Ship semantic search in minutes
The complete developer toolkit for Voyage AI embeddings, vector search, and RAG pipelines.
Terminal, browser, or desktop — embed, compare, benchmark, and deploy with MongoDB Atlas Vector Search.
$ curl -fsSL https://vaicli.com/install.sh | shClick to copy · Works on macOS, Linux & WSL
Why I built this around Voyage AI
After evaluating embedding providers, Voyage AI stood out for quality and cost. Here's what makes their models worth exploring.
SOTA Quality
voyage-3-large ranks #1 on MTEB with 67.29 score, outperforming OpenAI, Cohere, and Google.
83% Cost Savings
Starting at $0.02 per million tokens vs OpenAI's $0.13 — same quality, fraction of the cost.
Shared Embedding Space
Embed documents with voyage-3-large, query with voyage-3-lite. Same vector space, 4x faster queries.
Domain Models
Purpose-built models for code, finance, law, and multilingual — outperform general models by 10-20%.
Two-Stage Retrieval
Combine fast embedding search with neural reranking for optimal precision and recall.
Production Ready
Sub-100ms latency, 99.9% uptime SLA, and enterprise-grade security for production workloads.
vai is a community-built developer tool — designed to help you explore Voyage AI embeddings and MongoDB Atlas Vector Search. Not affiliated with Voyage AI or MongoDB.
Everything you need to work with embeddings
Six powerful tools in one CLI — from generating embeddings to benchmarking models and building RAG pipelines.
Embed
Generate SOTA embeddings with Voyage AI's models — #1 on MTEB at half the cost of OpenAI.
Try itCompare
Measure semantic similarity between texts with cosine, dot product, and euclidean distance.
Try itMultimodal
Embed images and text in a shared vector space — search images with text and vice versa.
Try it11 specialized models. One shared space.
Embed documents with voyage-4-large, query with voyage-4-lite — same vector space, 83% cost reduction. The new voyage-4 family sets SOTA on RTEB benchmarks.
| Model | Type | Context | RTEB | Price/1M | Best For |
|---|---|---|---|---|---|
voyage-4-large voyage-4 | Embedding | 32K | 71.41 | $0.12 | Best quality, multilingual |
voyage-4 voyage-4 | Embedding | 32K | 70.07 | $0.06 | Balanced quality & cost |
voyage-4-lite voyage-4 | Embedding | 32K | 68.10 | $0.02 | Lowest cost, high volume |
voyage-code-3 | Embedding | 32K | — | $0.18 | Code & technical docs |
voyage-finance-2 | Embedding | 32K | — | $0.12 | Financial documents |
voyage-law-2 | Embedding | 16K | — | $0.12 | Legal documents |
voyage-context-3 soon | Embedding | 32K | — | $0.18 | Contextualized chunks |
voyage-multimodal-3.5 | Multimodal | 32K | — | $0.12/M + $0.60/B px | Text + images + video |
voyage-4-nano soon voyage-4 | Embedding | 32K | — | Free (open-weight) | Edge / local / self-hosted |
rerank-2.5 | Reranking | 32K | — | $0.05 | Best reranking quality |
rerank-2.5-lite | Reranking | 32K | — | $0.02 | Fast reranking |
RTEB Benchmark — NDCG@10 (29 datasets)
voyage-4-large
71.41
voyage-4
70.07
Gemini Embed 001
68.66
voyage-4-lite
68.10
Cohere Embed v4
65.75
OpenAI v3 Large
62.57
Source: Voyage AI, January 2026
Pro tip: All voyage-4 models share the same embedding space. Index documents with voyage-4-large for best quality, then query with voyage-4-lite to save 83% on API costs — no re-indexing needed.
* = default dimension; also supports other dimension sizes
See the Shared Space in Action
Embed the same text with three different models and watch the vectors land in the same neighborhood. Cross-model similarity of 0.95+ — proven live, not just claimed.
Embed documents with the best model. Query with the cheapest. 83% cost savings. No re-vectorization. Ever.
Powerful from the command line
Simple, intuitive commands that get out of your way.
Install
Configure
Embed
Give your AI agent a knowledge base
The vai MCP server exposes 11 tools over the Model Context Protocol. Any compatible agent can search, embed, rerank, and ingest documents from your collections.
Install
Connect
Use
Agent calls vai_query automatically
11 tools, one protocol
Your agent picks the right tool for the task. No prompting required.
vai_query
Full RAG: embed, search, rerank
vai_search
Raw vector similarity search
vai_rerank
Rerank documents by relevance
vai_embed
Generate embedding vectors
vai_similarity
Cosine similarity between texts
vai_collections
List collections and indexes
vai_models
Browse Voyage AI models
vai_topics
Discover educational topics
vai_explain
Deep-dive on any topic
vai_estimate
Cost calculator for operations
vai_ingest
Chunk, embed, store documents
Built for your domain
Go from sample documents to a working knowledge base in under 30 minutes, with walkthroughs tailored to your industry.
Developer Documentation
Internal docs, API references, and runbooks: semantic search in minutes
Legal & Compliance
Semantic search across legal documents, powered by a model trained on legal text
Financial Services
Earnings calls, risk reports, and policy docs, searchable with a model trained on financial text
Healthcare & Clinical
From clinical guidelines to searchable AI, using your own infrastructure
A full desktop experience
The Vai desktop app wraps the entire CLI experience in a beautiful Electron interface. Embed text, compare documents, run benchmarks, and explore vector search concepts — all with a visual interface.
Signed & notarized
Auto-updates
Vai Desktop
Enter text to embed...
[0.0234, -0.0891, 0.1456, 0.0023, ...]
voyage-4 · 1024 dimensions · 138ms
Community-Built. Globally Used.
Exploring Voyage AI embeddings from the terminal, browser, and desktop.
Countries
Telemetry Events
CLI Activity
Web + Desktop Activity
“vai is now used by developers in 0 countries worldwide”
Delayed aggregate telemetry • Data through the previous UTC day