ZeroClaw

✨Key Features

• ▸Ultra-Lightweight & Fast: Compiled binary ~3-9MB, cold starts in <10-20ms, runs in <5MB RAM; ideal for edge/embedded devices or resource-constrained environments.
• ▸Modular & Swappable Design: Everything (LLM providers, communication channels, tools, memory backends, runtimes) is trait-based—swap OpenAI → Ollama → Anthropic or Telegram → Discord without code changes.
• ▸Supported LLM Providers: 20+ including OpenAI-compatible, Anthropic, Ollama (local/remote), custom endpoints; native tool calling for many.
• ▸Multi-Channel Integration: Telegram, Discord, Slack, WhatsApp, Matrix, Signal, Email, Webhook, and more out-of-the-box.
• ▸Memory & Autonomy: Hybrid vector+keyword search (SQLite-based), multiple backends; supports supervised, readonly, or full autonomous modes with cron/scheduled tasks.
• ▸Built-in Tools & Extensibility: Shell, file ops, git, browser (WebDriver), HTTP, screenshot, hardware access; plugin system via TOML manifests.
• ▸Security-First: Default localhost bind, one-time pairing, filesystem sandboxing, deny-by-default channels, encrypted secrets, no public exposure without tunnels (Cloudflare/Tailscale/ngrok).
• ▸Runtime Options: Native execution or Docker-sandboxed; future WASM/edge support planned.
• ▸Identity & Persona: Markdown-based (IDENTITY.md/SOUL.md) or AIEOS JSON schemas for consistent agent behavior.
• ▸Observability: Prometheus/OpenTelemetry integration.
• ▸Companion Tools: Python package for LangGraph-style tool calling on inconsistent providers.
• ▸Community & Status: ~11.5k stars, active development (commits minutes ago as of Feb 2026), MIT licensed, built by contributors from Harvard/MIT/Sundai.Club; includes bootstrap scripts, detailed docs, and multilingual READMEs.