TL;DR

CrewAI's Voyage AI embedder defaults to the old voyage-2 model. Voyage already accepts any model name we pass, so this is just a default-value change: update it to voyage-4 (Voyage's current recommended general-purpose model) so zero-config users get modern embeddings.

Context

CrewAI's RAG layer turns text into embeddings via a provider. When a user picks Voyage AI without naming a model, it falls back to a hardcoded voyage-2 — an old-generation model. The provider passes the model string straight through to the Voyage SDK with no allowlist, so voyage-4 already works; there's nothing to "add support" for. The stale default lives in three spots plus the docs:

• voyageai_provider.py (field default)
• types.py (type annotation)
• embedding_callable.py (runtime fallback)
• Docs: https://docs.crewai.com/en/tools/ai-ml/ragtool — source ragtool.mdx

Acceptance Criteria

• [ ] The default Voyage AI model is voyage-4 in all three places it's currently voyage-2 (field default, type annotation, runtime fallback).
• [ ] Zero-config Voyage users get voyage-4; explicitly naming a model (e.g. voyage-code-4) still passes through unchanged.
• [ ] Existing Voyage tests pass, updated only where they assume voyage-2 as the default.
• [ ] Docs list voyage-4 as the default and refresh the model options list.