Ask your ERP anything. In plain English. No training required.
A public-facing AI chatbot built as a native Odoo 18 module, powered by a RocketRide AI pipeline and Qwen (Alibaba Cloud) as the LLM backend. Visitors and staff can query live company data using natural language — no forms, no filters, no manual searching.
Enterprise ERP systems are powerful but hard to use. Most employees spend time navigating menus, applying filters, and cross-referencing data just to answer simple questions.
This project replaces that friction with a single chat interface:
"Who is Ethan?" → Ethan Hunt, HR Manager, HR Department. Work email: Ethan@gmail.com.
No clicks. No training. Just ask.
The chatbot is embedded directly in the Odoo website frontend. Visitors type a question, the system queries live Odoo data via an AI pipeline, and streams back a natural language answer.
Browser (Odoo Website)
│
│ GET /chatbot/stream (Server-Sent Events)
▼
Odoo Controller (main.py)
│
│ WebSocket ws://localhost:5565
▼
RocketRide AI Pipeline (hr_chat.pipe)
│
├── Agent Node (orchestrates tool use)
├── LLM Node (Qwen-Plus via Alibaba Cloud)
└── Tool Node (HTTP → Odoo HR API)
│
│ GET /api/v1/employees/search
▼
Odoo HR Module (hr.employee)
| Decision | Rationale |
|---|---|
| RocketRide pipeline | Visual, swappable AI workflow — change LLM or tools without touching application code |
| Qwen-Plus (cloud LLM) | Fast inference, strong multilingual support, OpenAI-compatible API |
| SSE streaming | Immediate feedback to the user — responses appear word by word |
| Separate HR API endpoint | Clean boundary between AI pipeline and Odoo ORM; API-key authenticated |
| No dependency on odoo-llm | Standalone module — only requires website and hr from Odoo core |
| Layer | Technology |
|---|---|
| ERP Platform | Odoo 18 |
| Frontend | Odoo Website (QWeb + Vanilla JS) |
| Streaming | Server-Sent Events (SSE) |
| AI Pipeline | RocketRide Engine (Docker) |
| LLM | Qwen-Plus (Alibaba Cloud / DashScope) |
| Data Layer | Odoo ORM — hr.employee |
| Auth | HMAC API key (pipeline → Odoo) |
- Natural language queries — ask about employees, departments, job titles in plain text
- Live ERP data — answers come from your actual Odoo database, not a static knowledge base
- Streaming responses — word-by-word output for a responsive feel
- Public-facing — no Odoo login required for visitors
- Role-based access — visitors see public info only; logged-in staff and HR managers see work emails
- Swappable LLM — change from Qwen to OpenAI, Anthropic, or a local Ollama model by editing one pipeline file
- Extensible pipeline — add new data sources (calendar, inventory, CRM) as new tool nodes in RocketRide
website_llm_chat/
├── controllers/
│ ├── main.py # /chatbot page + /chatbot/stream SSE endpoint
│ └── hr_api.py # /api/v1/employees/search — HR data API
├── pipelines/
│ └── hr_chat.pipe # RocketRide pipeline definition
├── tests/
│ └── test_rbac.py # Unit tests for RBAC pure functions
├── static/src/
│ ├── js/chatbot.js # Frontend SSE consumer + chat UI
│ └── css/chatbot.css
├── views/
│ ├── templates.xml # Chatbot page template
│ └── menu.xml # Website navigation entry
├── rocketride_client.py # Async RocketRide WebSocket client
├── rbac.py # Role → allowed fields mapping (pure, no Odoo imports)
├── __manifest__.py
└── CLAUDE.md # Architecture notes
- Odoo 18 instance
- Docker
- Qwen API key (Alibaba Cloud DashScope) — or swap for any OpenAI-compatible LLM
docker run -d \
--name rocketride-engine \
-p 5565:5565 -p 20003:20003 \
--env-file .env \
ghcr.io/rocketride-org/rocketride-engine:latestcp .env.example .env
# Fill in:
# ROCKETRIDE_URI=ws://localhost:5565
# ROCKETRIDE_APIKEY=your_rocketride_key
# ROCKETRIDE_QWEN_API_KEY=your_qwen_key
# ROCKETRIDE_ODOO_HR_API_KEY=your_shared_secret
# ODOO_HR_API_KEY=your_shared_secret (same value)# Add to your Odoo addons path, then:
./odoo-bin -d your_db -i website_llm_chatNavigate to http://your-odoo/chatbot — the chatbot is live.
The pipeline is modular. To add calendar or inventory queries:
- Add a new HTTP tool node in
hr_chat.pipepointing to a new Odoo API endpoint - Create the corresponding controller in
controllers/ - Update the agent's system prompt to describe the new capability
No changes to core application logic required.
- Conversation memory (multi-turn context)
- Role-based access (visitor vs. staff / HR manager)
- MCP Server integration — expose Odoo tools to Claude Desktop / Cursor
- Calendar & appointment queries
- True token-level streaming from LLM
Built as a learning project exploring enterprise AI agent architecture on Odoo 18. The goal: design a system where natural language becomes the universal interface to ERP data — extensible, provider-agnostic, and deployable on real infrastructure.
Core concepts demonstrated:
- AI pipeline orchestration with RocketRide
- Tool calling — LLM decides when and how to query live data
- Streaming UX on top of a synchronous AI backend
- ERP integration patterns — clean API boundaries between AI and business logic
LGPL-3.0