Intent Hub 🚀

A static routing system based on vector similarity that dispatches user requests to the right AI Agent via semantic matching.

English | 中文

📹 Video Demo: Watch on YouTube

---

🗺️ Navigation & Quick Start

⚡ One-Click Run (Recommended)

The project is fully containerized and supports one-command deployment.

1. Prepare environment:

   cp env.example .env

   Edit .env and set your LLM_API_KEY (for auto corpus generation) and other options.

2. Start everything:

   docker compose up -d

   Or with China mirror config:

   docker compose --env-file .env.china up -d --build

After startup:

• Admin UI (Frontend): http://localhost (port 80 by default)
• API (Backend): http://localhost:8000
• Vector DB (Qdrant): http://localhost:6333/dashboard

Note:

• First run will pull/build images and download the Embedding model; ensure network access.
• A data/ directory is created at the project root for persisting routes and settings.

---

⚙️ Configuration

Configuration priority: Environment variables > DB/JSON config > Defaults.

Environment variables (.env)

Key options in the root .env file:

• LLM_API_KEY: LLM API key (required for corpus enhancement).
• LLM_PROVIDER: Provider (e.g. deepseek, qwen, openai).
• EMBEDDING_MODEL_NAME: Embedding model name.
• QDRANT_URL: Qdrant connection URL.

See comments in env.example for details.

Data persistence

User config and route data live under ./data:

• data/routes_config.json: Routes and corpus config.
• data/settings.json: System settings.
• data/env.runtime: Auto-generated runtime env file (updated when you save settings in the UI).

• Saving settings in the UI writes to data/settings.json and updates data/env.runtime so containers keep the same config after restart.
• Disable sync: set INTENT_HUB_ENV_SYNC_ENABLED=false.
• Custom sync path: INTENT_HUB_ENV_SYNC_PATH=/app/data/env.runtime.
• Custom sync keys: INTENT_HUB_ENV_SYNC_KEYS=QDRANT_URL,LLM_PROVIDER,LLM_API_KEY.

In Docker, this directory is mounted as a volume so data survives container removal.

---

🏗️ Architecture

Frontend/backend split with vector search for fast intent routing.

🔹 Backend (Python / FastAPI)

• Stack: Python 3.9+ + FastAPI
• Vector store: Qdrant
• Models:
  • Embedding: Qwen-Embedding-0.6B (HuggingFace / local)
  • LLM: LangChain integration for DeepSeek, OpenAI, Qwen, etc.
• Responsibilities: Intent recognition, vector sync, auto corpus generation, route management API.

🔹 Frontend (Vue / Vite)

• Stack: Vue 3 + Vite
• UI: Element Plus
• Responsibilities: Route CRUD UI, corpus generation, system config, vector match testing.

---

📂 Project layout

intenthub/
├── data/                  # Persisted data (routes, settings)
├── intent-hub-backend/    # Backend (Python/FastAPI)
│   ├── intent_hub/        # Core (encoding, search, services)
│   ├── tests/             # Unit tests
│   └── run.py             # Entry point
├── intent-hub-frontend/   # Frontend (Vue/Vite)
│   ├── src/               # Pages, components, state
│   └── vite.config.ts    # Build config
├── docker-compose.yml     # Full-stack compose
├── env.example            # Env template
└── README.md              # This file

---

📄 License

Distributed under the MIT License. See LICENSE for details.