repomind - An open-source, AI-powered application using Agent

项目说明

RepoMind

<p align="center"><strong>Stop reading code. Start talking to it.</strong></p>

<p align="center">
<a href=" src=" alt="GitHub stars" /></a>
<a href=" src=" alt="MIT License" /></a>
<a href=" src=" alt="Next.js" /></a>
<a href=" src=" alt="Gemini" /></a>
</p>

<p align="center">
<a href=" RepoMind Live</strong></a>
·
<a href=" React Analyzed</strong></a>
·
<a href="
</p>

![RepoMind Demo]()

RepoMind is an open-source, AI-powered platform for understanding public GitHub repositories and de profiles with Agentic CAG. It combines deep code reasoning, architecture , 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 chunks.

understanding: generates architecture maps and flow diagrams for faster onboarding.

Security scanning built in: quick/deep scans with 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

Trending Repositories: Disco and highly-acti repositories directly from the landing page.

Real-Time Global Search: Instant, optimized repository search with auto-complete suggestions for seamless na.

!RepoMind Landing Page

Repo Profile Intelligence

!Repo Profile Example

Core Capabilities

!RepoMind 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.

!CAG vs RAG

!Architecture Example

Architecture Flowcharts

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

1) Query + Context Pipeline

2) Caching + Retrie

3) Security Scan + Flow

Security Scanning

RepoMind includes a scanning flow for application and dependency risks.

Quick scan for fast triage

Deep scan for broader co

Real-time dependency checks: Integrated with OSV.dev API for latest CVE data

reporting pipeline

Actionable fixes inside repo chat flow

!Security Report

Dashboard Experience

!Dashboard O

!Recent Scans

!My Repositories

!Starred Repositories

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

Open

Model En

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

GEMINI_FILE_SELECTOR_MODEL (default: gemini-3.5-flash-lite)

GEMINI_LITE_MODEL (default: gemini-3.5-flash-lite)

GEMINI_THINKING_MODEL (default: gemini-3.8-flash)

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 a 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 a 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 consecuti 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

De Commands

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

Impro tooling used by public users

Ship features quickly with direct product impact

High-Impact Areas

Query pipeline and reasoning quality: improve file selection, prompting, and response quality

Security scanner and : impro accuracy and reduction of false positives

Dashboard and De UX: impro, 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 notes

Pull Request Expectations

Keep scope clear and re

Add or update tests when behavior changes

Update docs when introducing user- changes

Include before/after screenshots for UI work

Changelog

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

Tech Stack

Next.js

React

Gemini

OSV (Open Source )

Prisma

KV

License

Licensed under the MIT License. See LICENSE.

---

<p align="center">
Built by <a href=" ·
<a href=" on GitHub</a> ·
<a href=" Issue</a>
</p>

项目信息

开发语言：TypeScript

Star 数：286

技术要点

该项目属于安全软件方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕An open-source, AI-powered application using Agentic CAG to chat with any public GitHub repository or developer profile, offering deep code analysis, visual architecture maps and security audits展开，提供了可运行的实现，对理解相关安全机制、工具原理与攻防技术具有较高的研究价值。

从技术实现角度看，此类项目的核心在于对底层原理的深入理解与工程化落地。开发者在研究过程中可以结合官方文档与社区讨论，梳理项目的初始化流程、核心模块职责与数据流转，逐步掌握其中的技术细节与设计思路。

使用与学习建议

1. 下载源码压缩包，解压到本地环境。

2. 根据项目类型准备对应的运行环境与依赖。

3. 按照 README 中的步骤完成配置与部署。

4. 在隔离的测试环境中运行验证，避免影响生产系统。

免责声明

本资源来源于开源社区，仅供安全研究、技术学习与合法授权范围内的攻防演练使用，请勿用于任何非法或侵害他人权益的用途。使用本项目产生的一切后果由使用者自行承担。