ReflexWare

Knowledge Base

Ingest documents from URLs or text, run semantic search, and ground your AI assistant.

The Knowledge Base is where you build the material your AI assistant draws on to answer visitors accurately — pages from your website, help articles, or any text you paste in. Once something is indexed here, your chat widget's AI assistant can search it and ground its answers in your actual content instead of guessing. This page explains how to add documents, search them yourself, and manage what's indexed.

Who this is for

Anyone on your team who wants the AI assistant to answer from your organization's own content — usually whoever owns your help docs, FAQs, or product content. Adding and searching documents doesn't require a special permission beyond being a member of your organization.

Where to find it

Select Knowledge Base in the left menu (under the same area as Storage and Domains).

Adding documents

Click Add Document to open the add-document dialog, which has three ways to bring in content:

1. Add URL

  1. Select the Add URL tab.
  2. Paste a public URL — for example, a help article or product page.
  3. Optionally give it a title; if you leave it blank, the page's own title is used.
  4. Select Ingest URL.

ReflexWare reads the page, converts it to text, and indexes it. This can take a few seconds depending on the size of the page.

2. Add Text

  1. Select the Add Text tab.
  2. Enter a title and paste the text you want indexed.
  3. Select Ingest Text.

Use this for content that isn't already published on the web — internal notes, a policy you've written up, or anything you'd rather paste directly than link to.

3. Batch URLs

  1. Select the Batch URLs tab.
  2. Paste one URL per line (up to 100 — extra lines beyond that are ignored).
  3. Optionally add a title prefix that's applied to every document in the batch.
  4. Select Queue [n] URLs.

Unlike the single Add URL tab, a batch is queued as a background job rather than ingested immediately, so you can submit a large list and keep working while it processes.

How ingested content becomes searchable

Whichever way you add a document, ReflexWare breaks it into smaller overlapping chunks, generates an embedding (a numeric representation of meaning) for each chunk, and stores them for semantic search. This means your AI assistant — and the search box on this page — can find the most relevant passage even if a visitor's question doesn't use the same words as your source content.

Managing your documents

The Documents table lists everything you've added, with:

  • Title — the document's title, with a link to its source URL if it came from one.
  • Source — URL, Text, or File.
  • Status — PENDING while it's being processed, INDEXED once it's searchable, or FAILED if something went wrong (hover the row for the error).
  • Chunks — how many pieces the document was split into.
  • Created — when it was added.

Select the trash icon on a row to delete a document. Deleting removes the document and everything indexed from it — this can't be undone.

Searching your knowledge base

Use the Semantic Search box at the top of the page to try out what your AI assistant would find. Type a question or a few keywords and select Search. Each result shows a match percentage and the matching chunk of text, ranked by relevance — this is the same search your chat widget's AI assistant uses behind the scenes when it answers a visitor.

How this connects to your chat widget

Adding documents here doesn't automatically change how your chat widget answers — you also need to turn on Ground answers in knowledge base in the widget's AI Assistant settings. Once that's on, the assistant searches your knowledge base before answering and cites the sources it used. See Chat Widget → Knowledge-base grounding with citations.

Reader sources

The Reader sources panel at the bottom of the page is a diagnostic view of the channels ReflexWare can use to read a URL (for example, a standard fetch vs. a headless-browser fallback for JavaScript-heavy pages). You generally don't need to act on this — it's there so you can see why a particular URL might have failed to ingest.

Browser tab

The Browser tab runs an isolated browser job: give it a task and an optional start URL, and it returns extracted text (and screenshots when a real browser is configured). The same engine powers the Browser Agent automation action. Runs are listed on this tab so you can inspect output without opening Automations. Failed or finished runs have Run again. In-progress runs have Cancel. Export CSV downloads the recent run list. If the job was started from an automation or the lead Browser button, the extract is stored on that lead — the run row links to that lead. Open the lead to see the latest extract on the Details tab (with a link back to this run, and Cancel while it is in progress). The lead's Activity tab also records each finished job and links back here. You can open a specific run from /knowledge?tab=browser&run=….

If the form is disabled, an admin still needs to turn the feature on on the server.

Troubleshooting

  • Ingesting a URL or text fails immediately. Your organization's knowledge base feature may not be turned on yet — ask whoever manages your ReflexWare deployment to confirm it's enabled.
  • A document shows FAILED. Hover the status badge to see the error. A common cause for URLs is a page that blocks automated readers; try pasting the content directly with Add Text instead.
  • Search returns no results. Make sure you have at least one document with an INDEXED status — documents still PENDING or FAILED aren't searchable yet.
  • The AI assistant isn't using my knowledge base. Confirm Ground answers in knowledge base is turned on for that chat widget. See Chat Widget.