Skip to main content

vai v1.31.0: voyage-4-nano local inference

· 3 min read
Michael Lynn
Michael Lynn
Principal Staff Developer Advocate, MongoDB

vai v1.31.0 adds voyage-4-nano local inference, giving developers a zero-API-key path into Voyage AI embeddings through a lightweight Python bridge built directly into the CLI.

vai pixel robot celebrating success

Release moment

The docs, CLI, and launch messaging now land as one story

This release is where the pixel robot and the product story line up cleanly: local-first onboarding, a friendly CLI, and documentation that carries the same visual cues for getting started, understanding the bridge, and shipping a working retrieval flow.

Go to local inference overview

What shipped​

This release makes local inference a first-class part of the vai story.

Highlights:

  • voyage-4-nano local embedding in vai
  • vai nano setup for one-time environment and model setup
  • --local support on core embedding and pipeline workflows
  • documentation updates across installation, models, commands, and guides
  • a clearer "start local, scale later" path for the Voyage 4 family

Why nano matters​

Before this release, the first-time vai experience usually started with an API key. With voyage-4-nano, you can now install the CLI, set up the local model, and generate embeddings immediately on your machine.

npm install -g voyageai-cli
vai nano setup
vai embed "What is vector search?" --local

That makes local inference the shortest path to a working embedding while still keeping the broader Voyage 4 workflow intact.

The Python bridge​

vai is a Node.js CLI. voyage-4-nano runs in Python. This release connects those two pieces with a lightweight Python bridge that:

  • creates the local runtime environment
  • installs dependencies
  • downloads and caches the model
  • serves local embedding requests back to the CLI

This is intentionally user-visible. Local inference is easy to use, but it is not magic. Python is part of the local runtime story, and vai nano setup handles that setup for you.

Shared embedding space​

The most important product detail is that voyage-4-nano shares embedding space with the rest of the Voyage 4 family:

  • voyage-4-large
  • voyage-4
  • voyage-4-lite

That means local-first is not a dead end. You can prototype locally, then move to API-backed Voyage 4 models later as your needs change.

Try it​

# One-time setup
vai nano setup

# Check readiness
vai nano status

# Smoke test local inference
vai nano test

# Embed locally
vai embed "What is vector search?" --local

# Run the ingestion pipeline with local embedding
vai pipeline ./docs/ --local --db myapp --collection knowledge --create-index

Updated docs​

This release also updates the docs site to make local inference easier to discover and understand:

What this changes for users​

If you are new to vai, local inference is now the simplest way to get started.

If you already use the API-backed Voyage 4 models, nano gives you a local development path that fits into the same family and the same broader product story.