AI / RETRIEVAL-AUGMENTED GENERATION
Documents in.
Evidence out.
RAG Gateway connects familiar AI clients to a shared knowledge base, with local answer generation and managed retrieval.
One gateway, shared retrieval.
A local model cannot answer from documents it has never seen. The Gateway handles ingestion, tenant-filtered retrieval, and evidence assembly so clients can work from the same sources.
What the project brings together
- FastAPI ingestion and retrieval APIs.
- MongoDB Atlas vector search and optional hybrid retrieval.
- Ollama for configured local generation.
- A Playground and retrieval evaluation workbench.
- Source labels for tracing answers back to evidence.
REFERENCE DATA FLOW
RetrievalAI Models API
EmbeddingsOllama
Generation
Conceptual architecture from the public project documentation.
The data boundary matters.
Local generation is one part of the pipeline. Document text and questions go to the embedding endpoint; chunks, metadata, and vectors reside in Atlas. Optional reranking also sends text to a managed service. Tenant filtering does not provide document-level authorization.
This is an independent personal project, unaffiliated with and not endorsed by MongoDB, Inc.