Skip to content
Corvid Docs

RAG QA Bot Real Providers

Real provider mode is opt-in. CI and normal tests use deterministic mock retrieval and mock LLM responses.

OpenAI Embeddings

Environment:

Terminal window
CORVID_RUN_REAL=1
OPENAI_API_KEY=<redacted-openai-key>
CORVID_RAG_EMBEDDER=openai:text-embedding-3-small

The committed demo keeps embeddings in examples/rag_qa_bot/seed/embeddings.json for offline tests. Real mode should rebuild embeddings from seed/kb/ before a live run, then record any provider trace through the redaction pipeline before promotion.

Ollama Embeddings

Minimum setup:

Terminal window
ollama serve
ollama pull nomic-embed-text

Environment:

Terminal window
CORVID_RUN_REAL=1
CORVID_RAG_EMBEDDER=ollama:nomic-embed-text
OLLAMA_BASE_URL=http://localhost:11434

OLLAMA_BASE_URL is optional when Ollama listens on http://localhost:11434. Use it when running Ollama on another host or port.

Mock Mode

Offline tests and CI use:

Terminal window
CORVID_TEST_MOCK_TOOLS={"retrieve_context":["Refunds over one hundred dollars require approval before money moves.","Refunds over one hundred dollars require approval before money moves.","Refunds over one hundred dollars require approval before money moves."]}
CORVID_TEST_MOCK_LLM=1
CORVID_TEST_MOCK_LLM_RESPONSE=Refunds over one hundred dollars require approval before money moves.

The mock returns the same grounded retrieval and answer strings consumed by the Corvid program; the RagAnswer typed surface is unchanged.