clacky-ai/openclacky

What it solves

OpenClacky is designed to reduce the high token costs associated with running AI agents, while maintaining capabilities comparable to high-end closed-source agents like Claude Code. It provides an open-source, model-agnostic alternative for coding and general task automation.

How it works

OpenClacky employs several strategies to minimize token consumption:

• Cache Optimization: It uses "Insert-then-Compress" and double cache markers to maintain a near 100% cache hit rate, ensuring system prompts are not mutated.
• Minimal Toolset: Instead of bloating the prompt with dozens of tools, it uses only 16 core tools and offloads extended capabilities to a "Skill" ecosystem via a single meta-tool.
• Background Compression: The agent automatically compresses long contexts during idle time to reduce cold-start costs.
• BYOK (Bring Your Own Key): Users can choose any OpenAI-compatible model (e.g., Claude, DeepSeek, GPT) to further optimize costs.
• Skill Evolution: Users can create skills in natural language; the agent then drafts the documentation and evolves the skill's accuracy based on execution results.

Who it’s for

• Developers looking for a high-performance AI coding assistant that is open-source and model-flexible.
• Users who want to avoid expensive subscription-based agents.
• Creators who wish to package and monetize vertical AI experts (Skills) through encrypted distribution.

Highlights

• Token Efficiency: Claims to be significantly cheaper than other open-source agents like OpenClaw and Hermes.
• MIT Licensed: 100% open source.
• Model Freedom: Supports any OpenAI-compatible API.
• Skill Ecosystem: Natural language skill creation and self-evolving capabilities.
• Multi-Interface: Available via CLI, Web UI, and Docker.