CodeJury
Terminal-first, knowledge-grounded multi-agent software delivery pipeline: scope requirements, implement changes, run tests, and gate pull requests with deterministic QA and ensemble code review.
What is it?
What it is
A terminal-first, multi-stage agentic coding platform that uses an ensemble of independent LLM judges (a 'jury') to review code changes, rather than a single LLM judge.
Why it exists
To solve the 'ingle point of judgement' failure in agentic coding, where a single LLM judge often shares the same blind spots as the authoring LLM, leading to unreliable code reviews.
Who should use it
Developers working on medium to large codebases who want to reduce the localization tax and improve code quality through a diverse LLM jury review system. Teams seeking to automate code review with multiple independent models to mitigate single-model biases. Users comfortable with the terminal who want to integrate AI into their workflow without vendor lock-in.
Who should avoid it
Users primarily working on trivial or very small changes where the tool's overhead outweighs benefits. Those unwilling to use the terminal interface (despite an optional web UI). Individuals requiring a completely zero-configuration tool (though setup is minimized). Users who prefer a single-model review system and are not interested in configuring multiple stages.
How it works
A quick walkthrough in plain English
How CodeJury works
Step 1 of 3
You interact with it
Open CodeJury, send a request, or connect it to your stack.
Features
Advantages
- Reduces bias through ensemble judgment
- Flexible customization of models per pipeline stage
- Efficient codebase understanding via persistent graph
- Improved search accuracy with hybrid retrieval
- Context retention across deliveries for better scoping
- Human oversight ensures accountability
- Structured implementation planning reduces errors
- Seamless integration with existing coding CLIs
- Transparent cost tracking for budget management
- Robust fallback mechanisms maintain functionality
- Real-time terminal experience with rich features
Disadvantages
- Requires ripgrep or codebase-memory-mcp for optimal performance
- Setup complexity with multiple dependencies
- Potential cost increase with premium model usage
- Learning curve for terminal-centric workflow
- Limited web UI functionality compared to desktop tools
- Dependency on specific LLM providers for some stages
Installation
native
pip install -e ".[semantic,treesitter]" then codejury
FAQ
Do I need to provide my own API keys to use CodeJury?
Not necessarily. CodeJury can run using your existing subscriptions to coding CLIs like Claude Code, Codex, Cursor, Aider, or Gemini CLI, which avoids per-token API billing. Alternatively, you can use any OpenAI-compatible API key; the default settings target Groq's free tier, allowing the entire pipeline to run for $0.
How does the 'Jury' review process differ from standard LLM code reviews?
Instead of relying on a single LLM judge—which can be biased or blind to its own errors—CodeJury uses an ensemble of independent, specialized judges (e.g., Correctness, Security, Architecture) running in parallel. A 'foreperson' LLM then synthesizes these diverse perspectives into a single, evidence-bound verdict.
What operating systems are supported?
CodeJury is cross-platform and supports macOS, Linux, and Windows. It is written in pure Python using libraries like Rich and Textual that ensure consistent terminal rendering across all three platforms.
How does CodeJury navigate large codebases efficiently?
CodeJury builds a persistent code graph and a local semantic search index (using fastembed and Qdrant) during the ingestion phase. This allows the agent to perform exact file:line lookups and call-graph impact analysis, reducing the 'localization tax' and token spend compared to agents that must grep or re-read files from scratch.
Can I customize the agents and the review panel?
Yes. You can use the `/model` command to assign different providers and models to each of the six pipeline stages (Knowledge, PM, Planner, Dev, QA, Review). You can also use the `/jury` command to seat, unseat, reorder, or even add custom judges with your own specific instructions or compliance checklists.
Featured in Videos
YouTube tutorials and walkthroughs for CodeJury
Alternatives
Similar projects ranked by category, topics, and text overlap.
