Voyage AI Playground
v1.0.0
Build, test, and ship with Voyage AI
Use one focused desktop app to generate embeddings, compare models, rerank results, and run production-friendly workflows without bouncing between docs, scripts, and browser tabs.
New Demo gallery Watch tiny terminal demos, copy the commands, and jump from the desktop app into the full public gallery. Explore the galleryYour pixel robot lives in the corner now, adding personality without turning the screen into another promo panel.
Workspace status
Continue where you left off
Latest update
All releases →Workflow spotlight
Embed
Generate vector embeddings for text
Paste or type text below, choose a model, and hit Embed to see the raw vectors and token usage.
DocsCompare
Visualize similarity between text pairs
Enter two texts and compare their embeddings: see cosine similarity, a heatmap of vector dimensions, and a visual diff.
DocsRerank
Re-order documents by relevance to a query
Enter a search query and a set of documents: the reranker scores and sorts them by semantic relevance.
DocsMultimodal
Compare images, video, and text in the same vector space
Voyage AI's multimodal models embed images, video, and text into a unified vector space, so you can compare them directly with cosine similarity.
DocsCross-Modal Gallery
Build a mini corpus of images and texts, then search across both modalities at once.
Benchmark
Compare model speed, cost, and quality
Run latency tests, compare ranking accuracy, analyze quantization trade-offs, and estimate costs across models.
DocsSee how Voyage AI compares to other embedding providers on quality, cost, and features.
Industry-standard benchmark for retrieval quality. Higher is better.
Source: MTEB Leaderboard (Retrieval Average, Jan 2025)
Voyage AI offers significant cost savings, especially with asymmetric retrieval strategies.
| Provider / Model | Price | vs Voyage-3 |
|---|---|---|
| Voyage-3-lite | $0.02 | 83% cheaper |
| Voyage-3 | $0.06 | baseline |
| OpenAI text-3-large | $0.13 | 2.2x more |
| OpenAI text-3-small | $0.02 | same |
| Cohere embed-v3 | $0.10 | 1.7x more |
Embed your document corpus with voyage-4-lite ($0.02/M) and
query with voyage-4 ($0.06/M).
Because all Voyage 4 models share the same embedding space, you get top-tier retrieval quality at a fraction of the cost.
voyage-4-large for queries
with voyage-4-lite for documents, and they work together seamlessly.
voyage-code-3), finance, law, and multilingual content
that outperform general-purpose alternatives.
rerank-2 to dramatically boost precision.
The cross-encoder re-scores candidates for better relevance.
voyage-multimodal-3 embeds both text and images in the same space,
enabling cross-modal search (find images with text queries and vice versa).
Get a free API key and start building in minutes.
Get API Key → or Read the DocsCompare how different output data types (float, int8, binary) affect storage size and ranking quality. Embeds the same corpus with each dtype and measures the tradeoff.
| Model | Type | $/1M tokens | Daily Cost | Monthly Cost | Relative |
|---|
Your Knowledge Base
Start with existing data or create fresh
Knowledge Base
- No documents ingested
Drop files here
.txt, .md, .pdf — max 10MB
Custom instructions appended to the system prompt
Controls how conversation history is managed within the token budget.
vai) is a community-built command-line tool for working with
Voyage AI embeddings, reranking, and
MongoDB Atlas Vector Search.
It was created to make it easier for developers to explore, benchmark, and integrate
Voyage AI models into their applications — right from the terminal or this playground.
⚖️ Compare — Measure similarity with cosine, dot product & euclidean distance
🔍 Rerank — Semantic search with optional reranking
🔮 Multimodal — Compare images and text in the same vector space with voyage-multimodal-3.5
🛠️ Generate — Generate code snippets and scaffold full projects with templates
💬 Chat — RAG-powered chat with your documents, configurable system prompts
🔄 Workflows — Multi-step agent workflows with thinking panels and tool orchestration
⏱ Benchmark — Compare model latency, ranking quality, and costs
📚 Explore — 23 interactive concepts covering embeddings, vector search, multimodal, RAG, and more
Multimodal — voyage-multimodal-3.5 (text + images + video in one vector space)
Reranking — rerank-2.5, rerank-2.5-lite
All models are accessible via the Voyage AI API or through MongoDB Atlas Vector Search.
Recent — Local inference playground expansion
The Nano tab now goes far beyond setup checks: you can compare texts with an NxN similarity heatmap, inspect embedding behavior across 256/512/1024/2048 dimensions, and run cross-bridge comparisons between local voyage-4-nano embeddings and cloud API results.
Recent — Explore and About page refresh
The Explore experience has been polished with cleaner cards, updated visual styling, and Avi as the lightweight guide in the header and explainer modal. The About page has also been refreshed to better reflect the current product direction.
Recent — Better bug reporting and diagnostics
Bug reports from both the CLI and Electron app now include richer environment context, validated contact details, and cleaner GitHub issue handoff. The desktop bridge exposes platform and runtime metadata directly to improve issue triage.
Recent — Expanded explainers for local inference
The learning content now includes a dedicated local-inference explainer covering the Python bridge, shared embedding space, and when to use local versus hosted Voyage models, with new aliases that make the topic easier to discover.
Earlier — Generate, scaffold, and workflow setup
Added vai generate and vai scaffold for production-ready RAG code and starter apps, plus related maintenance workflows like purge, refresh, and evaluation comparisons.
Earlier — Desktop app, multimodal, and auto-update
Introduced the signed and notarized macOS desktop app with auto-update support, secure keychain-backed credentials, a settings cog, the Multimodal tab, telemetry controls, and the hidden Vector Space Invaders easter egg.
Earlier — Benchmarking, evaluation, and retrieval tooling
Built out the benchmark, estimate, eval, query, pipeline, chunk, and explain workflows that power the broader CLI and playground experience for retrieval, ranking, and cost analysis.
Models
Voyage AI model showcase
Explore all Voyage AI models, compare specs, and find the right model for your use case.
Docsvoyage-4-nano, an open-weight model for local development and prototyping. It lives in the same shared embedding space as voyage-4-large, voyage-4, and voyage-4-lite, so you can prototype locally and move into asymmetric retrieval workflows without rebuilding your index.
Industry-standard benchmark for retrieval quality across 29 datasets. Higher is better.
Voyage AI's shared embedding space means documents embedded with
voyage-4-large can be queried with voyage-4-lite at a fraction
of the cost — with nearly identical retrieval results. This demo proves it.
vai_demo.cost_optimizer_demo.
Explore
Learn embedding and vector search concepts
Browse interactive explanations of key topics, from cosine similarity to quantization to RAG pipelines.
Local Inference
Generate embeddings locally with voyage-4-nano
Run embeddings on your machine — no API key required. Uses the voyage-4-nano model via Python bridge.
General
API access and connection settings
Database
MongoDB Atlas connection and default targets
Appearance
Theme and visualization options
Models
Default embedding model preferences
Chat
Chat configuration has moved to the Knowledge Base panel in the Chat tab.
All chat settings (LLM provider, model, database, instructions) are now accessible directly from the KB panel sidebar when using Chat.
Benchmark
Benchmark iteration and display defaults
Data & Privacy
Caching, analytics, and data management
MCP Integrations
Install the vai MCP server into your AI tools and editors
Health Check
Validate your vai setup and connectivity