one-shot, webapp) from one binary
- Native Rust implementation for performance and reliability
- Filesystem sandbox limits file tools to working directory
- Persistent memory stored as editable markdown files
- Knowledge bases use simple grep without embeddings to manage
- Agent teams enable multi-process coordination via shared mailbox
- Plan mode requires approval before executing multi-step work
- Scheduled jobs survive reboots via native daemon (launchd/systemd)
- Native PDF/DOCX/PPTX/XLSX read/edit/create in one conversation
- Shell escape with ! prefix runs commands without token cost
- Side-channel agents run concurrently without polluting main history
- /dream consolidator mines recent sessions for new insights
- Official Docker image supports linux/amd64 and linux/arm64
- API keys stored in OS keychain by default for security
- Hooks on 8 lifecycle events enable audit and gating workflows
- Session resume picks up exactly where you left off
Thclaws cons
- GUI requires Wayland/webkit2gtk dependencies on Linux headless servers
- No built-in hosting — must use Agentic Press or external host separately
- Knowledge bases use grep instead of vector search, limiting semantic capabilities
- Learning curve for advanced features like agent teams and plugins
- Requires Rust 1.85+, Node.js 20+, pnpm 9+ to build from source
- No official mobile app or tablet interface
- Enterprise features like OIDC SSO require separate Enterprise Edition
- Documentation is comprehensive but spread across multiple manuals
- Only 1k GitHub stars indicates smaller community than competitors
- Per-step retry budget in plan mode can delay execution significantly
Frequently asked questions about Thclaws
What is thClaws and how is it different from other AI agents?
thClaws is a native-Rust AI agent workspace that runs locally on your machine, codes, automates workflows, searches knowledge bases, and coordinates agent teams. It differs through multi-provider support (switch mid-session), open standards (MCP, AGENTS.md, SKILL.md), sovereign-by-design architecture keeping data on your machine, four interfaces from one binary, filesystem sandboxing, offline capability with Ollama, and no provider lock-in. Built in Thailand by ThaiGPT Co., Ltd.
Is thClaws free to use?
Yes, the thClaws client is free and open source forever under dual MIT/Apache-2.0 license. You only pay for LLM API usage when using cloud providers like Anthropic or OpenAI. Running entirely locally with Ollama costs nothing — no API key, no network required. Enterprise Edition with OIDC SSO and commercial support is a separate paid offering.
Which AI model providers does thClaws support?
thClaws supports Anthropic (native + Claude Agent SDK), OpenAI (Chat Completions + Responses/Codex), Google Gemini & Gemma, Alibaba DashScope (Qwen), DeepSeek, Z.ai (GLM Coding Plan), NVIDIA NIM, NSTDA Thai LLM (OpenThaiGPT, Typhoon, Pathumma, THaLLE), OpenRouter, Agentic Press, Azure AI Foundry, Ollama (local/local Anthropic-compatible/Ollama Cloud), LMStudio, plus a generic OpenAI-compatible slot for LiteLLM/Portkey/Helicone/vLLM/internal proxies. Auto-detected by model name prefix.
How do I install thClaws on my machine?
Download pre-built binaries from GitHub Releases or thclaws.ai/downloads. macOS: universal .dmg (Apple Silicon + Intel) or architecture-specific tarballs. Windows: .msi installer (sets PATH automatically) or portable ZIP. Linux: x86_64 or aarch64 tarballs. Docker: pull thclaws/thclaws:latest (multi-arch). Build from source requires Rust 1.85+, Node.js 20+, pnpm 9+. First run prompts for secrets backend (OS keychain or .env).
What are skills and how do I use them?
Skills are reusable expert workflows packaged as a directory with SKILL.md plus optional scripts. The agent automatically picks the right skill when a request matches the whenToUse trigger, or you can invoke explicitly as /<skill-name>. Install from git URL or .zip archive with /skill install. Skills follow open standards and are portable across compatible harnesses.
How does thClaws handle memory and project instructions?
Drop AGENTS.md or CLAUDE.md in your repo — thClaws walks up from cwd and injects every match into the system prompt. Persistent memory store holds longer-lived facts classified as user/feedback/project/reference, stored as markdown you can read, edit, or commit. Sessions live as JSONL under .thclaws/sessions/ — git-friendly and grep-friendly.
What are knowledge bases and how do they work?
Per-project and per-user wikis the agent searches and reads on demand. Drop markdown pages under .thclaws/kms/<name>/pages/, give each a one-line entry in index.md, and the agent gets a table of contents every turn plus KmsRead/KmsSearch/KmsWrite/KmsAppend/KmsDelete tools. No embeddings — uses grep + read following Andrej Karpathy's LLM-wiki pattern. Run /dream to consolidate insights.
How does thClaws ensure safety and security?
Filesystem sandbox scopes file tools to working directory. Destructive shell commands flagged before execution. Approve every mutating tool call unless opted into auto-approve. Permission requests label which agent asks (main/side-channel/subagent). API keys stored in OS keychain by default (macOS Keychain/Windows Credential Manager/Linux Secret Service) with .env fallback. Hooks enable audit on 8 lifecycle events with timeout-SIGKILL guarantees.
What are agent teams and when should I use them?
Agent teams orchestrate multiple thClaws processes through a shared mailbox and task queue, each in its own tmux pane and optional git worktree. One agent writes backend while teammate builds frontend in parallel; lead merges branches when done. Use for complex projects requiring parallel work streams. Lead and teammates coordinate via task queue and git worktrees, with /agent for concurrent side-channel specialists and Task tool for model-driven subagents up to 3 levels deep.