crabagent 0.6.1

CrabAgent - AI Agent Platform with dual-mode (CLI + Serve)

🦀 CrabAgent

AI Team Command Center — Build a team of specialized AI agents, delegate tasks in parallel, and watch them work in real-time. All from a local web dashboard.

CrabAgent is a local-first AI agent platform. Run it from any project directory via CLI or browser. Your data stays local, your API keys are encrypted, and you pick any LLM provider.

---

What Makes It Different

Feature	Description
🤖 AI Team	Create custom agent profiles; delegate tasks to multiple agents in parallel; each agent can use its own model
🧠 Team Memory	Persistent team knowledge + auto-extracted lessons; agents remember across sessions
📋 Task Board	Real-time right-side panel showing each agent's status (running/done/error), progress, and tool calls
@mention Delegation	Type @researcher search for X and CrabAgent auto-delegates, or click agents from the toolbar
🔀 Parallel Execution	Multiple agents run simultaneously — researcher searches while coder debugs while analyst compares
📊 Result Compare	Side-by-side view of all agent outputs, with one-click Markdown export
⏱ Scheduled Tasks	Agents run autonomously on a cron schedule with push notifications
🌐 Browser Automation	Playwright-powered headless browser — navigate, click, screenshot, extract
🖼️ Multimodal	Paste, upload, or drag images directly into chat; auto-detects vision-capable models
🔌 MCP Client	Connect external MCP servers (stdio + HTTP); tools auto-discover and prefix
🦀 Snapshots	Auto-snapshot files before changes; rollback anytime without Git
🔒 Privacy	All data stays local; API keys encrypted at rest; no telemetry

---

Quick Start

pip install 'crabagent[serve]'

crabagent init

# TUI — interactive REPL with slash commands
crabagent

# Web UI
crabagent --serve          # → http://localhost:5210
                           # Default login: admin / xcl1989

# Single-query CLI
crabagent "organize this directory"
crabagent -p deepseek -m deepseek-chat "write a Python script"

---

AI Team — Your Local Command Center

1. Build your team — Sidebar → 🤖 Team → + New Agent. Define each agent's role, goal, backstory, and model override.

2. Delegate tasks — Three ways to delegate:

• Type @researcher find competitor pricing in the chat — auto-detected and sent
• Click the 🤖 button next to the input → select agents → enter task → send
• Click an agent from the toolbar above the input to insert a @mention

3. Watch in real-time — The right-side Task Board shows every agent's progress:

• 🟣 Running — purple pulsing card with live step count and timer
• 🟢 Done — green card with elapsed time, tokens, and iteration count
• 🔴 Error — red card with error summary

4. Review results — Click any card to open the agent's full output, or click 📋 in the top toolbar for a split-pane comparison view of all results. Export to Markdown with one click.

5. Go parallel — Use delegate_parallel to run multiple agents simultaneously, or the web delegation modal to assign different tasks to different agents.

Built-in agents:

Agent	Role	Best For
🔍 Researcher	Web research	Search, browse, data collection
📊 Analyst	Data analysis	Comparison, pattern detection, reports
💻 Coder	Code expert	Write, review, debug, refactor
📝 Writer	Content writer	Write, edit, translate, format

---

Team Memory

Agents share persistent memory across sessions. Two types:

Type	Scope	Example
Team Knowledge (team)	All agents share	Tech stack, architecture decisions, user preferences
Agent Lessons (lesson)	Specific agent	Effective strategies, failed approaches, tool tips

How it works:

• Agents use memory_save() to store knowledge (e.g. user chooses a framework, records it)
• memory_recall(query) searches by keywords — split-term matching finds partial hits
• memory_replace(key, old, new) lets agents edit existing memories precisely
• Top team knowledge is auto-injected into system prompts on startup
• After sub-agent tasks complete, lessons are auto-extracted (high iterations → efficiency tip, etc.)

CLI: /memory list · /memory search <query> · /memory clear

---

Scheduled Tasks

Tasks run autonomously on a cron schedule. Define them via conversation or the ⏱ Tasks panel in the sidebar.

> Remind me to drink water every day at 11:00
> Check this product page every 30 minutes, notify me if price drops below 500

When a task completes, a notification appears in the bell icon. Click to jump to the execution's conversation — full message history, screenshots, and tool outputs are preserved.

---

Browser Automation

Controlled via Playwright. Install with pip install 'crabagent[browser]' then playwright install chromium.

Available tools: browser_navigate, browser_click, browser_type, browser_screenshot, browser_extract, browser_scroll.

> Open https://news.ycombinator.com and show me the top 5 stories
> Search for "Python async" on Google and extract the results

The browser starts lazily (first call only), shares one instance per conversation, and auto-closes on session end. Screenshots appear inline in the chat.

---

Image / Multimodal Support

Paste (Ctrl+V), upload, or drag images into the chat. CrabAgent auto-detects whether the current model supports vision — vision models get native multimodal content, non-vision models get a file-path placeholder.

• Max 5 images per message, 5MB each
• Supported: PNG, JPEG, GIF, WebP
• CLI: /image /path/to/image.png What's in this image?

---

MCP Client

Connect external MCP servers via stdio or HTTP. Tools auto-discover with the prefix mcp__{server}__{tool} and are visually distinguished in the chat.

Manage servers from the MCP panel in the sidebar — add, connect, disconnect, view tool counts and connection status.

---

Web Search & Custom Plugins

Web Search: Built-in web_search (DuckDuckGo, zero-config) and web_scrape. Optionally configure SearXNG for better results.

Custom Plugins: Drop a .py file in .crabagent/tools/:

name = "hello"
description = "Say hello to someone"
parameters = {
    "type": "object",
    "properties": {"name": {"type": "string", "description": "Name"}},
    "required": ["name"],
}
requires_permission = False

def run(name: str) -> str:
    return f"Hello, {name}!"

---

CLI Commands

Command	Description
/exit, /quit	Exit
/help	Show help
/clear	Clear conversation context
/model [name]	Switch model
/models	List available models
/provider [cmd]	Manage LLM providers (list/add/remove/set-default)
/sessions	List recent sessions
/session [id]	Load a session
/new	Start new conversation
/agents [cmd]	Agent team management (list/add/edit/toggle/rm)
/delegate [@agent] [task]	Delegate task to agent(s)
/memory [cmd]	Team memory (list/search/clear)
/molt [cmd]	Snapshot list/show/rollback
/todo [cmd]	Manage task list
/export	Export conversation to Markdown
/skills	List available skills
/image <path> [msg]	Send an image

---

Configuration

Set via environment variables or .env:

Variable	Default	Description
CRAB_DB_URL	sqlite+aiosqlite:///./crabagent.db	Database URL
CRAB_JWT_SECRET	Auto-generated	JWT signing key
CRAB_SERVE_HOST	0.0.0.0	Serve host
CRAB_SERVE_PORT	5210	Serve port
CRAB_MAX_ITERATIONS	50	Max agent iterations
CRAB_MAX_TOKENS	4096	Max response tokens
CRAB_BROWSER_HEADLESS	true	Browser headless mode

---

Installation

pip install 'crabagent[serve]'          # Web UI + API
pip install 'crabagent[browser]'        # Browser automation
pip install 'crabagent[dev]'            # Testing + linting

Docker:

docker compose up -d

---

Project Structure

CrabAgent/
├── .crabagent/
│   ├── skills/        # Domain skills (SKILL.md)
│   ├── tools/         # Custom plugin tools
│   └── molts/         # File snapshots
├── src/crabagent/
│   ├── cli/           # CLI entrypoint
│   ├── core/agent/    # Agent loop, tools, context, compression
│   ├── core/mcp/      # MCP client manager
│   └── serve/         # FastAPI + API + scheduler
├── frontend/          # React SPA
├── crabagent.db       # SQLite database
└── Makefile

---

Development

make install            # Build frontend + install (editable)
ruff check src/ tests/  # Lint
ruff format src/ tests/ # Format
pytest                   # Run tests

---

License

GNU Affero General Public License v3 (AGPLv3) for non-commercial use. Commercial use requires a separate license. Contact the author.

See LICENSE.