NullClaw

NullClaw runs AI agents on your own hardware or server for coding work and automated background tasks. It's aimed at developers who want an assistant that can carry out multi-step jobs with controls over its access and actions.

• Coding and task execution: Agents can run shell commands, read and edit files, search the web, and work with Git, subject to approval policies and sandbox restrictions.
• Model connections: Use Ollama, OpenRouter, Anthropic, or OpenAI-compatible APIs, including custom local and remote endpoints.
• Persistent memory: Saved sessions survive restarts, automatic context compaction supports longer conversations, and the default SQLite memory backend combines keyword and vector search.
• Interactive and background work: Talk to an agent through the terminal or a browser client, or run it as a headless worker receiving tasks from an orchestrator such as NullBoiler.
• Extensions and messaging: Connect MCP tools, delegate work to subagents, schedule tasks, and receive requests through channels including Telegram, Signal, Discord, and Slack.

The runtime is a self-contained Zig binary with no external runtime dependencies, supporting ARM, x86, and RISC-V hardware. It supports native execution, Docker sandboxing, and WASM through wasmtime. The agent runs locally, but choosing a cloud model provider sends model requests to that service. Browser dashboards with tool timelines and visual approvals use the separate NullHub ecosystem. NullClaw is open source under the MIT license.