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

[User query]
     │
     ▼
Embed Text (input_type=query)
     │  output: text → JSON float array
     ▼
MongoDB Atlas Vector Search (query_vector = {{embed.text}})
     │  json output: json[0], json[1], ... (individual documents)
     │  text output: stringified doc array (for reranker)
     ▼
Rerank Documents (query, documents = {{vector_search.text}})
     │  json output: json[0], json[1], ... (ranked results)
     ▼
Iterate node / [LLM answer node]

---

Prerequisites

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

---

Installation

From Dify Marketplace

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

Local / Debug Install

cp .env.example .env
# Fill in REMOTE_INSTALL_URL and REMOTE_INSTALL_KEY from Dify Plugin Management page
pip install -r requirements.txt
python -m main

Package for distribution

dify plugin package ./voyage_ai

---

Configuration

When installing the plugin, you will be prompted for:

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

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.

Parameter	Type	Required	Form	Description
text	string	✅	llm	The text to embed. Pipe a workflow variable directly here (e.g. {{sys.query}})
model	select	❌	form	Embedding model (default voyage-4)
input_type	select	❌	form	query (for search), document (for indexing), or empty for generic (default query)
output_dimensions	select	❌	form	Output dimensions: 256, 512, 1024, or 2048 (model default if blank)
truncation	boolean	❌	form	Truncate texts exceeding model token limit (default true)

Output (JSON message):

{
  "model": "voyage-4",
  "input_type": "query",
  "dimensions": 1024,
  "total_tokens": 6,
  "embedding": [0.01, -0.03, ...],
  "embedding_json": "[0.01, -0.03, ...]"
}

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

---

Rerank Documents

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

Parameter	Type	Required	Form	Description
query	string	✅	llm	The search query
documents	string	✅	llm	JSON array of document strings, e.g. ["doc one", "doc two"]. Also accepts a native list piped from another tool (e.g. Vector Search text output).
model	select	❌	form	Reranking model (default rerank-2.5)
top_k	number	❌	llm	Return only top K results (blank = return all)
truncation	boolean	❌	form	Truncate long documents (default true)

Output (JSON): Each ranked result is yielded as its own JSON message — json[0] is the highest-scored, json[1] is the second, etc. Directly iterable in Dify's Iterate/Loop nodes.

// json[0]
{"index": 0, "relevance_score": 0.847, "document": "This quarter..."}
// json[1]
{"index": 2, "relevance_score": 0.269, "document": "Photosynthesis..."}
// json[2]
{"index": 1, "relevance_score": 0.249, "document": "20th-century..."}

---

Supported Models

Embedding Models

Model	Dimensions	Context	Description
voyage-4-large	1024 (default), 256, 512, 2048	32K	Best quality, multilingual
voyage-4	1024 (default), 256, 512, 2048	32K	Balanced quality/cost (recommended)
voyage-4-lite	1024 (default), 256, 512, 2048	32K	Lowest latency and cost
voyage-context-3	1024 (default), 256, 512, 2048	32K	Contextualized chunk embeddings
voyage-code-3	1024 (default), 256, 512, 2048	32K	Code and technical documentation
voyage-finance-2	1024 (fixed)	32K	Finance RAG
voyage-law-2	1024 (fixed)	16K	Legal RAG
voyage-3-large	1024 (default), 256, 512, 2048	32K	Previous generation general
voyage-3.5	1024 (default), 256, 512, 2048	32K	Previous generation general
voyage-3.5-lite	1024 (default), 256, 512, 2048	32K	Previous generation lite
voyage-code-2	1536 (fixed)	16K	Previous generation code

Reranking Models

Model	Context	Description
rerank-2.5	32K	Highest accuracy (recommended, 200M free tokens)
rerank-2.5-lite	32K	Fast and cost-effective (200M free tokens)
rerank-2	16K	Previous generation, multilingual
rerank-2-lite	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.