feat(ai): default Anthropic installs to pgvector RAG (4/5)

[gariasf]

Summary

Closes the Anthropic RAG loop. PR 4 of 5, stacked on #1985.

Background

The provider-agnostic vector store stack already shipped to main:

• VectorStore::Pgvector — local pgvector adapter with chunk + embed pipeline
• VectorStore::Embeddable — embeddings via any OpenAI-compatible endpoint (EMBEDDING_URI_BASE, EMBEDDING_ACCESS_TOKEN)
• VectorStore::Registry — VECTOR_STORE_PROVIDER env routes between openai, pgvector, qdrant
• db/migrate/20260316120000_create_vector_store_chunks.rb — pgvector table + index

This PR makes Anthropic the first-class citizen of that stack.

What changed

• VectorStore::Registry.adapter_name now returns :pgvector when Setting.llm_provider == "anthropic" and no explicit VECTOR_STORE_PROVIDER override is set. Anthropic has no hosted vector store, so the local pgvector adapter is the only correct default. Explicit VECTOR_STORE_PROVIDER=... still wins.
• Assistant::Function::SearchFamilyFiles updates its "no adapter" error message to mention pgvector + EMBEDDING_URI_BASE as the supported Anthropic-only path, instead of pointing exclusively at OpenAI.

What didn't change (intentional)

• Embedding generation is decoupled from the chat provider. Anthropic installs point EMBEDDING_URI_BASE at Voyage AI (Anthropic's recommended embeddings partner), a local Ollama instance, or OpenAI's text-embedding-3-small. No new gem; no new credentials beyond what already exists.
• VectorStore::Pgvector, VectorStore::Embeddable, the migration, and the vector_store_chunks schema are unchanged.

Setup checklist for Anthropic-only installs

# 1. Enable pgvector in Postgres (one-time)
psql -d sure_production -c "CREATE EXTENSION vector;"

# 2. Run the existing migration
bin/rails db:migrate

# 3. Configure embeddings — examples:
#    OpenAI embeddings (cheap, no chat-provider dep)
EMBEDDING_URI_BASE=https://api.openai.com/v1/
EMBEDDING_ACCESS_TOKEN=sk-...
EMBEDDING_MODEL=text-embedding-3-small
EMBEDDING_DIMENSIONS=1536

#    Voyage AI
EMBEDDING_URI_BASE=https://api.voyageai.com/v1/
EMBEDDING_ACCESS_TOKEN=pa-...
EMBEDDING_MODEL=voyage-3

#    Ollama (fully local)
EMBEDDING_URI_BASE=http://localhost:11434/v1/
EMBEDDING_MODEL=nomic-embed-text
EMBEDDING_DIMENSIONS=1024

With LLM_PROVIDER=anthropic set, VectorStore.adapter now resolves to pgvector automatically — no extra VECTOR_STORE_PROVIDER needed.

Test plan

• test/models/vector_store/registry_test.rb — 3 new tests:
  • defaults to pgvector when LLM_PROVIDER=anthropic
  • routes to pgvector even when an OpenAI key is also present
  • explicit VECTOR_STORE_PROVIDER=qdrant still wins on Anthropic installs
• All existing registry tests untouched and still pass
• Full suite: 4380 runs, 18086 assertions, 0 failures, 26 pre-existing skips, 1 pre-existing libvips env error
• bin/rubocop clean
• bin/brakeman --no-pager clean

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 91dd230e-f36f-4053-9fc7-01a290e430a6

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feature/anthropic-rag-pgvector

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}