OpenClaw AI — Frequently Asked Questions

OpenClaw itself is completely free and open-source (MIT License). You only pay for the AI provider API you connect it to (OpenAI, Anthropic, etc.). If you use Ollama with local models, there are zero costs — everything runs on your own hardware.

OpenClaw is written in Python and has a Python SDK and a CLI. It can work with any programming language through file operations, shell commands, and code execution. The Python SDK supports direct embedding in Python applications.

Python 3.10 or later is required. Python 3.12 is recommended for best performance. Check your version with python3 --version.

Yes. OpenClaw supports Windows 10/11, macOS 12+, and all major Linux distributions. See the installation guide for platform-specific instructions.

Only if you use cloud providers (OpenAI/Anthropic). File content is sent to the LLM API so the agent can read and understand them. Use Ollama with local models for full data privacy — nothing leaves your machine. You can also configure max_file_size limits to prevent large files from being sent.

By default, OpenClaw can read and write files, but will not delete files without explicit confirmation (in interactive mode) or unless the task specifically asks for deletion. Use --dry-run to preview all file operations before they execute. You can restrict operations via allow_file_write: false in config.

Yes, but follow best practices: (1) Never include API keys or secrets in plain text files. (2) Use .openclawignore to exclude sensitive directories. (3) Run with --dry-run for unfamiliar tasks. (4) Use local Ollama models if your codebase is proprietary.

Yes, with the browser extra: pip install openclaw[browser]. It can fetch web pages, search the web, and extract structured data. Install playwright for full JavaScript rendering support.

Yes. By default, command execution is interactive — the agent explains what it will run and asks before executing. You can enable autonomous execution with allow_command_exec: true in config, but only do this in trusted environments.

It depends on the model's context window. GPT-4o has 128K tokens (~100K words). For large codebases, OpenClaw uses chunking and summarization to handle more content than fits in the context window. Configure max_context_files: 50 to limit how many files are included.

Each openclaw run call starts a fresh context. For persistent memory across sessions, use named agents with the --agent flag — they maintain a conversation history file at ~/.openclaw/agents/name/history.json. Use openclaw agent list to manage saved agents.

OpenClaw is developer-focused and local-first. Key differences: (1) Works entirely offline with Ollama. (2) Native CLI — no web UI required. (3) Direct filesystem access without sandboxes. (4) Python SDK for embedding in applications. (5) Built for automation pipelines, not just chat-style interaction.

LangChain is a framework for building AI applications — it requires writing Python code. OpenClaw is a ready-to-use tool that can be driven from the command line with natural language. They're complementary: use LangChain for custom applications, use OpenClaw for quick automation tasks.