Settings

How this deployment is configured, and the taxonomy every screen in the app reads from.

Core configuration

How this deployment is wired up. These values come from environment variables and the model registry, so they are shown read-only — change them in .env.local (or the hosting provider’s environment) and restart.

Database
Supabase Postgres, and the corpus it holds
Connected

Project URL

https://wingpaaavolugdsmuvlm.supabase.co

NEXT_PUBLIC_SUPABASE_URL

Service role key

Configured

SUPABASE_SERVICE_ROLE_KEY

Documents

0

documents table

Indexed passages

0

document_chunks with an embedding — what semantic search reads

Models
What every AI call uses. Per-capability overrides and spend limits live under AI in the sidebar.
Active

API key

Configured

GEMINI_API_KEY

Web search grounding

Enabled

Lets an answer cite the live web alongside uploaded guidelines

Text model

gemini-3.8-flash

GEMINI_MODEL_DEFAULT

Embedding model

gemini-embedding-2 · 1536d

GEMINI_MODEL_EMBEDDING — must match the vector() column width

Retrieval
How a program area changes which documents ground an answer
Storage ready

Selecting a program re-ranks the corpus; it never filters it. A question that needs two programs still reaches both, and choosing a program can never leave an answer with no sources.

On the selected program
+0.06 to the similarity score
Cross-cutting clinical guidance
+0.03 — consolidated guidelines stay reachable from every program
Policy and systems material
−0.04 — kept available, but never outranks a treatment protocol

Document storage

compass-dev

R2_BUCKET_NAME — where PDFs and previews are kept