@tanstack/ai-cohere weekly downloads statistics

@tanstack/ai-cohere

Cohere adapter for TanStack AI. It gives you two things: multimodal embeddings with embed-v4.0, and document reranking with Cohere's rerank models — the precision step for RAG and search pipelines.

For chat, summarization, or media, use OpenAI, Anthropic, or Gemini.

Install

Setup

Get your API key from the Cohere Dashboard and set it as an environment variable:

Usage

Embedding Adapter

Multimodal Inputs

embed-v4.0 embeds text, images, and fused text+image items (one vector per input item). Fuse parts by nesting them in an array — the outer array is the item list:

Cohere does not fetch remote image URLs. Pass base64 data or a data: URI, or enable allowUrlFetch in the adapter config to have the adapter download http(s) URLs and inline them.

Rerank Adapter

With Explicit API Key

Both adapters read COHERE_API_KEY from the environment. To pass a key explicitly, use the create* factories:

Supported Models

Embedding Models

• embed-v4.0 - Multimodal embedding model (text + images, Matryoshka dimensions via the top-level dimensions option)

Rerank Models

• rerank-v3.5 • rerank-english-v3.0 • rerank-multilingual-v3.0

Features

• ✅ Embeddings (batch, one request per input array) • ✅ Multimodal embedding input (text + images + fused text/image items) • ✅ Dimension reduction (dimensions → Cohere output_dimension) • ✅ Document reranking • ❌ Chat / text generation • ❌ Image generation

Tree-Shakeable Adapters

This package uses tree-shakeable adapters, so you only import what you need:

Read the docs ->

• Reranking Guide — object documents, RAG pipelines, options, and the result shape. • Cohere Adapter — models, configuration, and explicit API keys.

License

MIT

Description from the npm listing.