Documentation

Search infrastructure for everything your company knows.

Hayfork indexes your websites and files, lets you group them into collections, and answers questions over them with sources. People use it through a console or your product. Agents use it through one API.

Four things to know

Everything in Hayfork is one of these, and every guide uses the same words for them.

TermWhat it is
WorkspaceYour isolated account. Sources, documents, collections, keys and usage all belong to exactly one workspace. Nothing is shared between workspaces.
Data sourceWhere content comes from: a website you crawl or a set of uploaded files. A source owns its documents and can be re-synced on its own.
CollectionA search scope made of one or more data sources. Every search names a collection. A source can sit in many collections, and its documents count once.
API keyHow your product and your agents authenticate. Keys carry scopes, so a key embedded in a website can only search, never change anything.

How content becomes an answer

When you connect a source, the text of each page or file is pulled out, converted to Markdown, split into chunks along its headings, and written to the index. Documents move from queued to processing to indexed, and you can watch that happen in the console.

A search runs two rankers over the collection at once: a full-text ranker for exact words and a vector ranker for meaning. The two rankings are merged, and each hit comes back with the chunk that matched, its heading path such as Billing › Invoices › Refunds, and a link to the source. The retrieve endpoint returns the same matches as full passages with citation numbers, ready to paste into a prompt.

Where things live

Plan limits

Limits are enforced by the API. A request over the limit returns 429 with a message saying which limit and how much room is left. The console shows monthly search usage in the sidebar.

PlanData sourcesDocumentsSearches / month
Free23001,000
Team105,00050,000
BusinessUnlimited25,000500,000
Self-hostedUnlimitedUnlimitedUnlimited

A document is one page, file or record, counted once however many collections include it. Re-syncing a source is free. Searches from people and from agents count the same.