Skip to content
OdyssAI

OdyRAG — the knowledge-graph layer

Clients query knowledge. OdyRAG is where that knowledge is built and tended — the engine’s knowledge-graph layer.

OdyRAG is the engine’s knowledge-graph layer: the management dashboard for LightRAG knowledge graphs, served by OdyssAI-X at /odyrag. Clients can query the graphs; OdyRAG is where you inspect them, ingest into them, and keep them healthy.

Powered by LightRAG.

Two organs, one dashboard

Section titled “Two organs, one dashboard”

The stack keeps knowledge in two distinct organs, and OdyRAG is the lens on both:

OrganWhat it isUsed for
Memory (LightRAG)Small, living graphs — one per shard instance.”What do we already know?” — facts and their relations, queried per request.
Corpus (Qdrant)A large document store, embedded for retrieval.”Find me the passage” — bulk document RAG.

They answer different questions. OdyRAG manages the LightRAG side and ingests from the corpus into it when you want documents folded into the living graph.

What you do in OdyRAG

Section titled “What you do in OdyRAG”
  • Shard instances. See the running LightRAG instances and their state. Each is a graph you can inspect.
  • Batch inject from Qdrant. Pull a slice of the document corpus into a graph, so its facts become part of what a query against that graph returns. This is the bridge from “a pile of documents” to “a queryable graph of what they say”.
  • Query the knowledge graph. Run a retrieval directly against a graph to see exactly what comes back for a question — the operator’s view of what a client gets.
  • Watch ingestion. Long ingests stream progress; logs are tailable on the host (~/lightrag-alpha-ingest/logs/).

Build side, query side

Section titled “Build side, query side”

The same data seen from two sides:

clients   → query             (read: retrieve from a graph, per request)
OdyssAI-X → OdyRAG dashboard  (build: inspect graphs, ingest the corpus, query directly)

Clients query the graphs; an operator manages the underlying graphs — and feeds documents into them — in OdyRAG.

A note on care

Section titled “A note on care”

The LightRAG stores are powerful but the on-disk JSON stores are fragile: a flush is a full, non-atomic rewrite. Treat ingestion as the moment to be careful — let a batch finish, watch the logs, and keep a backup of a graph before a large re-ingest. (The stack ships a quarantine + re-embed path from the graphml for recovery.)

Section titled “Read next”