# Jarvis: A Multi-Agent AI Code Review System to Automate PR Review Processes

> Jarvis is a code review system based on the Agentic AI architecture. It collaborates through four professional agents—Planner, Coder, Reviewer, and Executor—to implement a complete automated process for PR analysis, issue identification, automatic repair, test generation, and validation.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-06-04T06:44:59.000Z
- 最近活动: 2026-06-04T06:58:57.382Z
- 热度: 141.8
- 关键词: 代码审查, Agentic AI, 多代理系统, GitHub, 自动化测试, RAG, Go, DevOps
- 页面链接: https://www.zingnex.cn/en/forum/thread/jarvis-ai-pr
- Canonical: https://www.zingnex.cn/forum/thread/jarvis-ai-pr
- Markdown 来源: floors_fallback

---

## Jarvis: A Multi-Agent AI Code Review System to Automate PR Review Processes

Jarvis is a multi-agent code review system based on the Agentic AI architecture, designed to automate the GitHub PR review process. It collaborates through four professional agents—Planner, Coder, Reviewer, and Executor—to complete the full process of PR analysis, issue identification, automatic repair, test generation, and validation. The project is maintained by whyankush07 and open-sourced on GitHub (link: https://github.com/whyankush07/ci-jarvis) under the MIT license. Its tech stack includes Go, Next.js, PostgreSQL, Redis, and OpenAI/Gemini API, among others. The core goal is to address pain points in traditional code reviews such as repetitive work, inconsistent quality, and long feedback cycles, while retaining the final decision-making power of human developers.

## Pain Points of Traditional Code Reviews

Traditional code review, as a key link in software quality assurance, faces many challenges:
1. **Repetitive work**: Reviewers need to repeatedly check fixed patterns like missing semicolons and unused variables;
2. **Omission of boundary conditions**: Humans easily overlook edge cases and exception handling;
3. **Inconsistent quality**: Different reviewers have varying levels of strictness and focus areas;
4. **Long feedback cycle**: It may take hours or even days from PR submission to receiving feedback;
5. **Lack of project context**: Reviewers may not understand the project's specific architecture and conventions.

Jarvis systematically solves these problems through its multi-agent collaboration architecture.

## Jarvis's Agentic AI Architecture and Workflow

Jarvis adopts the Agentic AI architecture, consisting of four professional agents:
- **Planner Agent**: Understands PR context, decomposes tasks, and formulates execution strategies;
- **Coder Agent**: Generates repair code, provides refactoring suggestions, and completes feature implementation;
- **Reviewer Agent**: Checks code standards, identifies defects, and enforces coding norms;
- **Executor Agent**: Runs tests, performs static analysis, verifies code compilability, and generates reports.

The workflow is: GitHub PR trigger → GitHub Webhook → Redis task queue → Go orchestrator coordinates agents → Executor validation → Human approval → Update PR. The workflow features include asynchronous processing, parallel execution, human approval, and traceability.

## Detailed Explanation of Core Features and Tech Stack

### Core Features
- **Agent-based code review**: Multi-agent collaboration for comprehensive code analysis;
- **Repository-aware intelligence**: Uses RAG technology to understand project-specific code patterns (code embedding, similarity search, context injection);
- **Automated test generation**: Automatically generates unit/integration tests for modified code;
- **Human approval workflow**: Developers can selectively accept/reject modifications or request re-generation;
- **Review history**: Stores complete review sessions and decisions, supporting auditing and learning;
- **Sandbox execution**: Uses Docker containers for isolated validation of code security;
- **Extensible architecture**: Supports adding new agents and tool integrations.

### Tech Stack Details
- **Backend**: Go (concurrency processing), Fiber (web framework), PostgreSQL (persistence), Redis (task queue);
- **Frontend**: Next.js 15, TypeScript, Tailwind CSS;
- **AI**: OpenAI/Gemini API, vector embedding, RAG pipeline;
- **Infrastructure**: Docker, Docker Compose, GitHub Webhooks.

## Project Value and Significance of Jarvis

The project value of Jarvis is reflected in:
1. **Efficiency improvement**: Automates repetitive review work, freeing up developers' creative time;
2. **Quality assurance**: Systematic processes reduce human omissions and improve review consistency;
3. **Knowledge inheritance**: RAG technology captures project-specific patterns and conventions, facilitating team knowledge precipitation;
4. **Learning platform**: Provides practical opportunities for technologies like Agentic AI, multi-agent orchestration, and RAG;
5. **Open-source collaboration**: Community-driven continuous improvement, promoting the development of AI-assisted development tools.

It represents a practical application of AI in the field of software engineering, providing a reference Agentic AI architecture case for AI-assisted development tools.

## Learning Value and Contribution Opportunities

### Learning Value
Jarvis provides developers with a practical platform to learn modern technologies, including:
- Agentic AI system design and multi-agent collaboration;
- RAG technology implementation and prompt engineering;
- Go backend development and distributed system design;
- Docker sandbox and containerized deployment.

### Contribution Opportunities
The project welcomes contributors at all levels:
- **Documentation**: Improve user guides and API documentation;
- **Development**: Frontend/backend feature implementation, new agent design;
- **AI optimization**: Prompt template improvement, agent behavior optimization;
- **Testing and DevOps**: Increase test coverage, improve deployment processes.

### Development Roadmap
Divided into six phases: Basic architecture → Core agents → Execution and retries → RAG integration → GitHub integration → Advanced features (observability, performance optimization, etc.).