CodeLance - GitHub Repository Intelligence

Agentic RAG·Hybrid Retrieval·Architecture-Aware AI

CodeLance

Repository Intelligence for Modern Developers

Get an instant AI-generated summary of any repository the moment it is indexed.

Ask architecture questions — why Postgres over MongoDB, why BullMQ over direct calls.

Trace any feature end-to-end across files with exact filename and line number citations.

Onboard into any codebase in minutes, not days.

Built without LangChain — engineered from scratch under real constraints: rate-limited APIs, free-tier Gemini, parallel BullMQ queues designed to stay within limits

Get Started

Parsed Languages

TypeScript

JavaScript

Python

Java

Rust

Architecture Pipeline

ZIP Stream Fetch

Single HTTP request · no rate limit risk

BullMQ — Ingestion Queue

Parse → chunk → store with null embeddings

BullMQ — Embedding Queue

3 parallel workers · respects Gemini free-tier limits

HNSW Vector Search

pgvector ANN · 3072-dim Gemini embeddings

PostgreSQL Full Text Search

websearch_to_tsquery · keyword precision

RRF Re-ranking

score = 1/(60+vRank) + 1/(60+kRank)

Agentic AI Loop

Gemini · 4 tools · up to 5 iterations

Cited Answer

filename · line number · function name

Node.jsPostgreSQLpgvectorBullMQRedisGemini APINext.jsDocker

Core Architecture

Built for Real Codebases

Engineered with hybrid retrieval, semantic search, intelligent re-ranking and grounded citations to understand repositories beyond simple code search.

Intelligent Search

Hybrid retrieval combining pgvector semantic search with PostgreSQL full-text search for precise repository understanding.

Code Analysis

AI reconstructs repository context across multiple files to explain architecture, dependencies and implementation details.

Source Citations

Every answer is grounded with exact filenames, functions and line numbers so every explanation remains verifiable.

Async Pipeline

BullMQ-powered ingestion pipeline processes repositories reliably while respecting external API limits.

Agentic Reasoning

Multi-step reasoning plans retrieval, ranking and synthesis before generating architecture-aware responses.

Language Agnostic

Supports all major programming languages through a language-agnostic chunking strategy.

Hybrid Retrieval

pgvector + PostgreSQL

Agentic Workflow

Multi-step reasoning

Grounded AI

Exact source citations

Universal Support

All major languages

See it in action

A real question, asked against a real codebase — no scripted demo

CodeLance.app — SHAKSHYAM23/TicketFlow

CodeLance answering a question about the TicketFlow repository, showing the agent's cited sources

Every repo is parsed into chunks first — this one, 78 files down to 253 searchable pieces.

The agent decides which tools to call based on the question, not a fixed retrieval step.

Answers stay grounded in the actual code — no filler, no guessing when something isn't there.

Agentic RAG • Semantic Search • Repository Intelligence

Ask Better Questions.

CodeLance understands your repository beyond simple code search. Explore architecture, implementation details, design decisions, and engineering trade-offs through natural conversation.

💬 How does authentication work?

💬 Why was PostgreSQL chosen instead of MongoDB?

💬 What happens if we replace JWT with Sessions?

💬 Explain the complete request flow.

💬 How would you improve this architecture?