OCA Neural Reviewer

AI-powered PR review tool for OCA (Odoo Community Association) repositories.

Uses local AI models (via Ollama) and Kubernetes (via Minikube) to automatically test and review Pull Requests across OCA's 800+ repositories.

Developed by POP Coop — a cooperative committed to open source and community-driven development.

What It Does

For each selected OCA repo:
  └─ List open PRs for your Odoo version
       └─ For each PR (skips already reviewed):
            ├─ Detect changed addons
            ├─ Create K8s Job with oca-ci image
            ├─ Clone repo, checkout PR, run Odoo tests
            │
            ├─ PASSED → Approve + AI suggests test improvements
            │           + request reciprocal reviews @mentioning
            │           all participants
            │
            └─ FAILED → AI analyzes root cause + suggests fix
                        + suggests better tests
                        + @mentions all participants

Quick Start

git clone https://git.pop.coop/pop/oca-reviewer.git
cd oca-reviewer
chmod +x oca-reviewer scripts/*.sh
./oca-reviewer

The interactive wizard will guide you through:

1. GitHub authentication (token or gh CLI)
2. Hardware allocation (CPU, RAM, disk for Kubernetes)
3. Odoo version (14.0, 16.0, 17.0, 18.0)
4. AI model selection (Qwen3-coder:30b recommended)
5. Repository selection (checkbox TUI with all OCA repos)
6. Review behavior (approve, comment, suggest tests)

Prerequisites

Software	Purpose
Docker	Container runtime
Minikube	Local Kubernetes
kubectl	Kubernetes CLI
gh	GitHub CLI
dialog	TUI dialogs
Ollama	Local AI inference
python3	JSON parsing

Install on Manjaro/Arch

sudo pacman -Syu docker minikube kubectl github-cli dialog python
curl -fsSL https://ollama.com/install.sh | sh
ollama pull qwen3-coder:30b

Install on Ubuntu/Debian

sudo apt install docker.io kubectl dialog python3
# minikube: https://minikube.sigs.k8s.io/docs/start/
# gh: https://cli.github.com/
curl -fsSL https://ollama.com/install.sh | sh
ollama pull qwen3-coder:30b

Usage

# First run (setup wizard)
./oca-reviewer

# Review all selected repos
./oca-reviewer review

# Review specific repos
./oca-reviewer review server-tools l10n-brazil

# Force re-review everything (ignore cache)
./oca-reviewer review -f

# Change repo selection
./oca-reviewer repos

# Reconfigure settings
./oca-reviewer config

# Check infrastructure status
./oca-reviewer status

# Destroy everything
./oca-reviewer teardown

Architecture

┌─────────────────────────────────────────────────────────┐
│                    Your Machine                         │
│                                                         │
│  ┌──────────┐  ┌────────────┐  ┌────────────────────┐  │
│  │  Ollama   │  │  Runboat   │  │     Minikube       │  │
│  │(AI Model) │  │ Controller │  │ ┌────────────────┐ │  │
│  │          │  │  (docker)  │──▶│ │ runboat-builds │ │  │
│  └────┬─────┘  └────────────┘  │ │  ┌──────────┐  │ │  │
│       │                        │ │  │ OCA CI   │  │ │  │
│       │                        │ │  │ Job/Pod  │  │ │  │
│  ┌────┴───────────────────┐    │ │  └──────────┘  │ │  │
│  │    oca-reviewer        │    │ └────────────────┘ │  │
│  │  (orchestrator + gh)   │    │ ┌────────────────┐ │  │
│  └────────────────────────┘    │ │ runboat-infra  │ │  │
│                                │ │  ┌──────────┐  │ │  │
│                                │ │  │PostgreSQL│  │ │  │
│                                │ │  └──────────┘  │ │  │
│                                │ └────────────────┘ │  │
│                                └────────────────────┘  │
└─────────────────────────────────────────────────────────┘

AI Review Quality

The tool uses a carefully crafted system prompt that teaches the AI model OCA-specific patterns:

• Modifiers can be domains (not just booleans)
• Extensibility hooks (_get_*, _prepare_* returning False/[] are intentional)
• OCA testing patterns (TransactionCase, SavepointCase, @tagged)
• setup_modifiers overriding is valid in mixins
• fields_get() returns metadata, not field objects

Recommended Models

Model	VRAM	Quality	Speed
qwen3:8b	6 GB	Basic	Fast
qwen3-coder:30b	16 GB	Recommended	Moderate
qwen3:32b	24 GB	Excellent	Slow
qwen3:72b	48 GB	Best	Very slow

Features

• Smart tracking: Only re-reviews PRs with new commits (SHA-based cache)
• Reciprocal reviews: @mentions all PR participants and links your open PRs
• Test suggestions: AI analyzes coverage gaps even when tests pass
• Multi-repo: Processes 800+ OCA repos with interactive selection
• Resilient: Falls back from Runboat to direct K8s Jobs if needed
• Resource-aware: Configure CPU/RAM allocation via wizard

Configuration

All settings are stored in .config/:

File	Content
.config/settings.env	All configuration
.config/repos.txt	Selected repositories
.config/token	GitHub token (gitignored)
.config/reviewed.tsv	Review tracking cache

Contributing

Contributions welcome! This is a cooperative project.

1. Fork on git.pop.coop
2. Create a feature branch
3. Submit a merge request

License

AGPL-3.0 — See LICENSE

Credits

• POP Coop — Initial development
• OCA — Odoo Community Association
• Runboat by Stéphane Bidoul (@sbidoul)
• oca-ci images by OCA
• Ollama for local AI inference