Index / NullClaw · updated Sep 3, 2026

NullClaw

nullclaw/nullclaw · healthy · rank 13 of 43 by stars

A 678 KB static Zig binary that delivers a full autonomous AI assistant stack with ~1 MB RAM and sub-2ms boot. It's the extreme-minimalist, runs-on-anything answer to heavyweight agent frameworks.

#zig#lightweight#edge#autonomous-agent#static-binary
Compare vs OpenClaw
GitHub ↗

Facts

Repository

Contributors103
Open issues93
Last commitJul 13, 2026
Release cadence~8 days
Latest releasev2026.5.29

Runtime

LanguageZig
Memory1 MB
Boot time2 ms
Deploymentedge · desktop · self-hosted · cloud
Setup difficultyMedium
Plugin ecosystemEmerging

Posture

LicenseMIT
Local-firstYes
Cloud dependencyOptional
Privacy postureStrong

Community

Sentiment72% positive
Reddit mentions1
Web results10

Security breakdown

Composite 88 / 100 · how these are scored

Sandboxing9 / 10

higher is safer

API security8 / 10

higher is safer

Network isolation8 / 10

higher is safer

Telemetry safety9 / 10

higher is safer

Shell access risk4 / 10

higher is riskier

Model access

Read from the repository, not written by a model · 51 files examined

5 providers · runs locally · gateway support

Direct
AnthropicAzure OpenAIGoogleOpenAIVertex AI
Gateway
OpenRouter
Local
LM StudioOllamavLLM
Compatible
OpenAI-compatible

Pinned models

glm-4.7Zhipu
gpt-5.4OpenAI
gpt-4oOpenAI

glm-4.7 released 2025-12-22 — 9 months old (from the public model catalogue)

Pin last edited 2 months ago

Several defaults are pinned; the project picks one per provider rather than shipping a single default.

Evidence

Decision

Why choose NullClaw over OpenClaw?

Why choose this

  • Orders of magnitude smaller footprint: 678 KB binary, ~1 MB RAM
  • Sub-2ms cold start versus typical runtime-heavy stacks
  • True portability across ARM, x86, and RISC-V with no runtime deps

Tradeoffs

  • Smaller community and ecosystem than OpenClaw
  • Zig toolchain narrows the contributor and hiring pool
  • UI layer (nullhub) is still in beta

Best fit

  • Running agents on $5 SBCs and constrained edge hardware
  • Users who want a single dependency-free static binary
  • Security-conscious self-hosters wanting layered sandboxing

Avoid if

  • You need a mature GUI or managed cloud offering
  • You want a large plugin marketplace and community templates
  • You prefer mainstream languages with bigger contributor pools
High Confidence82%

Strong evidence from a detailed README, active recent commits, and a current release supports the performance and security claims. Community sentiment evidence is thin (1 Reddit match), so sentiment and ecosystem maturity scores carry more uncertainty.

AI layer reviewed Aug 24, 2026 · how this is written

Star activity

8,061 stars today

Overview

NullClaw is a fully autonomous AI assistant infrastructure written entirely in Zig, shipping as a single 678 KB static binary with ~1 MB peak RSS and sub-2ms startup on modern hardware. Despite its size, it is feature-complete: 50+ providers, 19 channels, 35+ tools, 10 memory engines, MCP support, subagents, streaming, voice, tunnels, and hardware peripherals — all behind pluggable vtable interfaces so every core system is swappable.

Security is a first-class design goal: pairing-based access, multi-layer sandboxing via landlock, firejail, bubblewrap, or Docker, explicit allowlists, workspace scoping, and encrypted secrets. Recent commits show active hardening work — stderr isolation for agent runners, allocation-free gateway bind paths, and config validation to prevent silent message drops — backed by 5,300+ tests and CI/nightly pipelines.

Compared to OpenClaw, NullClaw trades ecosystem breadth for extreme minimalism and portability: it runs on anything with a CPU, from Apple Silicon to 0.8 GHz edge cores and RISC-V boards, with zero runtime dependencies. The companion nullhub UI (beta) plus nullboiler, nullwatch, and nulltickets round out an emerging ecosystem for orchestration, observability, and task tracking.

Nominate a clone

Add a new Claw

Paste a GitHub repository and tell us why it belongs on the tracker.

Opens a prefilled issue on GitHub — every nomination is public. Comfortable with a PR? Adding the repo to projects.json is faster.