Core Features Overview

OpenClacky is a fully-featured local AI Agent platform. This guide helps you quickly understand all core capabilities.

Session Management

A session is a persistent conversation context between you and the Agent, storing complete message history and task records.

Start the server and visit http://127.0.0.1:7070. From the sidebar you can create, switch, rename, and delete sessions. Each session is fully isolated, and automatically restored after server restart.

📖 Session Management

Long-term Memory System

After complex sessions, the Agent automatically saves important knowledge — decisions, architecture choices, your preferences — to ~/.clacky/memories/ for future sessions to reuse. Fully automatic, no manual trigger needed.

The Agent retrieves relevant memories automatically, or you can ask explicitly:

Recall the decisions we made about the authentication system

📖 Memory System

Smart Compression & History Review

When conversation history gets long, the Agent automatically compresses earlier messages while keeping recent context intact — so you can work on long, multi-task sessions without worrying about costs.

You can also ask the Agent to review at any time:

What did we discuss earlier?

Skill System

Skills are reusable AI capability packages that give the Agent specialized expertise.

Just describe your needs and the Agent picks the right Skill automatically. You can also invoke them directly with /, for example /browser-setup, /deploy.

OpenClacky comes with built-in Skills for browser automation, IM integration, scheduled tasks, code exploration, project deployment, and more. Third-party and custom Skills are also supported.

📖 How to Use a Skill · Built-in Skills

Scheduled Tasks

Create automated tasks that run on schedule without manual intervention.

Just describe what you want:

Create a task that fetches AI news every morning at 9 AM

All tasks can be managed in the WebUI "Tasks Panel": view schedules, enable/disable, trigger manually, and view execution logs.

Note: Keep openclacky server running for scheduled tasks to fire automatically.

📖 Scheduled Tasks

Channel Connection (IM Integration)

Connect OpenClacky to Feishu, WeCom, or WeChat to chat with the Agent directly in IM.

/channel-setup

After setup, @mention the bot in group chats or chat directly. Each IM conversation is linked to an independent Session.

📖 Channel Configuration

Browser Automation

Built-in browser automation lets the Agent control Chrome/Edge to handle logins, web scraping, form filling, UI interactions, and more.

Run /browser-setup once to configure, then just describe what you need:

Open GitHub and save a screenshot to the desktop

📖 Browser Automation Tool

Project Rules File (.clackyrules)

Create a .clackyrules file in your project root to define your tech stack, code conventions, and common commands. The Agent loads it automatically whenever you start in that directory — no need to re-explain your project every time.

📖 Project Rules File

Brand Skills (Licensed Skills)

Businesses and creators can package their expertise into licensed Skills and distribute them to buyers.

Activate your license: Click the Settings icon (⚙) in the top-right of WebUI → "License Activation" → enter your key and click Activate. The WebUI then navigates to the "Brand Skills" tab automatically — click "Install" to download and use.

The "Skill Store" tab also offers publicly available Skills that anyone can install without a license.

Create Custom Skills

Encapsulate your specific workflows into Skills for reuse or sharing.

/skill-creator

The Agent guides you through defining the use case, writing instructions, and testing. Finished Skills can also be published to the OpenClacky marketplace.

📖 Skill Basics · Publish to Marketplace

Skill Evolution (Auto-Create & Auto-Improve)

Your skill library grows as you use the Agent — you don't have to write it all upfront. When a task finishes, OpenClacky automatically decides whether it learned something worth keeping:

• Scenario 1 — Auto-create a new Skill. If you completed a complex task using only raw tools (no existing Skill), the Agent evaluates whether the workflow is reusable. If yes, it invokes skill-creator in quick mode and writes a new Skill to ~/.clacky/skills/. No interruption, no prompts.

• Scenario 2 — Auto-improve an existing Skill. If you invoked one of your Skills via a slash command and the run was long enough to be meaningful, the Agent reflects on it: Were the instructions clear? Any edge cases missing? If there are concrete, actionable improvements, it updates the Skill's SKILL.md for you.

Both scenarios run in an isolated subagent, so your main conversation is never polluted. The two are mutually exclusive by design — you never get a duplicate Skill auto-extracted from a task that was already served by an existing one.

Scope rules. Only Skills in ~/.clacky/skills/ or .clacky/skills/ are eligible for auto-improvement. Built-in default Skills and licensed brand Skills are never modified by evolution. Only the main Agent triggers evolution, not subagents — no recursive blow-up.

Very few agent platforms ship this capability. It is a deliberate design choice: the Skill system should compound over time, the same way memory does, turning day-to-day usage into a growing, personalized toolkit.

You can disable the feature in your agent config if you prefer strict control:

skill_evolution:
  enabled: false

CLI Terminal Mode

If you prefer working in the terminal, just run openclacky to enter chat mode directly — no server needed.

openclacky

Key commands:

• /config — Configure your model and API Key
• /xxx — Invoke a specific Skill
• Shift+Tab — Switch between L3 (semi-auto) and L4 (fully-auto) mode
• exit or Ctrl+C — Quit

CLI mode supports all tools and Skills, sharing the same configuration and memory system as the Web UI. It's the preferred choice for power users who live in the terminal — and an open-source alternative to Claude Code.

📖 CLI Reference

Quick Start Recommendations

Beginner path: Install & start → Configure API Key (/config) → Try a conversation → Explore Skills → Configure browser → Create scheduled tasks

Advanced usage: Define .clackyrules → Connect IM → Create custom Skills → Publish to marketplace

Power users: Use CLI mode and integrate it into your daily terminal workflow

FAQ

Q: Where are sessions saved?

A: ~/.clacky/sessions/, automatically restored after server restart.

Q: How to reduce Token costs?

A: Built-in smart compression and Prompt Caching optimize automatically for long conversations.

Q: Do scheduled tasks run when my computer sleeps?

A: No — openclacky server must be running and the computer awake.

Q: Which LLM providers are supported?

A: OpenAI, Anthropic, DeepSeek, MiniMax, OpenRouter, and more. See Agent Configuration.

📖 More questions: FAQ

Next Steps

• Skill Creation Practices — Tips for using OpenClacky efficiently
• Harness Engineering — OpenClacky's agent runtime
• Publish Your First Skill — Become a creator