OpenClacky: The Token-Efficient Open-Source AI Agent — Complete Guide

OpenClacky is the most token-efficient open-source AI agent on the market today. It matches Claude Code on capability while costing roughly 0.8× the tokens of the baseline — and only about one-third of other open-source agents. Built on two years of agentic R&D, it is 100% open source under the MIT license and works with any OpenAI-compatible model through a bring-your-own-key (BYOK) setup.

Why token efficiency matters

For heavy agent workloads, token spend is the real cost of running an AI assistant. OpenClacky attacks that cost at every layer instead of cutting features:

The Skill ecosystem — the soul of the agent

OpenClacky keeps only 16 core tools and offloads everything else to Skills, invoked through a single invoke_skill meta-tool:

• Invoke with / — instant browse, fuzzy search, direct call across hundreds of Skills.
• Create Skills in natural language — describe what you want; the agent drafts SKILL.md, breaks down steps, and validates it. No code required.
• Self-evolving — after each run the agent updates the Skill from execution context, so the next call is more stable and accurate.
• Open and compatible — supports Claude Skills, Markdown Pack, and custom formats.

Model freedom and integrations

Bring any OpenAI-compatible API key: official direct, aggregate routing, or compatible relays. Use Claude for code and auto-route subtasks to a cheaper model like DeepSeek. Connect channels including Discord, Telegram, Feishu, WeCom, and WeChat, backed by MiraclePlus, ZhenFund, Sequoia China, and Hillhouse Capital.