Voyage AI Plugin for Dify

A Tool-type Dify plugin that lets Chatflow, Workflow, and Agent applications generate vector embeddings and rerank documents using Voyage AI models hosted on MongoDB Atlas.

Companion plugin: Use the MongoDB Atlas Tool plugin to perform vector search with the embeddings generated here. The Embed Text output can be piped directly into the Vector Search tool's Query Vector field.

Tools

Tool	Description
Embed Text	Generate a vector embedding for a single text string
Rerank Documents	Rerank a list of document strings against a query by relevance score

Typical Workflow

Prerequisites

A MongoDB Atlas account with the Voyage AI preview feature enabled
A Model API key from Atlas → AI Models (key starts with )

Installation

From Dify Marketplace

Search for Voyage AI in the Dify Plugin Marketplace and click Install.

Local / Debug Install

Package for distribution

Configuration

When installing the plugin, you will be prompted for:

Credential	Required	Description
Voyage AI API Key	✅	Atlas model API key starting with

To create a key: Atlas UI → your project → AI Models → Create model API key.

Tool Reference

Embed Text

Generate a vector embedding for a single text. Returns the embedding as a JSON float array string, ready to pipe into the MongoDB Atlas Vector Search tool.

Type	Required	Form	Description
string	✅	llm	The text to embed. Pipe a workflow variable directly here (e.g. )
select	❌	form	Embedding model (default )
select	❌	form	(for search), (for indexing), or empty for generic (default )
select	❌	form	Output dimensions: 256, 512, 1024, or 2048 (model default if blank)
boolean	❌	form	Truncate texts exceeding model token limit (default true)

Output (JSON message):

Output (text message): The embedding as a plain JSON array string — pipe directly into Vector Search's field.

Rerank Documents

Rerank a list of documents against a query. Returns documents sorted by relevance score (highest first).

Type	Required	Form	Description
string	✅	llm	The search query
string	✅	llm	JSON array of document strings, e.g. . Also accepts a native list piped from another tool (e.g. Vector Search output).
select	❌	form	Reranking model (default )
number	❌	llm	Return only top K results (blank = return all)
boolean	❌	form	Truncate long documents (default true)

Output (JSON): Each ranked result is yielded as its own JSON message — is the highest-scored, is the second, etc. Directly iterable in Dify's Iterate/Loop nodes.

Supported Models

Embedding Models

Dimensions	Context	Description
1024 (default), 256, 512, 2048	32K	Best quality, multilingual
1024 (default), 256, 512, 2048	32K	Balanced quality/cost (recommended)
1024 (default), 256, 512, 2048	32K	Lowest latency and cost
1024 (default), 256, 512, 2048	32K	Contextualized chunk embeddings
1024 (default), 256, 512, 2048	32K	Code and technical documentation
1024 (fixed)	32K	Finance RAG
1024 (fixed)	16K	Legal RAG
1024 (default), 256, 512, 2048	32K	Previous generation general
1024 (default), 256, 512, 2048	32K	Previous generation general
1024 (default), 256, 512, 2048	32K	Previous generation lite
1536 (fixed)	16K	Previous generation code

Reranking Models

Model	Context	Description
	32K	Highest accuracy (recommended, 200M free tokens)
	32K	Fast and cost-effective (200M free tokens)
	16K	Previous generation, multilingual
	8K	Previous generation lite, multilingual

Support

GitHub Issues: https://github.com/mongodb-developer/dify-plugins-voyageai/issues
MongoDB Developer Community: https://www.mongodb.com/community/forums/

License

Apache 2.0 — see LICENSE for details.