ZeroClawAn AI proxy running time frame, located in OpenClaw as a very lightweight alternative, prepared in Rust language. The project was opened in February 2026 and in two days more than 3,400 stars were acquired in GitHub. Its core advantage is its impressive performance: only 7.8 MB, nearly 200 times lower than OpenClaw, run-time memory; cold start time is less than 10 milliseconds, 400 times faster than OpenClaw. The binary file is only 3.4MB and can be deployed on low-capacity equipment such as berry pie. In terms of architecture, ZeroClaw has adopted a tripit-driven plug-in design, defining eight core Trait interfaces, including Provider, Channel, Tool, Memory, etc., each component can be replaced by configuration to lock down the zero manufacturer. The security mechanism is also excellent, with internal matching codes, work area limits, white lists of commands and API key encryption. The project supports the AIEOS personality code, which allows the AI character configuration to migrate between systems. The built-in hybrid search engine combines vector similarity with keyword matching, and can process memory retrieval without external dependence。
ZeroClaw function
- EXTREMELY LIGHT: THE SINGLE BINARY FILE IS ONLY ABOUT 3.4MB, THE MEMORY IS LESS THAN 5MB RUNNING AND THE START TIME IS LESS THAN 10 MS。
- Multi-model compatibility: Native supports more than 22 AI service providers, including OpenAI, Anthropic, Mistral, OpenRouter, Ollama etc., compatible with OpenAI API formats。
- Multi-channel access: Supports multiple communication channels such as CLI, Telegram, Discord, Slack, Webhook。
- Modularized architecture: Models, corridors, memories, tools are plug-inable modules that are easily extended。
- Local priority: Default SQLite storage to support full-text search and vector search without external dependence。
- Security design: Internal sandboxes, white lists, work area isolation and twinning certification mechanisms to ensure safe operation。
Official website link:http://zeroclaw.org/zh