Ollama Code Review Extension
Supercharge Your Code Reviews with Self-Hosted AI
Welcome to the Ollama Code Review Extension – your new ally in building top-notch software! This extension seamlessly integrates Ollama's powerful self-hosted language models into your Azure DevOps pipeline, transforming code reviews into an intelligent and efficient process while keeping your code secure on your own infrastructure.
Get Started Now!
Enhance your development workflow with Ollama Code Review. Start receiving intelligent and actionable insights on your code changes using your own self-hosted AI models. Install the extension today and experience the future of code reviews with complete control over your data!
Why Choose Ollama Code Review?
- Self-Hosted & Secure: Keep your code and reviews completely private on your own infrastructure. No data sent to external cloud services.
- 4-Pass Verification (v2.8): Context check → Review → Format → Verify workflow ensures accurate, hallucination-free feedback.
- Per-Pass Model Selection (v2.8): Use different models for each pass - fast models for context/format, best models for review.
- Custom Pass Prompts (v2.8): Override the default prompt for any workflow pass.
- Agentic Context: AI can request additional files (imports, interfaces, base classes) for smarter reviews.
- Simplified Diff Format: Clear REMOVED/ADDED sections prevent AI confusion about what code exists.
- Automated Code Reviews: Say goodbye to manual code inspections! Let Ollama analyze your code changes, catching bugs, performance issues, and suggesting best practices.
- AI-Powered Insights: Leverage powerful open-source language models like CodeLlama, Llama 3, DeepSeek Coder, and more to receive insightful comments on your pull requests.
- Configurable Token Limit: Adjust for models with larger context windows (8k to 128k+).
- Debug Logging: Extensive logging for troubleshooting issues.
- OpenAI-Compatible: Works with Ollama and any OpenAI-compatible API endpoint.
- Enhanced Context: The AI receives the full file content, project metadata (README, dependencies), and language-specific project files for more informed reviews.
- Custom Best Practices: Define your organization's specific coding standards and have the AI enforce them during reviews.
- Multi-Language Support: Automatic detection and context gathering for JavaScript, TypeScript, Python, C#, Java, and more.
- Faster Reviews: Reduce the time spent on code reviews. Let Ollama handle the routine, allowing your team to focus on impactful work.
- Configurable and Customizable: Tailor the extension to your needs with customizable settings. Choose from various Ollama models, define file exclusions, and more.
- Cost-Effective: No API costs or per-token charges. Run unlimited code reviews on your own hardware.
What's New in v2.8
🔄 4-Pass Review Workflow - Enhanced accuracy with format enforcement:
- Pass 1 (Context Check): AI determines if it needs additional files to review properly
- Pass 2 (Review): AI generates the code review with full context
- Pass 3 (Format): Enforces consistent Summary + Issues Found structure
- Pass 4 (Verify): AI validates its own review against the actual code, removing hallucinations
🎛️ Per-Pass Model Selection - Optimize cost and performance:
- Configure different models for each pass (e.g., fast model for context, best model for review)
- Use smaller/faster models for formatting and context checks
- Reserve your most capable model for the actual review pass
📝 Custom Pass Prompts - Full control over each workflow stage:
- Override the default prompt for any pass with
pass1_prompt, pass2_prompt, pass3_prompt, pass4_prompt
- Customize context request behavior, review criteria, format rules, and verification logic
🧠 Agentic Context Requests - Smarter AI:
- AI can request imported files, interfaces, and base classes it needs
- Smart file fetcher with search fallback if AI gives incorrect path
- Maximum 3 additional files per review to stay focused
📊 Simplified Diff Format - Clearer input for AI:
- Transforms confusing git diff (
+/- prefixes) into clear sections
=== REMOVED (old code - no longer exists) ====== ADDED (new code - review this) ===- Prevents AI from thinking deleted code still exists
🏗️ Modular Codebase - Better maintainability:
- Clean separation: types, prompts, api-client, ollama, pullrequest, repository modules
- Easier to extend and customize for your needs
What's New in v2.5
📝 Custom System Prompt - Complete control over AI behavior:
- Override the entire system prompt with your own instructions
- Perfect for security-focused reviews or custom workflows
- Use when you need specific review formats
🔢 Configurable Token Limit - Adjust for your model:
- Set
token_limit based on your model's context window
- Supports values from 8192 to 131072+
- Enables use of larger context models
🐞 Debug Logging - Troubleshoot with ease:
- Enable with
debug_logging: true
- Logs system prompts, diffs, API requests/responses
- Token counts and file processing details
🌐 OpenAI-Compatible API - Works with more endpoints:
- Supports both Ollama native and OpenAI-compatible formats
- Use with any OpenAI-compatible API endpoint
🎉 Enhanced AI Context - The AI now receives:
- Complete file content (not just diffs) for better understanding
- Project metadata: README, package.json, requirements.txt, .csproj, .sln, etc.
- Language-specific dependency information
🎯 Custom Best Practices - Define your own project-specific coding standards:
- Add team conventions
- Enforce organizational guidelines
- Ensure consistency across your codebase
🌍 Improved Language Support:
- JavaScript/TypeScript: package.json analysis
- Python: requirements.txt parsing
- C#: .csproj, .sln, and packages.config support
- Java: pom.xml detection
Prerequisites
- A running Ollama instance accessible from your build agents
- Ollama models installed (e.g.,
ollama pull gpt-oss or ollama pull qwen2.5-coder)
Getting started
Set up Ollama:
- Install Ollama on your server or local machine following the Ollama installation guide
- Pull your preferred model:
ollama pull gpt-oss (recommended) or qwen2.5-coder, deepseek-coder-v2, codellama, etc.
- Ensure Ollama is running and accessible from your Azure DevOps build agents
Install the Ollama Code Review DevOps Extension.
Add Ollama Code Review Task to Your Pipeline:
trigger:
branches:
exclude:
- '*'
pr:
branches:
include:
- '*'
jobs:
- job: CodeReview
pool:
vmImage: 'ubuntu-latest'
steps:
- task: OllamaCodeReview@2
inputs:
ollama_endpoint: 'http://your-ollama-server:11434/api/chat'
ai_model: 'gpt-oss'
bugs: true
performance: true
best_practices: true
file_extensions: 'js,ts,css,html'
file_excludes: 'file1.js,file2.py,secret.txt'
token_limit: '16384'
debug_logging: false
enableMultipass: true
# Optional: Use different models per pass for cost optimization
# pass1_model: 'qwen2.5-coder:7b' # Fast model for context check
# pass2_model: 'gpt-oss:20b' # Best model for review
# pass3_model: 'qwen2.5-coder:7b' # Fast model for formatting
# pass4_model: 'qwen2.5-coder:14b' # Medium model for verification
additional_prompts: 'Fix variable naming, Ensure consistent indentation, Review error handling approach'
custom_best_practices: |
Always use async/await instead of .then() for promises
All public methods must have JSDoc comments
Database queries must use parameterized statements
If you do not already have Build Validation configured for your branch already add Build validation to your branch policy to trigger the code review when a Pull Request is created
FAQ
Q: What agent job settings are required?
A: Ensure that "Allow scripts to access OAuth token" is enabled as part of the agent job. Follow the documentation for more details.
Q: What permissions are required for Build Administrators?
A: Build Administrators must be given "Contribute to pull requests" access. Check this Stack Overflow answer for guidance on setting up permissions.
Bug Reports
If you find a bug or unexpected behavior, please open a bug report.
Feature Requests
If you have ideas for new features or enhancements, please submit a feature request.
Learn More
Visit our GitHub repository for additional documentation, updates, and support.
Securing Your Ollama API with nginx
If you want to expose your Ollama API over the internet or add authentication, you can use nginx as a reverse proxy with Bearer token authentication:
server {
listen 443 ssl http2;
server_name ollama.example.com;
# SSL certificates
ssl_certificate /etc/letsencrypt/live/ollama.example.com/fullchain.pem;
ssl_certificate_key /etc/letsencrypt/live/ollama.example.com/privkey.pem;
# -------------------------------
# PROXY TO OLLAMA
# -------------------------------
location / {
proxy_set_header Authorization $http_authorization;
# Validate Authorization header
set $expected "Bearer YOUR_SECRET_TOKEN_HERE";
if ($http_authorization != $expected) {
return 401;
}
proxy_pass http://127.0.0.1:11434;
proxy_http_version 1.1;
proxy_set_header Connection "";
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
proxy_buffering off;
}
# Optional: return proper WWW-Authenticate header
error_page 401 = @unauth;
location @unauth {
add_header WWW-Authenticate "Bearer realm=\"Ollama API\"" always;
return 401 "Unauthorized";
}
}
Important: Replace ollama.example.com, SSL certificate paths, and YOUR_SECRET_TOKEN_HERE with your actual values.
Then use the Bearer Token field in the extension configuration to authenticate:
- task: OllamaCodeReview@2
inputs:
ollama_endpoint: 'https://ollama.example.com/api/chat'
ai_model: 'gpt-oss'
bearer_token: '$(OllamaApiToken)' # Store as pipeline variable
Supported Ollama Models
This extension works with any Ollama model, but these are particularly well-suited for code reviews:
- gpt-oss - Excellent for code review with strong reasoning (Recommended)
- qwen2.5-coder - Advanced code analysis and understanding
- deepseek-coder-v2 - Latest version optimized for code understanding
- codellama - Meta's specialized code model (stable)
- llama3.3 / llama3.2 - Latest Llama models with improved reasoning
- mistral-large / mixtral - Efficient general-purpose models
- codegemma - Google's code-focused model
Run ollama list on your Ollama server to see all available models.
