ZeroClaw 🦀

Claw Done Right. The Ultra-Lightweight AI Agent Runtime.

View on GitHub

What is ZeroClaw?

ZeroClaw is a highly efficient, Rust-based alternative to the OpenClaw agent runtime. Designed to address the resource intensity and performance bottlenecks of existing solutions, ZeroClaw offers a radically optimized environment for running autonomous AI agents.

ZeroClaw

Image source: https://github.com/openagen/zeroclaw

In contrast to traditional runtimes which consume significant memory and CPU even when idle, ZeroClaw operates with a minimal footprint, making it accessible to a wider range of hardware, from powerful servers to low-cost edge devices.

The project focuses on portability, security, and speed. By compiling to a single static binary, it eliminates complex dependency chains and "it works on my machine" issues. ZeroClaw enables developers to deploy agents that boot instantly and execute tasks with near-native performance.

Key Features

🏎️ Ultra-Lightweight

ZeroClaw is engineered for efficiency. With a memory footprint of less than 5MB, it is 99% smaller than the OpenClaw core. This allows you to run multiple agents concurrently without exhausting system resources.

💰 Minimal Cost

Run your agents on hardware costing as little as $10. ZeroClaw is 98% cheaper to operate compared to requiring a Mac mini or similar hardware for other runtimes.

⚡ Lightning Fast

Experience startup times that are 400x faster. ZeroClaw boots in under 10 milliseconds, even on modest 0.6GHz cores. No more waiting for heavy runtimes to initialize.

🌍 True Portability

A single self-contained binary works across ARM, x86, and RISC-V architectures. ZeroClaw removes the headache of managing environment-specific dependencies.

Benchmark Snapshot

MetricOpenClawNanoBotPicoClawZeroClaw 🦀
LanguageTypeScriptPythonGoRust
RAM> 1GB> 100MB< 10MB< 5MB
Startup (0.8GHz core)> 500s> 30s< 1s< 10ms
Binary Size~28MB (dist)N/A (Scripts)~8MB3.4 MB
CostMac Mini $599Linux SBC ~$50Linux Board $10Any hardware $10

* Local machine quick benchmark (macOS arm64, Feb 2026) normalized for 0.8GHz edge hardware.
Notes: ZeroClaw results measured with /usr/bin/time -l on release builds. OpenClaw requires Node.js runtime (~390MB overhead). PicoClaw and ZeroClaw are static binaries.

Reproduce ZeroClaw numbers locally:

cargo build --release
ls -lh target/release/zeroclaw

/usr/bin/time -l target/release/zeroclaw --help
/usr/bin/time -l target/release/zeroclaw status

Architecture

ZeroClaw adopts a modular, trait-based architecture. This means every subsystem—including the memory provider, the communication channel, or the tool execution environment—is defined by a simple interface (trait). This design allows for:

  • Swap-ability: Change implementations with a simple configuration update. No code rewriting required.
  • Security: Components are isolated. Core systems are sandboxed by design.
  • Evolution: New capabilities can be added by implementing the relevant traits without disturbing the core logic.
Provider  <--->  [ Runtime Adapter ]  <--->  Channel ^ Memory    <--->  [ Security Policy ]  <--->  Tools v Observer  <--->  [ Identity Config ]  <--->  Tunnel

Detailed Capabilities

Identity System & AIEOS

ZeroClaw supports the AI Entity Object Specification (AIEOS), allowing for portable and standardized AI personas. You can define your agent's identity using simple JSON configurations or traditional markdown files. This ensures your agent maintains its behavioral integrity across different models and environments.

Features include:

  • Import/Export: Move identities between AIEOS-compatible systems.
  • Rich Profiling: Define psychology, traits, linguistics, and motivations.
  • Backwards Compatibility: Fully supports OpenClaw's markdown-based identity files (IDENTITY.md, SOUL.md).

Full-Stack Memory Engine

ZeroClaw includes a custom-built memory system that requires zero external dependencies. There is no need for complex setups like Pinecone or Elasticsearch. The memory engine handles vector embeddings and keyword search locally and efficiently.

  • Auto-Recall: context is automatically retrieved based on the task.
  • Hybrid Search: Combines vector similarity (0.7 weight) with keyword matching (0.3 weight).
  • Storage Options: Choose between SQLite, Markdown, or ephemeral memory.

Security First

Designed for safety, ZeroClaw implements strict controls to prevent agents from acting unexpectedly.

  • Pairing Requirement: New connections require a secure pairing code.
  • Workspace Scoping: File access is restricted to the specific workspace directory by default.
  • Command Allowlist: Only explicitly allowed commands (like git, npm, cargo) can be executed.
  • Encrypted Secrets: API keys are encrypted at rest using a local key file.

Quick Start Guide

Get up and running with ZeroClaw in seconds.

# 1. Install
git clone https://github.com/openagen/zeroclaw.git
cd zeroclaw
cargo build --release
cargo install --path . --force

# 2. Setup (Choose one)
# Quick setup (no prompts)
zeroclaw onboard --api-key sk-... --provider openrouter

# Or interactive wizard
zeroclaw onboard --interactive

# Or quickly repair channels/allowlists only
zeroclaw onboard --channels-only

# 3. Running Agents
# Chat
zeroclaw agent -m "Hello, ZeroClaw!"

# Interactive mode
zeroclaw agent

# Start the gateway (webhook server)
zeroclaw gateway                # default: 127.0.0.1:8080
zeroclaw gateway --port 0       # random port (security hardened)

# Start full autonomous runtime
zeroclaw daemon

# 4. Management & Diagnostics
# Check status
zeroclaw status

# Run system diagnostics
zeroclaw doctor

# Check channel health
zeroclaw channel doctor

# Get integration setup details
zeroclaw integrations info Telegram

# Manage background service
zeroclaw service install
zeroclaw service status

# 5. Migration
# Migrate memory from OpenClaw (safe preview first)
zeroclaw migrate openclaw --dry-run
zeroclaw migrate openclaw

# Dev fallback (no global install): prefix commands with cargo run --release -- 
# Example: cargo run --release -- status

Frequently Asked Questions