ZeroClaw

Overview

ZeroClaw is a security-first Rust AI agent runtime built by students from Harvard, MIT, and the Sundai.Club community, now organized as ZeroClaw Labs.^[1] The project prioritizes security, efficiency, and portability — a single binary that runs on $10 hardware while talking to ~20 LLM providers and 30+ channels.^[2]

As of June 2026, the repo sits at roughly 31.9K GitHub stars with 4.7K forks — about double its late-February count.^[1] The OpenClaw security crisis (CVE-2026-25253 plus the ClawHavoc malicious-skills campaign) pushed security-conscious users toward hardened alternatives, and ZeroClaw's deny-by-default posture made it a fixture of post-crisis "OpenClaw alternatives" coverage.^[3]

Funding is not publicly disclosed. The project is free and open source under MIT OR Apache-2.0.

Key Stats (as of June 11, 2026)

Benchmark vs Alternatives

Launch-era comparison published by the project:

Independent testing broadly confirms the efficiency claims — LumaDock measured ZeroClaw idling at ~3.9MB RAM with cold starts under 10ms, versus "hundreds of megabytes" for OpenClaw.^[4]

Installation

# One-line installer (prebuilt binary or source build)
curl -fsSL https://raw.githubusercontent.com/zeroclaw-labs/zeroclaw/master/install.sh | bash

# Or clone and run
git clone https://github.com/zeroclaw-labs/zeroclaw.git
cd zeroclaw
./install.sh

# Minimal kernel only (~6.6MB)
./install.sh --minimal

# One-shot setup: pick a provider, write a working config
zeroclaw quickstart

# Interactive chat
zeroclaw agent -a <alias>

# Run always-on as a system service
zeroclaw service install
zeroclaw service start

Pre-built binaries cover Linux (x86_64, aarch64, armv7, musl), macOS (x86_64, aarch64), Windows (x86_64), and Android (aarch64), with platform guides for FreeBSD, NixOS, and Docker.^[1]

Security Features

A "YOLO mode" escape hatch exists for trusted dev environments.^[2]

Channel Support

ZeroClaw now ships 30+ channel adapters — CLI, Telegram, Discord, Slack, iMessage, Matrix, WhatsApp, email, voice, and webhooks all deliver into the same agent loop.^[1] IDE/editor integration arrives via the Agent Client Protocol (JSON-RPC 2.0 over stdio).

Provider Support

• Ollama — Local models
• OpenRouter — Access to all major models
• Anthropic — Claude direct
• OpenAI — Including Codex subscription auth
• ~20 cloud providers — With fallback chains and routing when a provider flakes

Architecture

User → Channel (30+) / Gateway (REST/WS) / ACP → ZeroClaw Runtime → Sandbox → LLM Provider
                              ↓
                  Memory (SQLite + FTS5 + Vector)

Trait-driven design — Core systems (providers, channels, tools, memory, tunnels, peripherals) are Rust traits, making everything swappable.

Hybrid memory — SQLite with FTS5 full-text search plus vector cosine similarity for semantic retrieval.

SOP engine — Event-triggered Standard Operating Procedures (MQTT, webhook, cron, peripheral) with approval gates and resumable runs.

Strengths

• Security-First — Supervised autonomy, OS sandboxes, tool receipts, workspace scoping
• Extreme Efficiency — ~4MB idle RAM, sub-10ms startup, verified by independent testing^[4]
• Active Development — Four releases between May and early June 2026; commits landing daily as of June 11^[1]
• Portable — Single binary across ARM, x86, RISC-V, plus GPIO/I2C/SPI on embedded boards
• No Lock-In — OpenAI-compatible plus pluggable custom endpoints
• Migration Path — Imports OpenClaw data
• Active Community — Discord, Telegram, Reddit, and multilingual docs

Weaknesses / Considerations

• Young Project — Launched February 13, 2026; still pre-1.0 (v0.8.0 in beta)
• Security Gaps in Practice — LumaDock found modules like PromptGuard and LeakDetector exist but "aren't wired into the production response pipeline yet"^[4]
• Rust Build Time — Source compilation requires 2-4GB RAM (prebuilt binaries avoid this)
• Smaller Ecosystem — Far fewer integrations than OpenClaw's 1,000+ SaaS connections; thinner docs in places^[4]
• Impersonation Issues — The project has warned about fake domains and unauthorized forks

What Developers Say

From the Show HN for James Library, a multi-agent research lab built on ZeroClaw:^[5]

"ZeroClaw's pure Rust implementation provided exactly the zero-overhead foundation the project required... Because they solved the core runtime execution so elegantly, I was able to spend my time building the higher-level orchestration on top of it." — cwoodyard, Hacker News

Not everyone is impressed by the differentiation. In a thread comparing the Claw-ecosystem agents, one commenter dismissed it as simply:^[6]

"ZeroClaw: Claw But In Rust" — tao_oat, Hacker News

The broader sentiment in that thread was cautious — commenters noted the Claw forks are "popping up way too fast" for quality software and recommended strict access controls (virtual cards, restricted accounts) regardless of which runtime you pick.^[6]

Who Built It

Students and members from Harvard, MIT, and the Sundai.Club community, now operating as ZeroClaw Labs with an official site and hosted docs.^[7] The project maintains communities in English, Chinese, Russian, Japanese, French, and Vietnamese.

Ideal User

• Security-conscious deployments requiring isolation, approval gates, and audit receipts
• Edge/embedded systems with under 10MB RAM available
• Hardware projects — Raspberry Pi, STM32, Arduino, ESP32 via the Peripheral trait
• Rust developers wanting to contribute or customize
• Privacy-focused individuals who need data to stay local

Bottom Line

ZeroClaw is what you run when OpenClaw's security model keeps you up at night — and after the January-February 2026 CVE and ClawHavoc episodes, a lot more people are losing sleep. Stars roughly doubled to ~32K between late February and June 2026, and the project graduated from student hack to organized lab with hosted docs, daily commits, and a steady release cadence heading toward v0.8.0.

The efficiency numbers hold up under independent testing — about 4MB idle RAM and sub-10ms startup means this genuinely runs on $10 hardware. The security architecture (supervised autonomy, OS-level sandboxes, tool receipts) is the most considered in the Claw ecosystem, though reviewers note some advertised modules aren't fully wired into the production pipeline yet.

The caveats are maturity and ecosystem depth. It's still pre-1.0, documentation is thin in places, and OpenClaw's integration catalog remains far larger. For security-critical deployments or edge hardware, ZeroClaw is the strongest option in the ecosystem. For maximum integrations, OpenClaw still wins on breadth.

---