LinkingMemGraph-native RAG Engine

Rust + Python engine unifying vector search, graph traversal, and LLM reasoning in a single pipeline — Query → Embed → HNSW → BFS → Score → LLM answer.

System Design

API Reference

View on GitHub

Hybrid Vector + Graph

CSR graph and HNSW vector index share one pipeline — no separate vector database, no Neo4j + Qdrant split.

Multimodal

Text and image nodes share the same vector space. Query by text, by image URL, or both — caption or CLIP embedding backends.

Multi-hop Reasoning

The LLM can request additional graph expansion iteratively via /query/multihop before producing a final answer.

Any LLM Provider

OpenAI-compatible client works with OpenAI, Ollama, Gemini compat, Groq, LM Studio, vLLM — just point OPENAI_BASE_URL.

Production-minded

Delta store + WAL crash recovery, hot-swap graph merge, Prometheus metrics, per-key rate limiting, distributed ingest.

Bring Your Own Plugin

The engine speaks a plain HTTP/Unix-socket plugin interface — swap in a custom embed/extract/generate server in any language.