From uploaded files to cited answers
Bee's document pipeline is built into every tier: upload, chunk, embed, index, cite. Vision and video input on Comb, Buzz, and Hive take it past plain text.
One pipeline, end to end
Document intelligence on Bee is not a bolt-on vector database you assemble yourself. Upload a document and the pipeline handles chunking, embedding, and indexing; ask a question and answers come back grounded with citations to the source passages.
The same capability is exposed everywhere you use Bee - the workspace, the OpenAI-compatible API, and the MCP server's document tools - so a knowledge base built once serves every surface.
Capacity that scales with your plan
- →Bee Cell (free) - device-local text and code retrieval with no Bee document-count cap. Files are selected explicitly and only retrieved excerpts are sent when the user enables Chat retrieval.
- →Bee Brood - up to 100 documents of hosted retrieval.
- →Bee Comb - up to 1,000 documents.
- →Bee Buzz - up to 5,000 documents, with the index shared across the team.
- →Bee Hive - up to 25,000 documents.
- →Bee Swarm - up to 100,000 documents.
Documents aren't only text anymore
Comb, Buzz, and Hive accept image and video input natively. Screenshots, scanned pages, diagrams, and screen recordings go into the same conversation as text documents - ask what changed between two dashboard screenshots, or what a recorded demo actually shows.
For questions that must span an entire corpus at once, Comb and Buzz also serve a 1M-token long-context mode: load the full document set into the window instead of retrieving pieces of it.
Grounded answers over confident guesses
Every retrieval answer carries citations back to source passages, so reviewers verify rather than trust. That's the difference between an assistant you can use for real work and one you have to babysit.
Isolation by default
Retrieval indexes are per tenant. Shared cross-member document context remains in progress until tenant-boundary validation is complete. Documents are encrypted at rest with AES-256-GCM under per-tenant keys.
Frequently asked questions
- Do answers include citations to my documents?
- Yes - retrieval answers are grounded with cite-backs to the source passages, so you can verify every claim against the underlying document.
- What file inputs are supported?
- Text documents flow through the retrieval pipeline on every tier, and Comb, Buzz, and Hive additionally accept images and video natively in conversation.
- How many documents can I index?
- Bee Cell has device-local text and code retrieval with no Bee document-count cap. Hosted libraries start on paid plans and scale from 1,000 documents on Comb through 25,000 on Hive and 100,000 on Swarm, with hosted storage caps published on the pricing page.
- Can my whole team search the same documents?
- Not generally today. Shared cross-member document context remains in progress; seats, pooled usage, and organisation administration are available separately.
- Is RAG better than long context for document Q&A?
- They solve different shapes of the problem: retrieval scales to corpora far beyond any window, while the 1M-token long-context mode on Comb and Buzz reasons over everything simultaneously. Bee serves both, so you pick per question.
Related
Start on the free tier
Bee Cell is free - no card. Scale to paid tiers, the API, or sovereign deployment when you are ready.