Mixedbread

Mixedbread is an end-to-end search and retrieval platform built for AI agents. Instead of stitching together a 3-stage pipeline of parser + embedding model + vector DB + reranker, teams hit a single API that ingests text, PDFs, tables, images, audio, and video, then returns relevant results in under 200ms. The platform owns the full stack: their own embedding and reranker models (mxbai-embed-large, mxbai-rerank, and follow-on models), their own parsing for complex documents, and a managed multi-modal search engine — all behind one client library available for Python and TypeScript. Mixedbread claims #1 on BrowseComp-Plus (an agentic deep-research retrieval eval) with the fewest search calls of any setup tested, plus strong results on MADQA and OfficeQA Pro. The product is positioned at agent builders who want 'no infrastructure, no embeddings to tune, no 3-stage pipeline to manage' — you upload, you search, you ship.

Key capabilities at a glance: Single API for text, PDF, image, audio, and video search; Sub-200ms search latency across 100+ languages; In-house embedding and reranker models (mxbai family); #1 on BrowseComp-Plus agentic retrieval benchmark; Automatic document parsing — no manual chunking; Bring Your Own Cloud (BYOC) deployment for enterprise.

Where Mixedbread wins: One API replaces a 4-piece DIY RAG stack — meaningful time savings; Sub-200ms retrieval helps keep agent loops snappy; In-house models tuned to work together avoids 'embedding/reranker mismatch' bugs; Strong BrowseComp-Plus result is credible third-party evidence for agentic retrieval quality; Multimodal ingest (PDFs, images, audio, video) in one index is rare.

Trade-offs to weigh: Lock-in to mxbai models — you can't swap in OpenAI/Cohere embeddings; Less control than a DIY stack for teams that need custom chunking or reranking; Newer than Pinecone or Weaviate; long-term roadmap less proven; $20/mo Scale plan is good for most teams but enterprise pricing isn't public.

Best-fit scenarios include: Powering retrieval for AI research agents; Multi-modal search across enterprise knowledge; Replacing a hand-built RAG pipeline with a managed API; Search across PDFs, screenshots, and call recordings in one index.

Pricing structure: Starter (Free) — $5 one-time credits, 3 workspace users, 10 stores, 100 requests/min, community Slack support. No card required. | Scale ($20/month) — $20/month included credits, unlimited workspace users, 10,000 stores, 1,200 queries/min, 360 ingestion/min, priority Slack support. | Enterprise (Custom) — Volume-based discounts, dedicated infrastructure, BYOC, point-in-time recovery, dedicated support.