Knowledge sources

Everything an Appricity agent says comes from your knowledge: the documents, transcripts and web pages you give it. Appricity breaks each source into passages, indexes them for both meaning and keywords, and retrieves the most relevant ones for every question.

What can I add?

• Files: PDF, Word (.docx and .doc), PowerPoint (.pptx and .ppt), Markdown, HTML and plain text, up to 50 MB each by default.
• Pasted text: a document body or a call transcript. Transcripts of product demos can be marked as demo recordings, which splits them by topic and makes them available as style examples.
• Web pages: a single page, or a crawl that follows links on the same site from a starting page (up to 500 pages and 6 links deep; 50 pages and 3 deep by default). A crawl stays under its starting page's path: start at your documentation and only documentation pages are read; start at your home page to read the whole site. You can change a crawl's page limit, depth and skipped paths at any time from the source's page; they apply on the next re-ingest or scheduled refresh.

PDFs need to contain real text. A scanned, image-only PDF can't be read; export a text version instead.

Which screen should I use?

• Documents (in Work) is the simple way in, open to every user: drop files in, watch them go from Reading… to Ready to answer from, and remove them when they're out of date.
• Sources (in Knowledge) is the full workbench for admins: files and whole folders, pasted text, web pages and crawls, with a language per source, scopes, release binding, the passage view and re-ingest.
• Ingest (in Knowledge) is for bulk: up to 100 files or 100 web addresses in one batch, with duplicate detection and optional tag suggestions.

How do I know a source is ready?

Each source shows its progress: pending, then processing steps (reading, indexing and so on), then ready. Only ready sources are used in answers. If a source fails, its detail page shows why, and Re-ingest runs it again from the start.

What does bulk import do about duplicates?

Before a batch starts you can Check for duplicates. A file whose content exactly matches a source you already have is flagged, and you choose per file: Skip (the default), Replace existing, or Create new anyway. Ingestion starts once every duplicate has a decision.

With Suggest entity tags switched on, Appricity proposes which product-graph entities each passage is about. You approve or reject the suggestions on the batch page; nothing is tagged until you approve it.

Can I see what the brain extracted?

Yes. A source's detail page lists its passages, twenty at a time, and for a crawl the pages it visited. From there you can tag passages with entities from your product graph.

What happens when I delete a source?

It stops being used in answers straight away.

Does anything else happen when I add knowledge?

Yes. Shortly after new sources are ready, Appricity refreshes your company profile from them, without overwriting anything you've edited by hand.

Can I keep knowledge for one team separate?

Yes. In bulk import you can assign files to a scope, a partition inside the workspace. Sources without a scope are available to everyone in the workspace.