The outcome
Choose an embedding and reranking pipeline from labeled task evidence with explicit quality, latency, cost, and no-answer gates.
Voyage AI is a embedding and reranking platform for semantic search, retrieval-augmented generation, code and domain retrieval, and multimodal indexing. Creating high-quality text or multimodal embeddings and reranking candidate results for semantic search, RAG, code search, and domain-specific retrieval. This guide narrows that broad capability into one repeatable outcome, with checkpoints that keep the source material and your judgment in the loop.
Before you begin
Set the boundary before the tool starts.
Choose one real task, identify who will use the result, and decide what evidence or test will make the result acceptable. Gather only the source material needed for that task. If the work contains confidential, personal, regulated, or client-owned information, confirm that the platform and account are approved before sharing it.
AI should make the work easier to inspect. If the workflow removes the source, the owner, or the review step, redesign the workflow.
Step by step
A workflow you can repeat.
- 01
Build labeled queries with relevant, partially relevant, irrelevant and forbidden documents, including typos, paraphrases, multilingual, long-tail, ambiguous, no-answer, stale, adversarial and duplicate cases.
- 02
Embed queries with input_type=query using the exact corpus-compatible model, retrieve a generous candidate set, and measure recall at k before adding any reranker.
- 03
Run a current reranker on only authorized candidates, set truncation deliberately, sweep candidate and top-k sizes, and measure ranking quality, no-answer behavior, tail latency, token use, errors and cost.
- 04
Inspect failures against source text, especially relevant passages truncated from long documents, near-duplicate inflation, metadata mismatch, query-document role reversal, and scores treated as calibrated confidence.
- 05
Set release thresholds and monitoring samples, canary the full pipeline, track model and index versions per result, and roll back both together when compatibility or quality changes.
Working standard
What good use looks like.
- Measure retrieval recall before reranking.
- Never treat relevance score as truth.
- Roll back model and index together.
Embedding compatibility, dimensions, tokenizers, limits, and recommended models change; automatic truncation can silently remove decisive content, and relevance scores are not truth or permission checks. Voyage says hosted-API customers can opt out of storage and training use for zero-day retention through an organization-admin control. Keep keys server-side, confirm current data settings, preserve source permissions and deletion, separate query and document inputs, version every vector with its model configuration, evaluate reranking on labeled data, and reindex deliberately when compatibility changes.
Official references
Check the current product documentation.
Features, plan limits, availability, and data controls change. These official pages are the starting points used for this collection.