Offline mode: local embedding-based RAG with lightweight models (Gemma 4 uncensored)

[tashifkhan]

Goal

Allow core retrieval and response workflows to run fully offline using local embeddings and a lightweight local LLM.

Proposed Scope

• Implement a local vector store pipeline, offline indexing/retrieval, and a local inference path with model/runtime configuration.

Acceptance Criteria

• Users can toggle offline mode and complete end-to-end RAG queries without network access.

Target Date

• 22 Aug 2026 (IST)