nullclaw  by nullclaw

Ultra-lightweight, autonomous AI assistant infrastructure

Created 1 week ago

New!

1,962 stars

Top 22.0% on SourcePulse

GitHubView on GitHub
1 Expert Loves This Project
Project Summary

NullClaw is an AI assistant infrastructure designed for extreme efficiency and portability. Written entirely in Zig, it offers a 678 KB static binary with ~1 MB RAM usage and sub-8 ms startup, making it ideal for resource-constrained environments and edge devices. Its pluggable architecture and focus on security provide a flexible yet robust foundation for autonomous AI agents.

How It Works

The project leverages Zig's capabilities to compile into a single, static binary with no runtime or VM overhead, achieving near-zero memory usage and instant startup times on modest hardware. Its core design employs vtable interfaces for all subsystems—including AI model providers, communication channels, tools, and memory—enabling complete pluggability and swapability without code modifications. This architecture prioritizes lean defaults, security through sandboxing and explicit allowlists, and extreme portability across architectures.

Quick Start & Requirements

To build, clone the repository and run zig build -Doptimize=ReleaseSmall. The primary executable is then ./zig-out/bin/nullclaw. Key requirements include a Zig compiler (version 0.15 mentioned) and libc; SQLite is used for memory but is not a strict external dependency. Quick setup involves nullclaw onboard --api-key <key> --provider <provider> or an interactive wizard.

Highlighted Details

  • Impossibly Small: A 678 KB static binary with ~1 MB peak RSS, requiring only libc.
  • Instant Startup: Boots in under 2 ms on Apple Silicon and under 8 ms on a 0.8 GHz edge core.
  • Feature-Rich: Supports 22+ AI providers, 11 communication channels, 18+ tools, hybrid vector/FTS5 memory, multi-layer sandboxing, and hardware peripheral access.
  • Robust Security: Implements pairing, strict sandboxing (Landlock, Firejail, Docker), encrypted secrets, workspace scoping, and resource limits.
  • Hybrid Memory System: Combines SQLite with FTS5 for keyword search and vector embeddings, offering configurable weighting and automatic hygiene.

Maintenance & Community

The project uses CalVer for versioning (YYYY.M.D) with potential for daily releases. No specific details on maintainers, sponsorships, or community channels (like Discord/Slack) are provided in the README.

Licensing & Compatibility

NullClaw is distributed under the MIT license, which permits commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

As a pre-1.0 project, NullClaw does not offer stability guarantees; its configuration and command-line interface are subject to change between releases.

Health Check
Last Commit

1 day ago

Responsiveness

Inactive

Pull Requests (30d)
59
Issues (30d)
50
Star History
2,049 stars in the last 9 days

Explore Similar Projects

Feedback? Help us improve.