Nova: An AI‑Powered Code Review System Integrated with CI/CD Pipelines

Inspired by various AI coding assistants and AI Code Review solutions, the Nova project aims to provide an AI‑driven code review tool with the following characteristics:

1) Tight integration with existing CI processes, specifically GitLab‑Runner + SonarQube.

2) Full‑project context understanding, supporting multiple programming languages and delivering low‑cost implementation.

3) High extensibility, allowing any model or AI‑Agent platform and any context‑search method such as RAG or AST search.

Nova (named after the astronomical term for a new star) symbolizes fresh, high‑quality code review capabilities.

Process Overview

The system consists of two main components: “Retrieve Code Context” and “Code Review”. Both components can be replaced by alternative implementations (e.g., using RAG instead of AST search, or a different LLM instead of Dify).

Deployment Architecture

Deployment is straightforward: after building a Nova Docker image, only a single job needs to be added to .gitlab-ci.yml . The architecture diagram (omitted) shows integration with GitLab CI, SonarQube, and optional Dify configuration.

Code Safety

The tool supports filtering of review files and context files by path, as well as content‑sensitive‑word filtering.

Feature Demonstration

Key features demonstrated include:

Support for Dify configuration.

Seamless integration into the GitLab‑CI pipeline.

Connection to SonarQube for quality gate enforcement.

Understanding of project context, illustrated by successfully locating the definition of gschatsdk.NewSdk() .

Challenges and Outlook

1) Achieving more precise code review suggestions to meet quality thresholds. 2) Providing customized audit prompts for different programming languages and projects.