Skip to main content
The core AI engine inside every EasyClaw container.

​
What is ZeroClaw?

ZeroClaw is a fast, fully autonomous AI agent framework built in Rust. It runs inside each user’s container and handles everything between receiving a message and sending a response. ZeroClaw is not a wrapper around an AI API. It’s a complete agent runtime that manages:
  • Conversation context — Message history, token budgets, context windowing
  • Tool execution — Shell commands, file operations, browser automation, integrations
  • Memory — Long-term memory with daily summaries persisted to disk
  • Channel bridging — Routing messages between WhatsApp, Signal, Telegram, iMessage, Email
  • Model connections — API calls to Anthropic, OpenAI, Google, and OpenRouter
ZeroClaw supports 30+ AI providers and 14 messaging channels. EasyClaw exposes a curated subset of these for the best user experience.

​
Configuration

ZeroClaw is configured through a TOML file at ~/.zeroclaw/config.toml: 
[autonomy]
level = "full"
max_actions_per_hour = 100
max_cost_per_day_cents = 5000

[agent]
max_tool_iterations = 20
max_history_messages = 50

[gateway]
require_pairing = false

​
Autonomy levels

LevelDescription
FullUnrestricted. The AI can execute any tool, run any command, and access any file. This is the default.
SupervisedThe AI must request approval before executing potentially dangerous actions.

​
Agent settings

SettingDefaultDescription
max_tool_iterations20Maximum tool calls per turn. Prevents infinite loops.
max_history_messages50Messages kept in the active context window. Older messages are summarized into memory.
max_actions_per_hour100Rate limit on tool executions.
max_cost_per_day_cents5000Daily spending cap on AI API calls ($50).

​
Tool system

ZeroClaw gives the AI access to a set of tools. When you ask your assistant something, the AI decides which tools to use based on your request.

​
Built-in tools

ToolDescription
ShellExecute commands in the container
File readRead file contents
File writeWrite or edit files
File searchSearch for files by name or content
BrowserBrowse the web with a headless browser
Memory readRecall information from long-term memory
Memory writeSave information to long-term memory

​
Integrations

ZeroClaw supports the Model Context Protocol (MCP) for connecting external tools. EasyClaw provides 250+ integrations — Google Workspace, GitHub, Slack, Linear, Notion, and more — each running as a separate MCP server inside your container.

​
Memory

ZeroClaw maintains two types of memory:

​
Active context

The recent conversation history (up to max_history_messages). This is what the AI “sees” when generating a response.

​
Long-term memory

Daily summaries written to ~/.zeroclaw/memory/. When the active context fills up, older messages are summarized and stored. The AI can recall these summaries using the memory tool. This gives your assistant persistent memory across sessions. It remembers your preferences, ongoing projects, and past conversations — even after your container is frozen and restored.

​
Gateway

ZeroClaw includes a gateway server that accepts WebSocket connections. This is how the web console and desktop app communicate with your AI in real-time: 
Browser → Public API → WebSocket proxy → Container gateway → ZeroClaw
The gateway handles real-time message streaming, file uploads and downloads, session pairing, and health checks.

​
Open source

ZeroClaw is developed as an independent project. The core agent framework is designed to be provider-agnostic and channel-agnostic — EasyClaw is one of many possible deployments.

Learn about OpenClaw

The open-source ecosystem behind EasyClaw.