RepoMind

Stop reading code. Start talking to it.

Try RepoMind Live · See React Analyzed · Contribute

RepoMind is an open-source, AI-powered platform for understanding public GitHub repositories and developer profiles with Agentic CAG. It combines deep code reasoning, architecture visualization, and security scanning into a fast, browser-first experience.

Why RepoMind

• Zero setup for public repos: paste a repo or profile and start immediately.
• Context-aware analysis: selects full, relevant files instead of fragmented vector chunks.
• Visual understanding: generates architecture maps and flow diagrams for faster onboarding.
• Security scanning built in: quick/deep scans with verification-focused reporting.
• Live step feedback: chat shows real-time processing stages (selecting files, reading files, preparing answer).
• Contributor-friendly codebase: clear TypeScript services, tests, and modular features.

Product Highlights

Landing Experience

Repo Profile Intelligence

Core Capabilities

Agentic CAG Architecture

RepoMind uses Agentic Context-Augmented Generation (CAG). Instead of relying only on embeddings, it prunes and loads coherent source context to answer architecture and implementation questions reliably.

Architecture Flowcharts

These diagrams provide a quick mental model of how RepoMind processes queries, optimizes performance, and validates security findings.

1) Query + Context Pipeline

graph TD
    A[User enters repo or profile query] --> B[Context-Aware Engine]
    B --> C[Repository/Profile metadata fetch + file tree]
    B --> D[File selection via GEMINI_FILE_SELECTOR_MODEL]
    D --> E[25 files for Lite or 50 files for Thinking]
    E --> F[Fetch selected files with SHA-aware cache reuse]
    F --> G[Build context with repo folder structure + selected files]
    G --> H[Answer generation via Lite or Thinking model]
    H --> I[Streaming response with text-only live steps]

Loading

2) Caching + Retrieval Strategy

graph TD
    A[Request selected files] --> B{SHA-keyed file cache hit?}
    B -->|Yes| C{Compressed entry?}
    C -->|Yes| D[Decompress]
    C -->|No| E[Return cached payload]
    D --> E
    B -->|No| F[Fetch only missing/new files from GitHub API]
    F --> G{Payload <= 2MB?}
    G -->|Yes| H[Gzip + store in KV]
    G -->|No| I[Skip cache write]
    H --> J[Return full context]
    I --> J
    J --> K[Cache query -> selected files - 24h]
    J --> L[Cache query answer - 24h]

Loading

3) Security Scan + Verification Flow

graph TD
    A[Start scan] --> B{Quick or Deep}
    B --> C[Collect scoped files + dependencies]
    C --> D[Deterministic security engine]
    D --> E{AI assist enabled?}
    E -->|Yes| F[AI validation pass]
    E -->|No| G[Verification gate]
    F --> G
    G --> H{Verified true?}
    H -->|Yes| I[Show finding in report]
    H -->|No| J[Hide as rejected/inconclusive]

Loading

Security Scanning

RepoMind includes a verification-first scanning flow for application and dependency risks.

• Quick scan for fast triage
• Deep scan for broader coverage
• Verified-first reporting pipeline
• Actionable fixes inside repo chat flow

Dashboard Experience

Getting Started

Use Hosted App

1. Open repomind.in
2. Enter owner/repo or a GitHub username
3. Ask architecture, code, or security questions

Run Locally

Prerequisites

• Node.js 18+
• npm
• GitHub token
• Gemini API key

Setup

git clone https://github.com/403errors/repomind.git
cd repomind
npm install
cp .env.example .env.local
npm run dev

Open http://localhost:3000.

Model Environment Variables

Repo chat now uses separate models for file selection and answer generation:

• GEMINI_FILE_SELECTOR_MODEL (default: gemini-3.1-flash-lite-preview)
• GEMINI_LITE_MODEL (default: gemini-3.1-flash-lite-preview)
• GEMINI_THINKING_MODEL (default: gemini-3-flash-preview)

Realtime Chat Feedback

Both repo chat and profile chat stream backend activity as live text status updates (no progress bar), including:

• Selecting files
• Reading repository context
• Preparing answer
• Tool activity (for example: fetching commits, fetching repos by age, Google search usage)

Repo Chat Commit Tooling

• fetch_recent_commits is available in both Lite and Thinking modes.
• When commit context is needed, repo chat fetches up to latest 10 commits for the current repository.
• Assistant messages can include:
  • Commits checked: just now or cached N min ago
  • Tools used: ...

Profile Chat Tooling

• Profile chat supports anonymous and logged-in usage.
• Profile context preloads latest 20 commits across repos, cached for 15 minutes.
• If a repo is explicitly asked for, tooling can fetch latest 10 commits for that repo.
• fetch_repos_by_age supports:
  • oldest (10 repos)
  • newest (10 repos)
  • journey (up to 20 repos, evenly spaced over repo age timeline)
• Cross-repo context is available via a chat-input toggle for logged-in users.

Auth Tiers, Limits, and Budgets

• Anonymous users:
  • Can use Lite mode in repo/profile chat
  • Cannot use Thinking mode
  • Cannot enable cross-repo profile context
  • Tool budget: 10/day per scope (repo, profile)
  • File-cache write budget: 10MB/day
  • File cache TTL: 30 minutes
  • Max cacheable file size: 128KB, except files used in two consecutive anonymous queries (can be cached up to 2MB)
• Logged-in users:
  • Can use Lite + Thinking
  • Can enable cross-repo profile context
  • Tool budget: 30/day per scope (repo, profile)
  • File-cache write budget: 20MB/day
  • File cache TTL: 1 hour

When anonymous usage limits are exceeded, UI prompts login with context-specific benefits.
When authenticated usage limits are exceeded, chat returns a contact message for extended limits.

Cache Isolation Rules

• Public repo file cache is shared by repository namespace.
• Private repo file cache is actor-scoped (non-shared).
• Query-to-selected-files cache TTL: 24h
• Query answer cache TTL: 24h

Developer Commands

npm run dev            # local dev server
npm run lint           # lint checks
npm run test           # test suite
npm run test:security  # security-focused tests
npm run build          # production build

Contributing

We want RepoMind to be one of the best OSS projects for AI-assisted code intelligence. Contributions of all sizes are welcome.

Why contribute

• Work on practical agentic AI for real repositories
• Improve developer tooling used by public users
• Ship visible features quickly with direct product impact

High-Impact Areas

• Query pipeline and reasoning quality: improve file selection, prompting, and response quality
• Security scanner and verification: improve detection accuracy and reduction of false positives
• Dashboard and DevTools UX: improve discoverability, workflows, and performance
• Docs, tests, and reliability: strengthen onboarding and confidence for contributors

Contribution Workflow

1. Fork the repository
2. Create a branch: git checkout -b feature/your-change
3. Make focused changes with tests
4. Run npm run lint and npm run test
5. Open a pull request with context, screenshots, and validation notes

Pull Request Expectations

• Keep scope clear and reviewable
• Add or update tests when behavior changes
• Update docs when introducing user-visible changes
• Include before/after screenshots for UI work

Changelog

Recent major updates are tracked in CHANGELOG.md, including improvements through v1.3.6 (2026-02-23).

Tech Stack

• Next.js
• React
• Gemini
• Prisma
• Vercel KV
• Vitest

License

Licensed under the MIT License. See LICENSE.

---

Built by 403errors · Star on GitHub · Report Issue