FORGE — AI Engineering Co-Pilot for Entire Organizations

Connect your codebase. Every engineer gets a senior engineer on-call 24/7 who knows every line.

FORGE indexes your entire GitHub/GitLab codebase using code-specific embeddings and makes it queryable in plain English. Voice-controlled, security-hardened, and context-aware across 1M tokens.

Features

• Codebase Semantic Search — nv-embedcode-7b-v1 code-specific embeddings (not text search)
• 1M Context Architecture Understanding — Nemotron-Ultra holds your entire codebase in memory
• Voice-Controlled Coding — Talk to your codebase with Parakeet ASR
• Automatic Test Generation — Qwen3-Coder-480B writes tests for any function
• Security Vulnerability Detection — Scans entire repo for OWASP vulnerabilities
• New Hire Onboarding — 10-minute codebase onboarding guides generated instantly
• Legacy Code Explainer — Understands and explains any code, any language
• PII/Secret Scrubbing — GLiNER strips API keys before any AI processing
• USD/3D Code Support — usdcode for metaverse/simulation teams

Tech Stack

• Frontend: React 18 + TypeScript + Tailwind CSS + Monaco Editor
• Backend: Node.js + Express + TypeScript
• Code Indexing: Tree-sitter for AST parsing
• GitHub Integration: Octokit
• AI: 13 NVIDIA NIM endpoints

Quick Start

git clone https://github.com/HayreBuilds/forge.git
cd forge
cp .env.example .env
# Add NVIDIA_API_KEY and GITHUB_TOKEN to .env

cd backend && npm install
cd ../frontend && npm install

cd backend && npm run dev
cd ../frontend && npm run dev

Environment Variables

NVIDIA_API_KEY=your_nvidia_api_key_here
GITHUB_TOKEN=your_github_personal_access_token
PORT=3001
CORS_ORIGIN=http://localhost:5173

API Endpoints

Method	Route	Description
POST	/api/repo/connect	Connect a GitHub/GitLab repository
POST	/api/repo/index	Index repository with code embeddings
POST	/api/query/search	Semantic code search
POST	/api/query/ask	Natural language codebase Q&A
POST	/api/generate/tests	Auto-generate tests for a function
POST	/api/security/scan	Security vulnerability scan
POST	/api/onboard/guide	Generate new hire onboarding guide
POST	/api/voice/query	Voice-to-code query

NVIDIA Endpoints Used

• qwen3-coder-480b-a35b-instruct — Primary code generation
• nvidia/nv-embedcode-7b-v1 — Code-specific semantic embeddings
• nvidia/nemotron-3-ultra-550b-a55b — Full codebase architectural reasoning
• deepseek-ai/deepseek-v4-flash — Fast code queries
• nemoretriever-parse — Internal documentation parsing
• llama-nemotron-embed-1b-v2 — Documentation embeddings
• nvidia/rerank-qa-mistral-4b — Result relevance reranking
• parakeet-ctc-1.1b-asr — Voice coding interface
• magpie-tts-zeroshot — Audio code explanations
• nvidia/usdcode — OpenUSD/3D code generation
• nvidia/llama-3.1-nemoguard-8b-topic-control — Engineering topic enforcement
• nvidia/nemojail-jailbreak-detect — Malicious code injection protection
• nvidia/gliner-pii — API key/secret scrubbing

License

MIT