Mixedbread

What is Mixedbread?

Mixedbread is a retrieval platform that lets AI agents search your own data, including PDFs, decks, videos, code, Slack and Drive. It handles parsing, OCR, transcription, chunking, embedding and indexing, then returns evidence as file, page and passage.

Key capabilities of Mixedbread

• Toast 1 search agent: runs the search loop for your agent with search, grep and inspect tools.
• Wholembed V3: an embedding model for text, images and other media across 100+ languages.
• mxbai-rerank: a listwise reranker that reorders candidates.
• Silo vector store: built for multi-vector embeddings and S3-native storage.
• Search modes: semantic, grep, reranking and agentic search chosen with one flag.
• SDKs and MCP: Python and TypeScript SDKs, an MCP server and a CLI.

How Mixedbread works

You upload files and Mixedbread parses, chunks, embeds and indexes them. Your agent queries through the SDK, MCP server or CLI, and gets evidence back with file, page and passage references. Authentication uses an MXBAI_API_KEY.

Who uses Mixedbread?

Mixedbread is for developers building AI agents and search features over private data, from small teams to enterprises that need custom limits, bring-your-own-bucket or bring-your-own-cloud options.

Mixedbread pricing

Starter is free with $5 in one-time credits and no card. Scale is $20 per month with $20 in monthly credits, then pay-as-you-go. Enterprise is custom. All plans use the same usage rates, such as $0.50 per 1M content tokens per month for storage.

Mixedbread alternatives

Voyage AI offers embedding models and rerankers through an API, so it covers the model layer without the full search platform. Mindee and Veryfi focus on extracting data from documents rather than agent retrieval. Mixedbread differs by bundling parsing, embeddings, reranking and a vector store.