OpenClacky
A token-efficient personal agent written in Ruby
Last verified:
What is OpenClacky?
OpenClacky is an open-source AI workbench that runs AI Agents locally on your machine, using Skills (capability packages) to complete professional tasks. Unlike web-based chat tools, OpenClacky runs on your own computer — it can read and write files, run commands, and control browsers to actually get work done for you. The tool matches Claude Code on capability at comparable cost while saving dramatically on token spend versus mainstream agent frameworks.
Key features include Skills — expert-designed AI workflows invoked with slash commands like /seo-audit, /commit, and /translate — with 20+ built-in Skills and the ability to install more from the marketplace or create your own. OpenClacky offers ultra-high token optimization through smart compression, long-term memory that remembers info across sessions, and prompt caching with a 95.1% 7-day network-wide cache hit rate. It supports flexible provider support connecting to Anthropic Claude, DeepSeek V4, Kimi, Minimax, GLM, MiMo, and any OpenAI-compatible API with automatic failover.
OpenClacky is designed for individuals who want to boost productivity, enterprises needing private deployment with per-employee license management, and domain experts who want to package their methodology into Skills and monetize them. The tool is backed by MiraclePlus, ZhenFund, Yingdong Capital, Variable Capital, Sequoia China, and Hillhouse Ventures, with 473,492 tasks completed and 953 GitHub stars in 30 days since launch.
OpenClacky pricing
Pricing model: Freemium
The OpenClacky client is free forever with no usage limits — it is MIT-licensed open source, not a free trial or restricted tier. Users pay only the model providers (or AI Keys at the same rate) for actual usage. Two ways to connect: Bring Your Own Key (BYOK) where you manage provider accounts yourself and pay model providers directly at their standard rates, or OpenClacky AI Keys which provides unified billing and connection stability billed at published provider rates with no markup. Per-token pricing is identical across both modes. AI Keys provides access and billing convenience — not expanded model capability. Client available for macOS, Linux, and Windows at no cost. The project is funded through thin margins on AI Keys usage, service fees from enterprise engagements, and revenue-share partnerships with independent creators.
OpenClacky pros
- 100% open-source MIT license with fully transparent code
- Runs locally on your machine — your data stays with you
- Claude Code-class capability at lower cost than any agent in its class
- 95.1% 7-day network-wide cache hit rate saves ~86% off input bill
- Only 16 core tools vs 40+ in Claude Code and 52 in Hermes
- Idle-time compression reduces tokens normally burned between turns by 30%+
- Bring Your Own Key (BYOK) — pay model providers directly at standard rates
- Supports Anthropic Claude, OpenAI GPT, DeepSeek, Kimi, Minimax, GLM, MiMo, OpenRouter
- 20+ built-in Skills pre-installed including /commit, /deploy, /translate, /seo-audit
- Skills self-evolve — AI reviews context and outcome after each run to improve reliability
- Long-term memory automatically captures and recalls key decisions across sessions
- Web UI allows running multiple tasks in parallel with separate sessions
- CLI mode for terminal users with no daemon or cold start
- Three-tier permission mode from step-by-step confirmation to full autonomy
- IM integration for Feishu, WeCom, WeChat, Discord, Telegram — invoke with @
- Browser automation to read pages, click buttons, fill forms, download files
- Project rules (.clackyrules) auto-loaded for stack and convention awareness
- Scheduled tasks via cron — describe in natural language, OpenClacky generates schedule
- Under 100 MB install vs Hermes at 48 GB
- Community marketplace for installing and monetizing Skills with revenue share
OpenClacky cons
- Linux does not have a desktop installer — must use command line
- Installer is not yet code-signed — Windows may show Unknown Publisher warning
- Requires user to provide their own API key — no free model usage included
- Built on Ruby — requires Ruby 3.1.0+ installation for CLI mode
- Web UI runs on local server at http://127.0.0.1:7070 — not a cloud service
- Limited to OpenAI-compatible APIs — proprietary non-compatible APIs won't work
- 16 tools means some capabilities offloaded to Skills ecosystem — may need to install more
- Auto-compression happens in background — user has limited control over compression settings
- Memory files stored locally in ~/.clacky/memories/ — no cloud backup option
- Creator Program requires packaging and encryption knowledge for monetization
Frequently asked questions about OpenClacky
Is the client actually free forever? Could pricing change later?
OpenClacky is MIT-licensed with source hosted on GitHub. Open source licenses cannot be revoked. Even if the team's commercial strategy were to change, the community retains the right to maintain a free version based on the current codebase.
How is the project funded?
Two primary channels: AI Keys, reselling model calls at provider cost with thin margins at scale; and enterprise plans with custom engagements. The client itself is not a revenue stream, so there is no commercial incentive to restrict features in the free version.
Are there any client-side differences between BYOK and AI Keys?
None. The client treats both modes as equivalent — every feature is available in either configuration. AI Keys is valued for connection reliability, regional reachability, and consolidated billing, not for model capability.
Under BYOK, does the API key pass through OpenClacky servers?
No. API keys are stored exclusively on the local client. Requests are dispatched from the user's device directly to the model provider. OpenClacky servers do not mediate the call and do not log the key.
Is commercial use permitted? Is additional licensing required?
MIT permits both personal and commercial use without additional licensing. For teams building branded products on top of OpenClacky, please consult the enterprise plan separately.
What models does OpenClacky support?
OpenClacky supports Anthropic Claude (claude-sonnet-4-6 recommended), OpenAI GPT (gpt-5.5), Kimi (kimi-k2.6), DeepSeek V4 (deepseek-v4-pro), Minimax (MiniMax-M3), GLM/ZhipuAI (GLM-5.1), Qwen/Alibaba (qwen3.7-max), MiMo/Xiaomi (mimo-v2.5-pro), and OpenRouter. Any OpenAI-compatible API works out of the box.
How do I install OpenClacky on macOS?
Option 1 (Recommended): Download openclacky-installer.dmg, open the .dmg file, drag OpenClacky into Applications, launch it, and click Open in Browser for Web UI. Option 2: Run the command line script in Terminal which installs Ruby and OpenClacky automatically. Supports both Apple Silicon (M1/M2/M3) and Intel chips.
What are Skills and how do I use them?
Skills are expert-designed AI workflows — no prompt engineering needed. Type / to browse and fuzzy-search Skills, then invoke instantly. Examples: /seo-audit https://example.com for SEO diagnosis, /commit for auto-generating Git commit messages, /translate for professional translation. You can build Skills by chatting with the agent, and Skills self-evolve after each run to perform more reliably.
How does token optimization work in OpenClacky?
Four things: Ultra-high cache hit rate (near 100%) through session never restarting and double-window cache markers; Minimal toolset (16 tools) with capabilities extending via invoke_skill meta-tool; Idle-time compression where the agent quietly compresses long context in background during between-turn gaps (which account for 30%+ of total session cost); and model routing choice where you can use Claude for main tasks and auto-route subagents to cheaper models like DeepSeek.
What is the Creator Program?
Professionals can package their workflows into encrypted Skills, deploy privately to every employee's desktop, and monetize them. Examples include QingClaw (10 years of legal expertise for law firms), JoyClaw (AI life planning assistant for high net worth individuals), and MediClaw (junior doctors accessing senior healthcare expertise). The program offers encrypted distribution, license management, and creator-defined pricing with revenue share through OpenClacky.