Yes, optionally, for two accuracy features — Wyregraf remains fully functional without it, on OpenAI and Anthropic alone.
Contextualized embeddings — Voyage's voyage-context-4 model embeds each chunk with awareness of its whole document, an alternative to Wyregraf's own contextual-blurb enrichment (when a Voyage contextual model is selected, the blurb option steps aside automatically so you never pay for both).
Reranking — Voyage's dedicated reranker (rerank-2.5-lite by default) can re-order a chat's retrieved candidates, as an alternative to using your accuracy LLM for the job.
To connect it, add your Voyage API key and models to an embeddings profile, then select that profile in the chat's Voyage Profile field and set Reranker Strategy to Voyage. If Voyage is unreachable or misconfigured, chats quietly fall back to their normal ranking — a failed reranker never breaks answers. Setup walkthrough: Hybrid Search & Reranking.
