Retrieval embeddings and reranking

How to use
Voyage AI.

Creating high-quality text or multimodal embeddings and reranking candidate results for semantic search, RAG, code search, and domain-specific retrieval.

Open Voyage AI ↗ Official download ↗
What it isembedding and reranking platform for semantic search, retrieval-augmented generation, code and domain retrieval, and multimodal indexing Workflows2 UpdatedSeptember 2026

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 covers the whole path in one place: official access, a first session that produces something reviewable, the checks that make output trustworthy, and the permissions worth limiting before you connect real work.

Troiana principle

AI should make the work easier to inspect. If the workflow removes the source, the owner, or the review step, redesign the workflow.

01

Access & setup

Find, install, and sign in to Voyage AI

Get into the official Voyage AI experience with the right account and a setup you understand.

  1. 01

    Start at https://www.voyageai.com/ and confirm the domain before entering account or payment information.

  2. 02

    Availability: Voyage AI is accessed through its web dashboard, hosted REST APIs, Python and TypeScript clients, and supported private deployment options including MongoDB listings in Azure.

  3. 03

    A Voyage account and API key, server-side secret storage, an owned and permissioned corpus, a vector index, labeled retrieval evaluations, token and rate budgets, approved data handling, monitoring, and a reindex and rollback plan.

  4. 04

    Sign in with the account you intend to keep using, then review plan, data, notification, and permission settings.

  5. 05

    Run one low-risk test task before connecting sensitive files, repositories, or workspace data.

  • Use official download pages.
  • Review permissions during setup.
  • Keep installers and applications updated.
02

First session

Your first useful Voyage AI session

Learn the interaction loop using a small task with a clear outcome.

  1. 01

    Define the search users, corpus owners, permissions, languages and modalities, relevance labels, freshness, latency, throughput, storage, data handling, cost, and deletion requirements.

  2. 02

    Create a server-side API key, choose a current embedding model by measured task fit, set input_type to document for the corpus and query for searches, and record model, dimension, datatype, tokenizer, and truncation choices.

  3. 03

    Index a small versioned corpus with stable IDs and metadata, retrieve candidates on labeled queries, then test whether a current reranker improves ranking enough to justify its latency and cost.

  4. 04

    Measure recall, ranking quality, no-answer behavior, permissions, multilingual and adversarial cases, latency, errors, and spend before canarying, monitoring drift, and preserving the prior index for rollback.

  • State the outcome before the background.
  • Provide the real source material.
  • Review the result before expanding the task.
03

Quality control

Check the quality of Voyage AI output

Turn generated code into a change you would put your name on in review.

  1. 01

    Restate the intended behaviour and the acceptance criteria before reading a single generated line.

  2. 02

    Read the diff rather than the chat summary; the summary describes intent, the diff describes what actually changed.

  3. 03

    Run the test suite, then write a test that would fail if the change were wrong, and confirm it passes for the right reason.

  4. 04

    Check the edges the model tends to skip: error paths, null and empty cases, concurrency, and migration or rollback behaviour.

  5. 05

    Trace every new dependency, network call, and file write, and confirm each one is genuinely required.

  • Review the diff, never the description of the diff.
  • A passing test suite proves the tests pass, not that the change is correct.
  • Keep changes small enough that a human can actually review them.
04

Privacy & permissions

Use Voyage AI safely

Give the tool enough repository access to help, and no more than that.

  1. 01

    Decide what the tool may read and what it may write before connecting a repository.

  2. 02

    Scope tokens and integrations to the narrowest repository, branch, and permission set that still works.

  3. 03

    Keep secrets out of the context window: use environment variables and secret stores, never inline credentials.

  4. 04

    Require review before generated code touches authentication, payments, permissions, migrations, or deletion paths.

  5. 05

    Audit what the integration retained after the session, and revoke access that is no longer needed.

  • 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.
  • Follow your organisation's approved-use policy.
  • Never treat fluent output as authorization to act.
05

Core workflows

Step-by-step ways to use Voyage AI for the work it does best.

Each workflow below has its own outcome, steps and review checkpoints.

01 WorkflowBuild a versioned Voyage retrieval index

Create embeddings from permissioned source material with stable identities, explicit model semantics, reproducible chunking, and deletion lineage.

 02 WorkflowEvaluate Voyage retrieval and reranking

Measure candidate recall and final ranking separately so a polished reranker result cannot hide missing evidence, silent truncation, or permission leakage.

Workflow 01

Build a versioned Voyage retrieval index

Create embeddings from permissioned source material with stable identities, explicit model semantics, reproducible chunking, and deletion lineage.

Outcome: Produce a searchable index that can be evaluated, refreshed, migrated, corrected, and deleted without mixing incompatible vectors or access scopes.

  1. 01

    Define corpus owners, users, source permissions, languages and modalities, freshness, chunking, metadata, access filters, relevance targets, latency, retention, and deletion requirements.

  2. 02

    Create a dedicated server-side Voyage key, confirm the organization's data opt-out setting, select an exact current model, and record tokenizer, input_type, dimension, datatype, normalization and truncation behavior.

  3. 03

    Parse approved copies, remove secrets and irrelevant metadata, assign stable source and chunk IDs, preserve version and access labels, and embed corpus chunks with input_type=document in bounded batches.

  4. 04

    Store model and preprocessing metadata beside every vector, enforce source permissions before ranking, and test duplicate, empty, oversized, malformed, multilingual, image, deletion, retry and partial-batch cases.

  5. 05

    Publish a versioned index, monitor ingestion failures and drift, re-embed changed sources deliberately, verify removal from index and backups, and keep the previous complete index for rollback.

What good use looks like

  • Version every vector with its model.
  • Filter permissions before returning results.
  • Make deletion traceable to the source.

Workflow 02

Evaluate Voyage retrieval and reranking

Measure candidate recall and final ranking separately so a polished reranker result cannot hide missing evidence, silent truncation, or permission leakage.

Outcome: Choose an embedding and reranking pipeline from labeled task evidence with explicit quality, latency, cost, and no-answer gates.

  1. 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.

  2. 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.

  3. 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.

  4. 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.

  5. 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.

What good use looks like

  • Measure retrieval recall before reranking.
  • Never treat relevance score as truth.
  • Roll back model and index together.
06

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 guide.

Explore the Troiana AI Hub →