Automate Code Reviews with AI: n8n + Ollama Workflow

Every pull request deserves a review, but not every team has the bandwidth. This tutorial shows you how to build an AI-powered code review pipeline using n8n and a local LLM (via Ollama) — completely self-hosted, no API costs, no data leaving your network.

What This Workflow Does

1. Receives a webhook when a PR is opened on GitHub/GitLab
2. Fetches the diff (code changes) from the PR
3. Sends the diff to Ollama (local LLM) for analysis
4. Formats the review into a structured report
5. Posts the review to Discord (or Slack, email — your choice)
6. Responds to the webhook confirming the review was posted

Total: 6 nodes, ~5 minutes to set up, zero recurring cost.

Architecture

GitHub PR Webhook → n8n → Fetch Diff → Ollama (local) → Format Review → Discord

The entire pipeline runs locally. Your code never touches a third-party API. This is critical for companies handling proprietary code or operating under GDPR constraints.

Prerequisites

• n8n installed (self-hosted or cloud)
• Ollama running with a code-capable model: ollama pull qwen2.5-coder:7b
• GitHub/GitLab webhook access to your repository
• Discord webhook URL (or Slack incoming webhook)

Step-by-Step Setup

1. Configure the Webhook Trigger

Create a new n8n workflow and add a Webhook node:

• HTTP Method: POST
• Path: /pr-review
• This generates a URL like https://your-n8n.com/webhook/pr-review

Register this URL in your GitHub repo: Settings → Webhooks → Add webhook → Payload URL → Select “Pull requests” events.

2. Fetch the Diff

Add a Code node to extract the diff URL from the webhook payload:

const prUrl = $input.first().json.pull_request.diff_url;
const title = $input.first().json.pull_request.title;
const author = $input.first().json.pull_request.user.login;
return [{ json: { prUrl, title, author, diffUrl: prUrl } }];

3. Send to Ollama for AI Review

Add an HTTP Request node:

• Method: POST
• URL: http://localhost:11434/api/generate
• Body (JSON):

{
  "model": "qwen2.5-coder:7b",
  "prompt": "Review this code diff. Focus on: bugs, security issues, performance problems, and code style. Be specific and reference line numbers.\n\nDiff:\n{{ $json.diff }}",
  "stream": false
}

This sends the diff to your local Ollama instance. The model runs on your hardware — no tokens billed, no data uploaded.

4. Format the Review

Add another Code node to structure the output:

const review = $input.first().json.response;
const title = $('Fetch Diff').first().json.title;
const author = $('Fetch Diff').first().json.author;

return [{
  json: {
    content: `**AI Code Review** 🤖\n**PR:** ${title}\n**Author:** ${author}\n\n${review}\n\n_Powered by Vorlux AI + Ollama (local inference)_`
  }
}];

5. Post to Discord

Add an HTTP Request node:

• Method: POST
• URL: Your Discord webhook URL
• Body: { "content": "{{ $json.content }}" }

6. Respond to Webhook

Add a Respond to Webhook node returning { "status": "reviewed" }.

Download the Workflow

This exact workflow is available as a ready-to-import JSON file:

Download ai_code_review.json →

Import it in n8n: Settings → Import Workflow → Upload File.

Why Local AI for Code Review?

Factor	Cloud API (GPT-4)	Local (Ollama + Qwen)
Cost	~$0.03-0.10 per review	EUR 0
Privacy	Code sent to OpenAI servers	Code stays on your machine
Speed	2-5 seconds (network)	1-3 seconds (local)
GDPR	Requires DPA with OpenAI	Fully compliant (local)
Availability	Depends on API uptime	Always available

flowchart LR
    GH["GitHub PR"] --> WH["n8n Webhook"]
    WH --> DIFF["Fetch Diff"]
    DIFF --> AI["Ollama Review"]
    AI --> FMT["Format Output"]
    FMT --> DC["Discord Alert"]
    
    style GH fill:#1E293B,color:#FAFAFA
    style AI fill:#059669,color:#FAFAFA
    style DC fill:#F5A623,color:#0B1628

Choosing the Right Model for Code Review

Not all models are equal for code review tasks. Based on our testing across hundreds of PRs:

Model	Strengths	Best For	Memory
Qwen 2.5 Coder 7B	Instruction-following, multi-language	General code review	~4.5GB
DeepSeek R1 14B	Chain-of-thought reasoning	Complex logic bugs	~10GB
Llama 3.3 70B	Deep analysis, architectural feedback	Architecture reviews	~40GB
Phi-4 14B	Fast inference, concise output	Quick PR checks	~9GB

For most teams, Qwen 2.5 Coder 7B provides the best balance of quality and speed. If your Mac has 32GB+ memory, consider running DeepSeek R1 14B for its superior reasoning on complex diffs.

Related reading

• Google Gemma 3: The First Multimodal Open Model That Fits on a Mac Mini
• Google Gemma 4: The Open Model Family That Changed Our Entire Stack
• Build a Local RAG Pipeline with n8n and Ollama: Query Your Company Documents with AI

Next Steps

• Replace Discord with Slack or email notification
• Add a GitHub comment node to post the review directly on the PR
• Use structured outputs (Ollama JSON schema) for machine-parseable reviews
• Chain with a test runner workflow for full CI/CD integration
• Explore n8n’s Model Context Protocol (MCP) support for connecting agents to external tools

---

Related resources

• 230 Workflows library — browse and download all workflows
• AI Models catalog — find the best model for code review
• Software stack — tools we use and recommend
• Contact us — need help setting this up?