~/work / odoo-ai
Odoo AI suite
Sole developer · Zamakan
AI modules that bring a self-hosted LLM into Odoo ERP, so business data never leaves the company's own servers. Built at Zamakan Management Consultancies and delivered into a client's production ERP.
- Year
- 2025–26
- Role
- Sole developer of the AI modules, as Odoo Technical Developer at Zamakan Management Consultancies.
- Platform
- Odoo 19 (Python and Owl / JavaScript)
- Stack
- Python · Ollama or any OpenAI-compatible endpoint · nomic-embed-text · pgvector · Web Speech API · Piper TTS
- Status
- Delivered into a client's production ERPPrivate client code, so there's no public link
- Odoo modules, from model provider to audit agent
- 4
- records audited per app in each run
- 3–5k
- per app audit, work that previously took a team days
- ~5 min
Overview
Odoo's AI features expect a cloud model provider. For clients who can't send ERP data to an outside API, I built modules that plug a self-hosted model into Odoo instead. I then used that foundation for voice features and automated audits that run on the company's own hardware.
Highlights
- 01
A local LLM provider for Odoo
Connects Odoo's AI app to Ollama or any OpenAI-compatible endpoint, for both chat and embeddings (nomic-embed-text). Retrieval runs on pgvector for RAG and semantic search, with guards so tool calls work reliably on a local model.
- 02
Voice in and out
Voice-to-text in Odoo Discuss using the browser's Web Speech API, and an AI voice-call dialog that answers out loud with Piper text-to-speech running on the server.
- 03
Odoo Brain, a scheduled audit agent
You give it a guidebook of business rules and pick the apps it covers. It reviews 3,000–5,000 records per app in about 5 minutes and records each issue with its severity, evidence and a suggested fix, producing reproducible audit reports.
- 04
Automation hub for dashboards
Scheduled AI runs over Odoo spreadsheet dashboards. It pulls the key numbers, applies rules, asks the model for a summary and logs alerts with severity, evidence, impact and a suggested fix, with record-level drilldowns for stakeholders.