Index / NullClaw · updated Oct 24, 2026
NullClaw
nullclaw/nullclaw · healthy · rank 15 of 50 by stars
A 678 KB static Zig binary that delivers a fully autonomous AI assistant stack with ~1 MB RAM and sub-2ms boot. It trades ecosystem maturity for extreme minimalism, portability, and security-first design.
Facts
Repository
Runtime
Posture
Community
Security breakdown
Composite 88 / 100 · how these are scored
higher is safer
higher is safer
higher is safer
higher is safer
higher is riskier
Model access
Read from the repository, not written by a model · 51 files examined
5 providers · runs locally · gateway support
Pinned models
glm-4.7 released 2025-12-22 — 10 months old (from the public model catalogue)
Pin last edited 19 days 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
- Dramatically smaller footprint: 678 KB binary and ~1 MB RAM vs typical runtimes
- Boots in <2 ms, enabling cold-start and microcontroller-class deployments
- Single static binary runs on ARM, x86, and RISC-V with only libc
Tradeoffs
- Much smaller community and skill/plugin ecosystem
- Lower development velocity; described in digests as quiet maintenance
- UI layer (nullhub) is beta, while OpenClaw ships a full Control UI
Best fit
- Running agents on $5 ARM boards and constrained edge hardware
- Users who want a single static binary with zero runtime dependencies
- Security-conscious self-hosters wanting layered sandboxing
Avoid if
- You need a large plugin/skill ecosystem like OpenClaw's registry
- You want a mature UI out of the box (nullhub is still beta)
- You prefer mainstream-language codebases for easy contribution
Strong evidence from README and commit history for architecture and performance claims; community sentiment is inferred from a small set of ecosystem digests with zero Reddit data, so popularity and long-term momentum are uncertain.
AI layer reviewed Oct 5, 2026 · how this is written
Star activity
8,132 stars today
Overview
NullClaw is a ground-up reimagining of autonomous AI assistant infrastructure written entirely in Zig. It ships as a 678 KB static binary with no runtime, VM, or framework overhead, peaking at roughly 1 MB RSS and booting in under 2 ms on Apple Silicon (under 8 ms on a 0.8 GHz edge core). Despite the tiny footprint it is feature-complete: 50+ LLM providers, 19 chat channels, 35+ tools, 10 memory engines, MCP support, subagents, streaming, voice, tunnels, and hardware peripherals.
Architecturally, everything is a vtable interface — providers, channels, tools, memory, tunnels, and peripherals are all pluggable and swappable. Security is a first-class concern: pairing-based auth, multi-layer sandboxing via landlock, firejail, bubblewrap, or Docker, explicit allowlists, workspace scoping, and encrypted secrets. Recent commits show active hardening work, including memory-safety fixes (use-after-free in cron delivery), secure HTTP fallbacks on Android, and careful diagnostics logging with redacted, truncated payloads.
Compared to OpenClaw, NullClaw targets the opposite end of the spectrum: instead of a rich skill registry and Control UI, it bets on minimalism, portability, and auditability. The surrounding Null ecosystem (nullhub UI, nullboiler orchestration, nullwatch observability, nulltickets task tracking) is emerging but still beta. Ecosystem digests characterize it as a low-velocity, quiet-maintenance project, so it suits users who prioritize footprint and control over community momentum.