[docs] Update README to accurately describe Soma embedding-free path

[szabta89]

The README.md had two places that implied Repo Mind Light always computes embeddings:

1. "What Repo Mind Light Does" — said "It also uses embeddings and a generative model to synthesize a final answer", which is inaccurate when the Soma backend is active.
2. "Cost And Model Considerations" — listed "computes embedding vectors during indexing and refresh" as a blanket cost, and said the CAPI token is needed "for embeddings and answer generation", without acknowledging the Soma path.

Changes

• Updated the high-level description to note that with Soma, embedding-API calls are skipped; with GraphRAG Zero (default), embeddings are still computed.
• Rewrote the cost section to distinguish the two backends, and clarified that a CAPI token is still required for answer generation and query rewriting even when embeddings are not used.

This aligns the packaging repo README with the source repo's documentation, which documents that graph_rag_zero: null + soma: {} skips all embedding-API calls during both indexing and querying.

Generated by Repo Mind Light AW Documentation Updater · ● 13.7M · ◷

• expires on Jul 31, 2026, 10:03 AM UTC