Skip to content

RepoGraph AI RepoGraph AI

RepoGraph AI RepoGraph AI is an autonomous, multi-agent developer cockpit powered by Gemini that maps, audits, refactors, and secures codebases — with every AI decision verified by the compiler, the graph, and the static

Code Video Cerebral Valley

Core trick

Ten parallel analyzer agents emit typed Pydantic records that build a dependency graph, and every AI-proposed edit must pass AST, blast-radius, and CWE checks.

Tags

dev toolssecurity

Description

RepoGraph AI RepoGraph AI is an autonomous, multi-agent developer cockpit powered by Gemini that maps, audits, refactors, and secures codebases — with every AI decision verified by the compiler, the graph, and the static analyzer before it reaches a human. The Problem Engineering teams suffer from silent architectural drift — circular dependencies, SOLID violations, and security vulnerabilities accumulating commit by commit. Current AI coding tools generate code confidently but never verify it, reason about architecture holistically, or check their own work. What It Does Upload any repository and RepoGraph AI deploys a coordinated team of Gemini agents that: Builds a live semantic graph using 10 parallel file analyzer agents communicating through typed Pydantic schemas Audits architectural health with SOLID principles scoring and circular dependency detection Validates design intent by reading your architecture diagram as an image and comparing it to actual code — genuine multimodal reasoning Autonomously fixes violations through a tri-agent pipeline: Architect plans, Coder implements, Reviewer critiques Verifies every generated change through a 4-gate pipeline: AST correctness → blast radius safety → CWE security scan → adversarial critic review Narrates architectural history by scrubbing through git commits and rebuilding the graph at every snapshot Exposes the entire suite as MCP tools callable directly from Cursor, VSCode, and Claude Desktop
radhika-singh-10/RepoGraph-AI★ 00 forksPythonApache-2.0Created May 20, 2026 · last push Jun 1, 2026

README excerpt

# 🗺️ RepoGraph AI

**RepoGraph AI** is a next-generation repository architecture visualizer and autonomous developer cockpit powered by **Gemini 3.5 Flash**. It maps, audits, refactors, and secures your codebase by substituting static AST parses with a collaborative team of parallelized Gemini agents communicating over strict typed Pydantic contracts.

---

### Project demoed in Google I/O Hackathon Round 1(23th May 2026). 
### Team name - <a href="https://cerebralvalley.ai/e/google-io-hackathon/hackathon/gallery?project=90">Archon</a>

---

## 🏗️ Core Architecture & Agent Layout

```mermaid
graph TD
    A[Workspace Directory] --> B[Concurrent Executor]
    B -->|Parallel Workers| C1[File Analyzer Agent 1]
    B -->|Parallel Workers| C2[File Analyzer Agent 2]
    B -->|Parallel Workers| C3[File Analyzer Agent N...]
    C1 -->|FileAnalysisResult JSON| D[Graph Orchestrator Agent]
    C2 -->|FileAnalysisResult JSON| D
    C3 -->|FileAnalysisResult JSON| D
    D -->|GraphOutput Schema| E[React Flow UI nodes & edges]
```

1. **Parallel File Analyzer Agents**:
   - Spawns concurrent `gemini-3.5-flash` requests (up to 10 workers in a `ThreadPoolExecutor`) to scan individual source files.
   - Workers semantically extract definitions (classes, functions), imports, technologies, and classifications using a Pydantic `FileAnalysisResult` response schema.
2. **Graph Orchestrator Agent**:
   - Aggregates worker scans to dynamically map imports, route handlers (`defines` relationship), and

Team