Embedded knowledge infrastructure
They upload it in your product. We handle chunking, embedding, retrieval, and keeping every customer’s data completely separate.
The uploader above is GTWY. Your users only ever see your product.
Why this is harder than it looks
Parsing, chunking strategy, embedding choice, reranking, index freshness, evaluation. Each one is a decision that quietly determines whether the answers are any good — and you only find out after your users do.
Each of your customers needs their own index, their own documents, their own retrieval scope — and absolute certainty that customer A never sees customer B’s data.
Which means a document management UI — upload, organise, delete, indexing status — that you now also have to build and maintain.
That’s three problems. We’ve solved all three.
How it works
What you get
Chunking, embedding, reranking, re-indexing — all handled.
How it works →Isolated at the index level, not filtered at query time.
Data isolation →Your users manage their own documents — upload, organise, delete, watch indexing status — inside your product, in your branding. This is the part a vector database doesn’t give you, and the part you would otherwise be building yourself.
Data sources →Files, URLs, Drive, Notion, S3, databases.
Data sources →With Agent Embed
Retrieval happens inside the agent run. There is no separate call and no separate bill — storage is the only thing metered on its own.
Reference architecture
One platform, many customers, each with their own knowledge base underneath it. Their architecture is the reference for yours.
Pricing
Included in the agent run. Not charged per query.
$0.10 per GB of indexed data, per day.
Questions
PDF, DOCX, TXT, MD, CSV, XLSX, PPTX and HTML, plus images through OCR. Web pages by URL or sitemap, and connected sources like Drive, Notion and S3. The full table with size limits is on the data sources page.
A typical document is searchable within seconds of upload finishing; large files and OCR jobs take longer. Indexing status is visible to your user in the upload UI, so nobody has to guess.
No. Each customer’s documents are indexed separately rather than pooled and filtered at query time, so another customer’s content is never a candidate for retrieval in the first place.
You choose, and you bring your own provider key. Embedding costs go straight to your provider account at your negotiated rate — we never mark them up, because we never touch them.
Their documents and their index are deleted. Their content is exportable before deletion if they want to take it with them.
Large enough that document count is rarely the constraint — per-customer storage caps are set by you, and the current file size and ingest limits are listed on the data sources page.
We’ll show you the upload panel running in your branding on the call.