Chatbox AI — LLM Assistant for Odoo 19
149 automated testsA personal R&D project: an AI assistant living inside Odoo Discuss that answers natural-language questions about live ERP data — and only the data the asking user is allowed to see.
- Problem
- Finding information in an ERP requires knowing which model to search and how to build domain filters. Answers are often scattered across records and PDF documents, out of reach for non-technical users.
- Solution
- A standalone Odoo 19 module: semantic search over configurable models using pgvector + OpenAI embeddings, document RAG over PDF attachments, and an LLM tool-calling loop (semantic search, record search, counting/grouping, menu & field introspection). Replies render as sanitized markdown with clickable links to real records, generated asynchronously via a job queue with a native typing indicator. Chat provider is switchable (Anthropic Claude, DeepSeek, any OpenAI-compatible API).
- My role
- Solo developer, end to end: design, spec, TDD implementation, security review, deployment on a live Odoo 19 instance.
- Result
- 149 automated tests. Security-hardened by design: retrieval always filtered by Odoo record rules, LLM HTML output sanitized with a trusted-link whitelist, tools gated per user group.
- Odoo 19
- Python
- PostgreSQL + pgvector
- OpenAI embeddings
- Anthropic API
- DeepSeek
- RAG
- Docker







