[sohanmaheshwar]

What

Migrates the agentic-rag-authorization example from Milvus + OpenAI to Elasticsearch + Mistral, and promotes it as the canonical version on main.

Changes

Vector store → Elasticsearch

• New elasticsearch_client.py singleton; milvus_client.py removed
• retrieval_node runs kNN search over an Elasticsearch dense_vector field (cosine)
• setup_environment.py creates the documents index and bulk-indexes the 50 docs

LLM + embeddings → Mistral

• mistral-embed embeddings (1024-dim) and mistral-small-latest LLM via langchain-mistralai
• Documents are embedded in a single batched embed_documents call (far fewer requests, avoids rate-limit failures mid-setup)

Docs & config

• README rewritten for the Elasticsearch + Mistral stack: Milvus/Weaviate references removed, consolidated to a single web-UI flow, Project Structure / Testing / Sample Scenarios sections dropped, and the state-flow diagrams are now Mermaid with the SpiceDB authorization boundary highlighted
• Added .env.example (the quick-start copies it; placeholder matches run_ui.py's preflight check)
• ARCHITECTURE.md async note corrected (the ES client and Mistral calls are synchronous)
• config.py, docker-compose.yml, run_ui.py updated for Elasticsearch + Mistral env vars and services

Tests

• test_elasticsearch_client.py added; test_milvus_client.py removed; retrieval/config tests updated for the new code paths

Notes

• The SpiceDB authorization layer is unchanged — it already follows client best practices (bulk checks, TOUCH writes, singleton client, fail-closed).

[tstirrat15]

LGTM, see comment