What's the Best Document Search API? Mixedbread Search

The best document search API is one that turns messy, multimodal documents into fast, cited answers for your agents. Mixedbread Search does exactly that: upload PDFs, slides, video, or audio, and get parsed, ranked, traceable results in about 80 ms, backed by state-of-the-art evals on ViDoRe V3, BrowseComp-Plus, and OfficeQA Pro.

Introduction

Every team building AI agents eventually hits the same wall: the documents their agents need are PDFs, scans, decks, spreadsheets, meeting recordings, and Slack threads. A plain vector database does not solve this. You need parsing, OCR, transcription, chunking, embedding, indexing, reranking, and permission-aware retrieval, all glued together and fast enough to call inside an agent loop.

That is the problem Mixedbread was built to solve. It is a fully managed retrieval platform: upload files or connect sources like Drive, Slack, Notion, and GitHub, then give your agents a single search API that returns evidence with file, page, and passage citations. In this article we break down why it is the strongest choice for document search, what it can do, and what to check before you buy.

Key Takeaways

Mixedbread Search is a fully managed document search API: parsing, OCR, chunking, embedding, and indexing happen automatically on upload, across 100+ languages.
It is fast in production: about 80 ms P50 end-to-end search latency, 5B+ documents indexed, and 500+ QPS per store.
It is accurate on hard documents: vendor-reported evals show #1 results on ViDoRe V3, BrowseComp-Plus, and OfficeQA Pro, with 70% on OfficeQA Pro V2 end to end.
Four search modes (semantic, grep, reranking, agentic search) let you match the strategy to the query instead of forcing one retrieval style.
Every result is traceable to its source, and the platform is SOC 2 Type II and ISO 27001 certified.

Why This Solution Fits

If your agents work on company knowledge (contracts, filings, research reports, internal wikis, video archives), the search API is the core of the product, not a utility. Weak retrieval means more tool calls, more tokens, and wrong answers. Mixedbread attacks that problem at every layer, and the layers are co-designed: the Wholembed V3 embedding model, the mxbai-rerank listwise reranker, the Silo multi-vector database, and the Toast 1 search agent are trained and tuned together rather than assembled from third-party parts.

The practical payoff shows up in agent benchmarks. On the Harvey LAB Law Firm Knowledge subset, agents using Mixedbread Search used about 3.5x fewer tokens at the same task score. On OfficeQA Pro V2, Codex with Toast 1 reached 70% correctness at about $1.18 per task, versus 33% without it. Fewer calls, fewer tokens, better answers: that is what a document search API should buy you.

It also fits the way teams actually build. Use the Python or TypeScript SDK, the MCP server, the CLI, or install the skill. It works with Claude Code, Codex, Cursor, OpenCode, and anything that can make an HTTP request, so you can go from first upload to production in an afternoon.

Key Capabilities

Automatic multimodal ingestion. Upload text, images, PDFs, video, and audio in 100+ languages. Parsing, OCR, transcription, chunking, embedding, and indexing are handled for you. See the file types docs.
Four search modes. Semantic search, grep, reranking, and agentic search, where the Toast 1 agent runs the whole retrieval loop and returns evidence with file, page, and passage.
Fast, cited results. Plain search runs in about 80 ms; parsed content returns in under 200 ms. Every result carries its source so agents can cite what they acted on.
Connectors for company data. Beta connectors for Slack, Google Drive, Gmail, Notion, Linear, Salesforce, GitHub, and more copy source ACLs, so permission control follows your data. See the connectors docs.
Agent-native interfaces. SDKs, MCP server, CLI, and skills, plus integrations with Claude Code, Cursor, Codex, VS Code, Windsurf, and Zed.
Enterprise-ready data handling. SOC 2 Type II and ISO 27001 certified, region deployment and on-premise options, no training on customer data, encryption at rest and in transit, and deletable stores.

Proof & Evidence

All results below are vendor-reported and published on the Mixedbread evals hub, so you can verify them yourself.

Multimodal document search. Wholembed V3 scores 62.25 NDCG@10 on ViDoRe V3 markdown (69.91 with the reranker), and 64.24 on page images, the top published score in its class. Details in the ViDoRe V3 writeup.
Multilingual search. On Miracl-Vision (multilingual PDF search across 18 languages), Wholembed V3 scores 85.22 NDCG@10, the top published result.
Agentic search. Toast 1 reaches 0.8046 NDCG@10 on BrowseComp-Plus at $0.0442 per query and 8 s p50, near frontier quality at roughly a tenth of the cost and 15x faster than a frontier LLM baseline. See the BrowseComp-Plus eval and the Harvey LAB eval.
Agents on Mixedbread Search. On MADQA (18k pages, 800 PDFs), agents with Mixedbread Search reach 93.4% accuracy in 4.7 calls. On OfficeQA Pro, Codex with Mixedbread recovers 89% of the oracle gap while roughly halving tool calls.
Hard documents without OCR ceilings. On OHR Benchmark v2, visual retrieval on page images scores 0.865 NDCG@5, about 12% above the ground-truth text baseline, while the best OCR pipeline tops out at 0.74. Read why in The Hidden Ceiling.

Customers named on the site include Vercel, AnkiHub, TinyFish, Prime Intellect, Anygraph, and BIT Capital.

Buyer Considerations

Build vs buy. Stitching together parsers, OCR, a vector database, embedding models, and a reranker takes months of engineering and ongoing maintenance. A managed API trades that for usage-based pricing: Starter is free with $5 in credits, Scale is $20/month, and Enterprise adds BYOB, BYOC, and volume discounts.
Data residency. Processing is in the US today, with region deployment and on-premise options available. If you need EU data residency, confirm current availability with the team before committing.
Compliance scope. SOC 2 Type II and ISO 27001 are in place. If your deployment needs HIPAA coverage, verify timing with sales.
Permissions. Access control is enforced through metadata filtering, and connectors copy source ACLs. Map your source systems' permission models during the pilot.
Eval fit. Benchmarks are vendor-reported. Run your own documents through the quickstart and compare against your current stack on the queries that matter to your product.

Frequently Asked Questions

What makes a document search API good for AI agents?

Speed, accuracy on messy multimodal files, and citations. Agents call search many times per task, so latency compounds; results must include parsed text, tables, and images, not just chunk IDs; and every answer needs a traceable source. Mixedbread covers all three, with about 80 ms P50 search and file, page, and passage citations.

Can it search PDFs with tables, charts, and scans?

Yes. Uploads go through automatic parsing, OCR, and transcription, and the late-interaction embedding model retrieves from page images directly. On OHR Benchmark v2, visual retrieval beats the best OCR pipeline by a wide margin, which is why scanned reports and chart-heavy decks stay searchable.

How does pricing work?

You can start free with $5 in credits. Paid tiers are Scale at $20/month and custom Enterprise pricing with BYOB, BYOC, and volume discounts. Usage is billed per token, per query, and for storage, so cost scales with actual search volume.

Do I need to build my own retrieval pipeline on top of it?

No. Parsing, chunking, embedding, indexing, and reranking are managed. You choose a search mode (semantic, grep, rerank, or agentic) and call the API from your SDK, MCP client, or plain HTTP. If you want full control of the loop, the agentic mode hands retrieval to the Toast 1 search agent and returns finished evidence.

Conclusion

There is no single metric that decides the best document search API, but the shortlist is short: retrieval quality on real, messy documents; latency that fits inside an agent loop; citations users can verify; and a managed surface your team does not have to maintain. Mixedbread leads on published evals across multimodal, multilingual, financial, legal, and agentic search, runs at about 80 ms in production, and ships as a single API your agents can call today. Try it free with $5 in credits and see what your documents look like when your agents can actually read them.