Summary

The first interface issue is already in place, but AgentRAG also needs one small reference implementation that proves the intended pipeline can actually run. An in-memory indexing and retrieval flow would be the best minimal slice.

Why this matters

Without one concrete implementation, the architecture risks staying abstract. A tiny reference pipeline would make the project easier to reason about and give future provider integrations a baseline.

Proposed scope

• add one in-memory implementation of the core indexing and retrieval flow
• support a small document set and a simple retrieval query path
• keep the implementation provider-agnostic and intentionally lightweight
• document how the reference pipeline maps onto the longer-term architecture

Acceptance criteria

• the repo can run a basic indexing and retrieval example end to end
• the example is simple enough to serve as a baseline for later vector-store work
• README or docs include the first concrete usage example

Notes

This issue is intentionally smaller than full RAG support. It is about proving the pipeline shape with a minimal working slice.