GraphEngine

"Why was this built this way?" Every developer asks this. GraphEngine answers it.

Point GraphEngine at any GitHub repository. It reads through commits, pull requests, and discussions — and builds a queryable knowledge graph of every engineering decision ever made.

No cloning. Just answers.

🔗 Try it live →

---

What it does

Most codebases are graveyards of undocumented decisions. Why was this architecture chosen? Why did they switch libraries? Why does this weird workaround exist?

GraphEngine reconstructs the reasoning behind your codebase by mining the conversations that happened around it — and lets you query that reasoning in plain English.

# Analyze any public GitHub repo
npm run dev ingest facebook react
npm run dev normalize facebook react
npm run dev extract facebook react

# Ask questions about specific components
npm run dev why facebook react fiber
npm run dev why facebook react classComponent
npm run dev why facebook react concurrentMode

---

Demo

Query	Answer
why was fiber introduced?	Extracted from 3 PRs, 12 commits — with citations
what changed in v18?	Timeline of decisions leading to React 18
why is reconciliation designed this way?	Grounded answer from actual PR discussions

---

How it works

GitHub Repo URL
      ↓
  Pull Requests + Commits + Issues (via GitHub API — no cloning needed)
      ↓
  Normalization → Decision Extraction (LLM-powered)
      ↓
  Knowledge Graph (relationships, timelines, decay modeling)
      ↓
  Why-Query Engine → Natural language answers with citations

No raw source code is ever downloaded. GraphEngine works entirely over the GitHub API using just a token.

---

Architecture

Layer	What it does
Ingestion	Captures GitHub events as immutable records
Normalization	Unified event format across PRs, commits, issues
Decision Extraction	LLM identifies decisions from raw discussions
Knowledge Graph	Maps relationships between decisions and outcomes
Decision Timeline	Tracks how decisions evolved over time
Knowledge Freshness	Decay + reinforcement modeling for stale decisions
Why-Query Engine	Natural language queries over the decision graph
Explainability	Every answer is grounded with source citations

---

Getting Started

Web interface (no install needed): 👉 https://why-graph-node.onrender.com

CLI:

# Clone and install
git clone https://github.com/rivv0/WhyGraph.git
cd WhyGraph
npm install

# Add your GitHub token and LLM Provider Key (Groq or OpenAI)
echo "GITHUB_TOKEN=your_token_here" > .env
echo "GROQ_API_KEY=your_key_here" >> .env

# Run
npm run dev              # Show all CLI commands
npm run web              # Start Web interface
npm run dev why <owner> <repo> <comp>  # Query from terminal

---

Why this is hard

• GitHub discussions are unstructured and noisy — extracting actual decisions requires understanding context, not just keywords
• Decisions aren't atomic events — they evolve across dozens of PRs over months
• Answers need to be grounded — hallucinated explanations are worse than no explanation
• Knowledge goes stale — a decision made in 2019 may have been reversed in 2022

GraphEngine handles all of this.

---

Tech Stack

Node.js SQLite JavaScript GitHub API LLM (decision extraction) Vector embeddings Canvas API

---

Built by

rivva <3

GraphEngine is what happens when you treat a GitHub repository not as code — but as a record of human decisions.