What is ZeroClaw?
ZeroClaw is a self-hosted AI assistant that runs entirely on your own machine. It works as an agent runtime: a single binary you configure to talk to an LLM, connect to your messaging apps, and take actions on your behalf.
Your keys, your conversations, and any files the tool touches never leave your hardware. ZeroClaw is a personal AI assistant that puts you in control of which language model to use, which messaging platforms to connect, and exactly what the agent is allowed to do.
Features & Benefits
- Multi-Channel Messaging: connect the AI assistant to Discord, Telegram, Signal, WhatsApp, Matrix, Slack, email, and over a dozen other platforms from a single running instance.
- LLM Provider Support: point the agent at Anthropic, OpenAI, Ollama, or roughly 20 other model providers, with each agent using its own provider independently.
- Shell and File Execution: let the AI assistant run shell commands, read and write files, search file contents by regex, and edit files by exact string match within your workspace.
- Headless Browser Automation: give the AI assistant control of a browser to fetch pages, interact with web interfaces, and extract content.
- Web Search: query the web through DuckDuckGo, Brave, Tavily, SearXNG, or Jina directly from within a conversation.
- Long-Term Memory: store facts, preferences, and notes that the AI assistant can recall across future sessions.
- Scheduled Tasks and Cron Jobs: create, update, and run scheduled jobs the agent manages on its own.
- Standard Operating Procedures (SOPs): define deterministic multi-step workflows with trigger matching, approval gates, and a full audit log.
- Subagent Delegation: spawn child agents to handle subtasks in parallel or in sequence.
- Voice Support: connect voice calls through Twilio, Telnyx, or Plivo, with text-to-speech output and local wake-word detection.
- Hardware Access: read and write GPIO, I2C, and SPI interfaces for use with boards like Raspberry Pi.
- MCP Extension Support: connect any Model Context Protocol server to extend what the AI assistant can do.
- Tool Risk Controls: classify every tool call by risk tier and require operator approval before the agent takes higher-risk actions.
What can ZeroClaw do?
- Run a private AI assistant on your own hardware
- Connect an AI assistant to Telegram, Discord, and Signal
- Automate tasks with scheduled cron jobs
- Execute shell commands from a chat conversation
- Search the web and fetch page content
- Read, write, and edit local files
- Automate browser interactions
- Store and recall long-term memory across sessions
- Route conversations to multiple messaging platforms
- Spawn subagents to handle parallel tasks
- Define repeatable multi-step workflows with approval gates
- Control hardware GPIO and I2C interfaces
- Switch LLM providers per agent
Real-World Applications
Developers who want a private AI assistant for daily coding work can set up ZeroClaw on their local machine and interact with it through a terminal UI or their existing chat apps. The AI agent can run shell commands, read and edit source files, search codebases by regex, and automate browser interactions. All of it stays on the developer’s hardware, with no API traffic going through a third-party cloud platform.
Privacy-conscious individuals or businesses that handle sensitive data may find ZeroClaw useful when they need AI help without sending that data to an external server. A small law firm or medical practice could use the AI assistant through email or Telegram while keeping client information fully on-premise. The tool risk controls let operators decide exactly what the agent is allowed to do before it acts.
Teams that run their own infrastructure can use ZeroClaw to build internal automation. The SOP system lets you define repeatable workflows with explicit triggers and approval steps. An engineering team could, for example, wire up a workflow that monitors a queue, drafts a pull request, and waits for human sign-off before proceeding.
Hobbyists and makers working with hardware can connect ZeroClaw to Raspberry Pi or similar boards via GPIO, I2C, and SPI interfaces. That means the personal AI assistant can read sensor data or trigger physical outputs as part of a broader automated routine.
