ZeptoClaw

ZeptoClaw is a fast, small, secure, local-first personal AI assistant infrastructure written in Rust. It ships as a single ~6MB binary that starts in ~50ms and uses ~6MB of RAM, bundling tools, long-term memory, messaging channels, 18 LLM providers, and sandboxed autonomy into one deployable unit. It explicitly positions itself as a synthesis of the Claw ecosystem: OpenClaw's integration breadth without the TypeScript footprint, NanoClaw's container isolation simplicity, NemoClaw's policy-gated guardrails without Docker/k3s overhead, and PicoClaw's edge efficiency with Rust's safety.

Architecturally, ZeptoClaw emphasizes defense-in-depth for agentic workloads: container isolation, prompt injection detection, a circuit-breaker provider stack, taint labelling, and safety scanning of model output. Recent commits show a strong focus on real-world edge deployment robustness — parsing reasoning_content from reasoning models (Qwen3, DeepSeek distills) served via Ollama/LiteLLM, sanitizing third-party MCP tool schemas so strict providers don't reject entire tool arrays, and coercing string-typed tool arguments from weak local models into their declared types. The project maintains rigorous engineering discipline with 3,900+ tests, clippy-clean code, and prompt security patching (e.g., RUSTSEC-2026-0285).

Compared to OpenClaw, ZeptoClaw trades ecosystem breadth for footprint, speed, and deploy-anywhere simplicity. One notable governance concern: the project recently removed its GitHub Actions CI checks in favor of local validation, which shifts trust onto maintainer discipline. It is best suited for individuals self-hosting a personal assistant on edge hardware or a VPS who value security controls and minimal resource usage over a large plugin marketplace.