Knowledge Base
The knowledge base extends your AI assistant's knowledge beyond the video transcript. By uploading documents, crawling web pages, or adding text, you give the AI more sources to draw from when answering viewer questions.
Why Use a Knowledge Base
Your video transcript is the primary knowledge source, but it can't cover everything. A 30-minute product demo might cover the main workflow but skip edge cases. A knowledge base lets you fill those gaps.
For example, upload your FAQ document and the AI can answer "what payment methods do you support?" even if the video only shows a Stripe checkout. Upload your API documentation and the AI can respond to developer questions that the video doesn't address.
Document Sources
File uploads — Upload PDFs, DOCX files, or other document formats. Keep'em extracts the text, chunks it into logical sections, and indexes it for AI search.
URL crawling — Provide a URL and Keep'em crawls the page (and optionally linked pages) to index the content. Useful for help centers, blog posts, or any web-based documentation. The number of pages per crawl depends on your plan.
Plain text — Paste content directly into the knowledge base. Useful for quick additions like short FAQ entries or product specs that don't exist as standalone documents.
How It Works
Documents go through a similar pipeline to the video transcript: they're chunked into logical sections (splitting on headings and natural boundaries), converted into vector embeddings, and stored alongside your transcript chunks.
When a viewer asks a question, the AI searches across all sources — video transcript chunks and document chunks — and uses the most relevant matches to generate its response.
Knowledge Base Limits
| Plan | Words per Event | URL Crawl Pages |
|---|---|---|
| Trial | 50,000 | None |
| Starter | 100,000 | 25 |
| Growth | 500,000 | 100 |
| Scale | 2,000,000 | 500 |
Best Practices
Start with your FAQ. If you have an existing FAQ or help center, that's the highest-value addition. These are the questions people actually ask.
Add product documentation. Technical details that would be impractical to cover in a video — API endpoints, configuration options, supported integrations — make your AI dramatically more useful.
Review chat insights. After your event has been live for a while, check the Top Questions analytics. If viewers are asking questions the AI can't answer, add documentation that covers those topics.