AG3NT
Local-first multi-agent platform built on DeepAgents. Gateway, Agent Worker, and Web UI for orchestrating autonomous AI agents.
Install / Use
claude mcp add AP3X-Dev -- npx -y github:AP3X-Dev/AG3NTIf the server publishes to npm under a different name, use that package instead — check the repo README.
MCP Server
Model Context Protocol server
Quality Score
Category
AI & Machine LearningSupported Platforms
Skill content
View source on GitHubAG3NT
<img src=".github/images/AG3NT_header.png" alt="AG3NT" width="100%"/>Local-first personal AI agent platform built on DeepAgents.
Features
- 🤖 Multi-Model Support - Anthropic, OpenAI, OpenRouter, Kimi, Google Gemini
- 🔌 Multi-Channel - CLI, TUI, Telegram, Discord adapters
- 🛠️ Agent Skills - Modular skill system with SKILL.md format
- 🌐 Browser Control - Playwright-based web automation (navigate, screenshot, click, fill)
- 🔒 Security - DM pairing, HITL approval for sensitive actions
- ⏰ Scheduler - Heartbeat checks and cron-based automation
- 🖥️ Multi-Node - Primary + companion device architecture
Repo Layout
ag3nt/
├── apps/
│ ├── gateway/ # Gateway daemon (HTTP + WebSocket + channels)
│ ├── agent/ # Agent worker (DeepAgents runtime)
│ ├── ui/ # Web dashboard (Next.js)
│ └── tui/ # Terminal UI client
├── skills/ # Bundled Agent Skills (SKILL.md format)
├── config/ # Default configuration templates
└── docs/ # Planning documents
🖥️ Web Dashboard
The AP3X-UI provides a comprehensive web interface for AG3NT:
Running the UI
Windows (Unified Script - Recommended):
.\start.ps1
This starts Gateway, Agent Worker, and UI together. Access at http://localhost:3000
Manual Start:
# Terminal 1: Start AG3NT Gateway
cd apps/gateway && npm run dev
# Terminal 2: Start AG3NT Agent Worker
cd apps/agent && .venv/Scripts/activate && python -m ag3nt_agent.worker
# Terminal 3: Start UI Dashboard
cd apps/ui && npm run dev
Access the dashboard at http://localhost:3000
UI Features
- Real-time chat with streaming
- Artifact library and management
- Skills and tools browser
- Subagent configuration
- MCP server manager
- Browser automation interface
- System monitoring and logs
Quick Start
Windows (One Command)
# Start everything: Gateway + Agent + UI
.\start.ps1
# Stop all services
.\stop.ps1
Manual Setup
1. Copy Configuration
# Create config directory
mkdir -p ~/.ag3nt
# Copy default config
cp config/default-config.yaml ~/.ag3nt/config.yaml
2. Start Gateway
cd apps/gateway
pnpm install
pnpm dev
Gateway runs on http://127.0.0.1:18789
3. Start Agent Worker
cd apps/agent
python -m venv .venv
# Activate virtual environment
# Windows:
.venv\Scripts\activate
# macOS/Linux:
source .venv/bin/activate
pip install -r requirements.txt
python -m ag3nt_agent.worker
Worker runs on http://127.0.0.1:18790
4. Start Web UI
cd apps/ui
npm install
npm run dev
UI runs on http://localhost:3000
5. Start TUI (Optional)
cd apps/tui
pip install -r requirements.txt
python ag3nt_tui.py
Milestone Status
| Milestone | Status | Description | |-----------|--------|-------------| | M1: Core Agent Runtime | ✅ Complete | DeepAgents integration, multi-model support | | M2: Modular Skill System | ✅ Complete | SKILL.md format, skill discovery, execution runtime, trigger matching | | M3: Gateway & Multi-Channel | ✅ Complete | HTTP/WS API, Telegram/Discord adapters | | M4: Planning & Memory | ✅ Complete | TodoListMiddleware, memory persistence | | M5: Secure Execution | ✅ Complete | HITL approval flow, DM pairing security | | M6: Scheduling | ✅ Complete | Heartbeat system, cron jobs | | M7: Multi-Node | ✅ Complete | WebSocket protocol, pairing, capability routing | | M8: Control Panel | ✅ Complete | Web UI, skill management, debug logs |
Active Development
See ROADMAP.md for detailed sprint planning and current priorities:
- Core Tools: Shell execution, web search, git operations
- Skill Execution: Runtime for skill entrypoints, MCP integration
- Testing: Unit and E2E test coverage
Documentation
- Agent Worker - Model providers and worker API
- Web Dashboard - Next.js web interface
- TUI Client - Terminal interface usage
- Gateway API - HTTP/WebSocket API reference
- Control Panel - Web-based control panel
- Multi-Node Architecture - Companion device support
- Skills - Skill format documentation
Environment Variables
| Variable | Description | Default |
|----------|-------------|---------|
| AG3NT_MODEL_PROVIDER | LLM provider (anthropic, openai, openrouter, kimi, google) | openrouter |
| AG3NT_MODEL_NAME | Model name | moonshotai/kimi-k2.5 |
| ANTHROPIC_API_KEY | Anthropic API key | - |
| OPENAI_API_KEY | OpenAI API key | - |
| OPENROUTER_API_KEY | OpenRouter API key | - |
| KIMI_API_KEY | Kimi/Moonshot API key | - |
| GOOGLE_API_KEY | Google Gemini API key | - |
License
MIT
Related Skills
momen-cursurrules-prompt-file
40.6kCursor rules for building custom frontends with Momen.app as headless BaaS with GraphQL API, actionflows, AI agents, and Stripe integration.
pyspark-etl-best-practices-cursorrules-prompt-file
40.6kCursor rules for PySpark ETL development with code style, joins, window functions, map operations, and Iceberg patterns.
semiotic-react-dataviz-cursorrules-prompt-file
40.6kCursor rules for Semiotic data visualization library with 30+ chart types, MCP server, and AI-assisted chart generation.
claude-mem
90.6kPersistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More
