OpenClaw FAQ

Real questions from Reddit, Hacker News, GitHub Issues, and the OpenClaw Discord. If your question is not covered here, check the What is OpenClaw? page for a full introduction or ask in the community.

General

What is OpenClaw?

OpenClaw is an open-source AI agent framework that connects large language models to your apps, devices, and services. Unlike a simple chatbot, it takes real-world actions — running commands, browsing the web, sending messages, controlling smart home devices, and more. It runs on your own hardware or on a managed platform. See the full What is OpenClaw? page for details.

Is OpenClaw free?

The software itself is MIT-licensed and completely free. However, most LLM providers charge per token, so running OpenClaw costs whatever you spend on API calls. Realistic monthly ranges:

• Light personal use (a few chats per day): $6 -- $13/mo
• Small business / power user: $25 -- $50/mo
• Heavy automation or team use: $100 -- $200+/mo

You can eliminate API costs entirely by running local models through Ollama, at the expense of weaker tool-calling capability.

What was OpenClaw called before?

The project has been renamed twice:

1. Clawdbot (November 2025) — the original name chosen by creator Peter Steinberger.
2. Moltbot (January 27, 2026) — a temporary rename after a trademark dispute with Anthropic over the "Claud-" prefix.
3. OpenClaw (January 30, 2026) — the final name, chosen three days later to signal open-source identity and avoid further trademark issues.

Who created OpenClaw?

Peter Steinberger, an Austrian developer best known for founding PSPDFKit. He released the first version in November 2025. When Steinberger joined OpenAI on February 14, 2026, the project was transferred to an independent open-source foundation to ensure its continued neutrality.

Setup and Hardware

Does OpenClaw work on Windows?

Only through WSL2 (Windows Subsystem for Linux). Native Windows is not supported because OpenClaw relies on Unix primitives like bubblewrap for sandboxing. WSL2 works fine for development and personal use, but is not recommended for production — use a Linux server or a managed platform instead.

How much RAM do I need?

• 1 GB: Absolute minimum. OpenClaw will run, but expect swapping under load.
• 2 GB: Practical minimum for a single-channel setup with a cloud LLM.
• 4 GB: Comfortable for multiple channels and moderate skill usage.
• 16 -- 32 GB: Required if you plan to run local models (Ollama). A 7B-parameter model needs roughly 8 GB of RAM; 13B needs 16 GB.

Can I run OpenClaw on a Raspberry Pi?

Yes. A Raspberry Pi 5 with 8 GB RAM is the recommended minimum for a usable experience. A Pi 4 with 4 GB can run it, but performance will be tight — expect slower response times and limited multitasking. Running local models on a Pi is not practical; stick to cloud LLM providers.

What is the best hardware for OpenClaw?

It depends on your goals:

• Simplest self-hosting: A Mac Mini M4 (16 GB) gives you excellent performance and can even run smaller local models.
• Budget server: A mini PC (Intel N100 or AMD Ryzen) with 8 -- 16 GB RAM. Fanless, low-power, runs 24/7.
• Cloud: A VPS with 2+ GB RAM from any major provider. Typically $5 -- $15/mo.
• Least effort: A managed platform like Claw Buff handles all infrastructure for you — no hardware decisions, no maintenance.

Models

Which AI models does OpenClaw support?

OpenClaw supports 20+ LLM providers out of the box, including OpenAI, Anthropic, Google Gemini, Mistral, Groq, Together, Fireworks, and more. About 80% of providers use the OpenAI-compatible API format, so adding a new one is typically just a baseUrl and API key — no code changes needed.

Can I use local models?

Yes. OpenClaw integrates with Ollama for fully local inference. Popular choices include Llama 3, Mistral, Qwen, and Gemma. Benefits: complete privacy, zero API cost, works offline. Tradeoff: local models are significantly weaker at tool calling and complex multi-step reasoning compared to frontier cloud models.

Does OpenClaw work offline?

Yes, if you use local models. Once the model weights are downloaded, no internet connection is required. Memory, skills, and tools all function offline. The only things that need internet are cloud LLM APIs and channel connections (Telegram, Discord, etc.).

How much does it actually cost to run?

This is the most common question, and the honest answer is: it depends entirely on your usage pattern. Realistic ranges:

Privacy and Security

Is OpenClaw safe?

The software itself is MIT-licensed and open-source, so you can audit every line. That said, security concerns are real:

• Exposed instances: Over 220,000 OpenClaw instances have been found publicly accessible on the internet with no authentication. If you self-host, you must configure authentication and firewall rules.
• CVEs: Security vulnerabilities have been discovered and patched. Keep your installation updated.
• ClawHavoc: A research project demonstrated that malicious skills could be uploaded to ClawHub. VirusTotal scanning is now active, but it cannot catch prompt injection attacks.

On managed platforms like Claw Buff, network isolation, authentication, and automatic updates are handled for you.

Does OpenClaw read my messages?

OpenClaw itself does not phone home or collect telemetry. However, your conversations are sent to whichever LLM provider you configure (OpenAI, Anthropic, etc.), and those providers have their own data policies.

For full privacy, run local models via Ollama. Your data never leaves your machine.

Where is my data stored?

• Persistent memory: Stored in MEMORY.md files on disk.
• Conversation history: Stored in a local SQLite database.
• Configuration: Stored in openclaw.json (JSON5 format).

All data lives on whatever machine runs OpenClaw. On managed platforms, data is stored in the provider's infrastructure.

Is OpenClaw GDPR compliant?

It depends on your setup:

• Self-hosted with local models: Yes — all data stays on your server, you are the data controller, and no data crosses borders.
• Self-hosted with cloud LLMs: Conversations are sent to US-based providers (OpenAI, Anthropic, etc.). You need to check each provider's Data Processing Agreement (DPA) and assess whether Standard Contractual Clauses (SCCs) cover your use case.
• Managed platform: Check the platform's own DPA and data residency policies.

Are ClawHub skills safe?

Not all of them. The ClawHavoc research project identified 1,184 malicious skills on ClawHub, including skills that exfiltrated environment variables and injected hidden instructions into agent prompts. In response, ClawHub now runs VirusTotal scanning on all uploads.

Channels and Usage

How many channels can run at once?

OpenClaw supports 12+ channels simultaneously — Telegram, Discord, Slack, WhatsApp, and others can all be active at the same time on a single instance. The practical limit is your server's RAM and the rate limits of each channel's API.

Does OpenClaw support voice?

Yes. OpenClaw supports text-to-speech (TTS) and speech-to-text (STT) on Telegram and WhatsApp voice notes. You can send a voice message, and the agent will transcribe it, process it, and optionally respond with a voice note. Voice support requires configuring a TTS/STT provider (e.g., OpenAI Whisper for STT, ElevenLabs or OpenAI TTS for synthesis).

Can I use OpenClaw for business?

Yes. The MIT license explicitly permits commercial use with no restrictions. You can deploy OpenClaw internally, build products on top of it, or offer it as a service. No license fees, no revenue sharing, no attribution requirements beyond keeping the MIT license notice in the source.

My Discord bot is online but not responding — what is wrong?

This is the single most common support question on the OpenClaw Discord. The fix:

Comparisons

OpenClaw vs ChatGPT?

They solve different problems:

OpenClaw is for people who want an agent that acts on their behalf across multiple platforms. ChatGPT is for people who want a conversation partner with minimal setup. They complement each other more than they compete.