StorageGitHub ↗
Browser-native · No server · Keys stay local

Ask any GitHub
repo anything.

Index any repo in your browser. Chat with its code using your own API key. Embeddings, retrieval, and storage — all on-device.

Star on GitHub★ —

Try an example

How it works

Under the Hood

01

Paste a GitHub URL

Any public repo. Private repos with a token.

02

Index in your browser

AST chunking + WebGPU embeddings. No server. Binary quantization shrinks vectors 32x. For ex: mlc-ai/web-llm goes from 2.1 MB to 67 KB.

03

Ask questions

Chat with your LLM of choice. Results cite real code.

01Ingestion + Indexing
Input</>

URL Trigger

Route params resolved, indexing starts

Ingest</>

GitHub Fetch

Repo tree + blobs at browser

Parse</>

AST Chunker

tree-sitter WASM semantic splits

Compute</>

Embedding Pipeline

transformers.js WebGPU vectors

Optimize</>

Binary Quantization

Compressed for fast Hamming search

Store</>

Entity-DB (IndexedDB)

Vectors + metadata persisted locally

index ready at query time
02Query-time Retrieval (CodeRAG-style)
Input</>

User Question

Natural language query to the repo

Expand</>

Query Expansion

Multi-query generation (CodeRAG-style)

SearchIndexedDB</>

Q1 • Original

Hybrid search. Hamming + regex

SearchIndexedDB</>

Q2 • Code-Style

Hybrid search. Hamming + regex

Merge</>

RRF Fusion

Reciprocal Rank Fusion over both paths

Rank</>

Preference Rerank

Matryoshka reranker on candidates

Select</>

Top-k Chunks

Best chunks assembled for context

Inference</>

WebLLM Worker

Qwen2-0.5B in a dedicated web worker

Output</>

Chat UI + CoVe

Streamed answer + verification loop