Keyline FAQ

View as Markdown

Common questions about the documents Keyline reads, the Keyline app, the Keyline API, and what to check when an answer looks wrong.

Documents, whichever way you reach them

These answers hold for both the Keyline app and the Keyline API, because both read the same documents the same way.

What file types does Keyline support?

Keyline accepts PDF, PNG, JPEG, and Office files (DOCX, XLSX, PPTX, DOC, XLS, PPT). Upload one and Keyline analyzes it and builds a search index automatically; see Getting started for the upload flow.

Is there a file size or page limit?

Per-file size and page limits apply. They are preliminary and get adjusted to fit customer needs, so tell us what you are working with. Your library itself keeps growing as you add files, and if an upload is rejected for size or length, split the document and upload the parts.

Will Keyline read scanned documents?

Yes. Every page is analyzed as an image, so a scan is an ordinary case. Architectural, electrical, and wiring drawings are a confirmed failure case in our own testing, so build elsewhere for those.

What languages does Keyline support?

Keyline can read and answer questions in multiple languages, including English and Japanese. You can ask a question in a different language than the source document.

What is the difference between the Keyline app and the Keyline API?

The Keyline app is the hosted interface you sign in to, where you upload files into a library, read them, and ask questions beside the page. The Keyline API is the same retrieval engine over HTTP with no interface of its own: it returns text, page numbers, and image crops for your own agent, application, or internal tool to render. An API call is scoped by a project id you copy out of the app, so the two work over the same documents.

Using the Keyline app

Who can see the files I upload?

Uploaded files and their conversations are visible to the members of the library you upload into. They stay inside that library.

Building on the Keyline API

These answers cover Keyline as an API you call from your own code. The landing page for that audience is Keyline for developers, the hands-on first run is the Keyline API quickstart, and the call-by-call reference is the agent runbook.

Does the Keyline API come with a user interface?

No. The API returns text, page numbers, and cropped images over HTTP, and your application renders them wherever your users already are: a chat transcript, a ticket view, a panel in your own tool. The Keyline app is what that data looks like when Helpfeel renders it.

How does the Keyline API differ from a document parsing API?

Parsing tools convert a document you hand them, per call. Keyline answers a question across a library you indexed earlier and decides for itself which pages to read. For structured output from one known file, a parsing API is the right tool.

Do I have to bring my own vector store?

Keyline builds and owns its own index over the documents you send it. Embeddings, chunking, and the database are handled for you.

Can I use my own model?

The reasoning that decides what to read is part of the product. Keyline returns text and images to your application, and you use whatever model you like from there.

How do I know when a document is ready?

Poll its status. Documents become deep-readable slightly before the library-wide search index finishes, so there are two states that both mean usable.

What does a citation look like?

A numbered reference that resolves to a file, a page number, and a cropped PNG of the block the claim came from. Region coordinates for that block arrive with the preview.

Is there a free tier?

Public Preview access comes with $50 in free credit, then usage-based pricing. Indexing has a real per-page cost, so pricing stays usage-based after the credit runs out. Rates and published limits live on the pricing page.

How do I get an API key today?

Troubleshooting

My upload seems stuck

Indexing runs once, on ingest, and a long document takes longer than a short one. Uploads also queue behind the files ahead of them, so a bulk import finishes in the background rather than all at once. Check the file's status before re-uploading it.

Keyline's answer did not match what I expected

Open the citation to see the block it came from: the page number plus a cropped image of the figure, table, or paragraph behind the claim. If that block does not support the answer, rephrase the question more specifically, for example naming the figure, table, or section you are asking about.