Skip to main content

Knowledge Curator

Not everything that enters the system deserves to be in the knowledge graph. Web pages contain navigation menus, cookie banners, and ads. Workflow results include debug output and status messages. Chat transcripts have small talk alongside substance.

The Knowledge Curator is the gatekeeper. It evaluates incoming content, keeps what's substantive, discards the noise, and routes the rest to the right place in your project's document library before it's classified into the knowledge graph.

What the Curator Does​

When the system captures content on your behalf — a web page an agent reads, the output of a tool, a chat transcript — the Curator decides what happens to it:

  • Filter — boilerplate, navigation, ads, cookie notices, and trivially short or duplicate content are dropped. Only substantive prose continues.
  • Extract — the author's original wording is preserved; the web chrome around it is stripped. No summarizing, no paraphrasing.
  • Route — content that arrives without a folder of its own is placed in the most specific matching folder in your library, or a new one is proposed when nothing fits.
  • Name its domain — the content is given a short, meaningful topic name that becomes its grouping in the knowledge graph.

Clean, structured content you upload yourself skips this step — you've already chosen where it lives, and there's no boilerplate to strip.

Describe Your Folders to Anchor Topics​

When you give a folder a description, that folder becomes an authoritative topic: content placed inside it is anchored to the folder's declared subject rather than re-guessed file by file. Describing your folders is the most reliable way to keep related content grouped under one stable topic in the graph.

Upload Flow​

You upload documents through the document browser — navigate to a folder, then drop files (or click to upload) directly into it. You choose the destination; the Curator does not suggest a folder for direct uploads.

Because you place the file yourself, the upload is immediate. If the project has Long-Term Memory enabled, the document is still enriched passively in the background — it's classified and woven into the knowledge graph without you having to wait. Automatic folder routing only applies to content the system captures for you (web pages, tool output), which has no folder of its own.

Images Are Knowledge, Too​

Images and illustrations aren't discarded when a document is processed — they're treated as knowledge carriers. A diagram, chart, or table is described in plain language, classified, and placed inline at its exact position in the document, so a chart showing quarterly revenue or an architecture diagram with labelled components becomes searchable, navigable knowledge rather than a binary blob agents can't reason about.