Python / gRPC / GraphRAG / RAG Infrastructure

Retrieval infrastructure at scale.

A distributed vector database prototype for AI retrieval workloads, with sharded gRPC nodes, persistent storage, GraphRAG expansion, snapshots, and LlamaIndex adapters.

Try search simulation View source
Live Output

Search simulation.

This browser demo mirrors the retrieval experience: query text is scored against seeded records, then graph-linked neighbors are included as context.

Architecture

Backend pieces that matter.

The original project exposes vector upsert, search, get, health, and snapshot APIs through Python gRPC services.

01

Distributed sharding

Rendezvous hashing routes records across vector nodes while keeping routing deterministic and explainable.

02

Persistent retrieval

SQLite is the default local store, with optional RocksDB-compatible storage through rocksdict.

03

Graph-aware context

GraphRAG expansion follows linked records so query results include nearby context, not only raw similarity hits.