> For the complete documentation index, see [llms.txt](https://docs.supademo.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.supademo.com/ai-demo-agent/agent-training.md). # Agent Training Train your Supademo AI Demo Agent on demos, videos, documents, links, and websites, then verify coverage with readiness scoring and knowledge gap insights. Agent training is the process of building your agent's **Knowledge Base** — the content library it draws from to answer questions and make recommendations. You train a Supademo AI Demo Agent by adding six source types (interactive demos, videos, documents, raw text, individual links, and crawled websites), then verifying coverage with the Readiness Check and closing holes surfaced by Knowledge Gaps. The more relevant, high-quality content you add, the better your agent performs.
*** ## Indexing Pipeline Content goes through an **indexing pipeline** after being added: text is extracted, chunked into segments, converted to embeddings, and stored in a vector database for fast retrieval. You'll see the index status for each source: | Status | Meaning | | ------------ | ----------------------------------------------- | | **Pending** | Queued for indexing. | | **Indexing** | Currently being processed. | | **Indexed** | Ready for the agent to use. | | **Failed** | Indexing encountered an error. Try re-indexing. | All Knowledge Base tabs support **bulk operations**: select multiple sources to reindex or remove them at once. You can also **reindex stale** content that may be outdated. *** ## Knowledge Source Types ### Demos Use this tab to add interactive Supademo demos from your workspace. These are the agent's primary asset — when a visitor asks about a feature, the agent can recommend and present the relevant demo directly in the chat. This allows users to click and explore different demos at their own pace and learn by doing. After clicking **Add**, easily search, filter, and multi-select demos and folders to add to your Agent's knowledge base. You can also add optional hints next to these demos to ensure they're referenced/pulled up at the right context within the conversation.
{% hint style="info" %} **Tips:** * Add demos that show complete workflows, not isolated clicks. * Name demos descriptively so the agent can match them to visitor questions. * Cover your most common use cases and feature areas. * Build demos specific to the agent use case (i.e. top-of-funnel sales, marketing) instead of reusing demos you've already created for other purposes (i.e. support docs) * Start with a focused set of interactive demos (i.e. 10-15 core demos) instead of demos that are too granular or enablement/CS/training-focused. {% endhint %} ### Videos Add video walkthroughs and tutorials. The agent recommends relevant videos when visitors want to see your product in action. These complement interactive demos by providing narrated context. For videos, you can add links from: * Your Supademo workspace (previously recorded screen recordings) * Youtube (by entering in a URL) * Loom (by entering in a URL)
### Documents Upload PDFs, slide decks, case studies, and other documents. The agent uses these to answer pricing, feature comparison, and technical questions accurately. Supported types include PDF and other document formats. ### Links and Websites You can add web content two ways: paste **individual links** to index specific pages one at a time, or crawl an entire **website**. For full-site crawls, Supademo uses Firecrawl to automatically crawl pages and index their content. This keeps your agent current with your latest pricing, features, and FAQs. After adding a website URL, the crawler discovers and indexes individual pages as child sources. You can drill into a website source to see all discovered pages, their index status, and manage them individually (reindex or remove specific pages). {% hint style="info" %} **Tips:** * Crawl **only specific pages on your marketing site** to avoid outdated content or hallucinations * Re-crawl periodically to pick up important content updates * Keep the content here focused on the Agent's use case (i.e. don't add all of your docs pages for a high-level sales use case for the AI Agent) {% endhint %} ### Text Add custom knowledge snippets — FAQs, talking points, objection handling scripts, or any information that doesn't live in a document or on your website. Text sources are **editable inline**, so you can revise a snippet directly in the Knowledge Base without deleting and re-adding it. This makes Text the fastest way to fill specific knowledge gaps identified in the Knowledge Gaps section. ### Third-Party Product Sources Sources can be **marked as showing a third-party product** — useful when a demo or video features a partner tool, an integration, or another product in your workflow rather than your own. Marking the source tells the agent the content isn't your product, so it presents and attributes it correctly in conversation. ## Content Hints {% hint style="success" %} **Content hints** are available on every content type in the knowledge base: demos, videos, PDFs, documents, and website pages. When adding or editing any asset, click "Add Hint" to tell the agent exactly when and why to recommend it. For example, hint a security whitepaper with "Surface when the prospect asks about compliance, SOC 2, or data privacy" so the agent pulls it up at the right moment instead of guessing from the title alone. Adding hints across all your assets can improve recommendation accuracy significantly. {% endhint %} *** ## Agent Readiness Check Before publishing, check your agent's **Readiness Score** to evaluate whether it has enough source material to answer common prospect questions. The readiness check scores your knowledge base against a set of common buyer and onboarding questions for your agent's use case, so you know what's covered before a real visitor asks. Click **Check readiness** in the right-side panel (or via the readiness card). The evaluation takes 15-30 seconds and grades your agent across categories of common questions.
### Score Interpretation | Score | Rating | Meaning | | ------------- | ---------- | --------------------------------------------------------------------------- | | **80-100%** | Strong | Strong source coverage. Your agent is well-prepared. | | **50-79%** | Moderate | Some source gaps remain. Consider adding more content for uncovered topics. | | **Below 50%** | Needs work | Significant gaps. The agent will struggle with many visitor questions. | ### What's Evaluated The readiness check tests whether your knowledge base can answer typical prospect questions across multiple categories. Each question is graded as: * **Full** — The agent has strong content to answer this. * **Partial** — The agent has some relevant content but may not fully answer. * **Fail** — No relevant content found. Use the detailed breakdown to identify exactly what content to add. Each failed or partial question includes an explanation and, where applicable, whether a recommendable asset exists. *** ## Evaluation Criteria and Conversation Triggers Beyond knowledge, you can train your agent on *judgment*: what to listen for in a conversation and what to do when it hears it. **Evaluation criteria** define what the agent should assess as each conversation unfolds — for example, whether the visitor matches your target profile, what problem they're trying to solve, or how far along they are in evaluating tools. The agent scores conversations against these criteria continuously, and the results feed the intent scoring you see in [Agent Analytics](https://docs.supademo.com/ai-demo-agent/agent-analytics). **Conversation triggers** are rules that fire when a criterion is met mid-conversation, so the agent acts at the right moment instead of waiting for the visitor to ask. Triggers can be enabled and disabled individually as you tune the agent's behavior. *** ## Qualification Definitions **Qualification definitions** tell the agent what a qualified buyer looks like for your business — the signals that separate a high-intent prospect from a casual browser. Define qualification criteria once, and the agent applies them in every conversation: asking the discovery questions needed to evaluate them, marking sessions as qualified when the definition is met, and pairing qualification with **triggers** so a qualified visitor is immediately routed to the right next step, such as a booking link or sales handoff. Qualified sessions surface in the Sessions table with their intent score, so your sales team can prioritize follow-up. See Agent Analytics for how qualification results appear in reporting, and [Capture Leads](https://docs.supademo.com/capture-leads) for collecting visitor contact details. *** ## Close the Loop with Knowledge Gaps Training doesn't end at publish. After each conversation, Supademo automatically detects questions your agent struggled to answer and groups them into **Knowledge Gaps** — a continuous feedback loop between what visitors ask and what your knowledge base covers. Each gap includes a severity, a recommendation (create a demo, upload a document, or add text context), and a CTA that takes you straight to the right Knowledge Base tab. Review gaps weekly, add the missing content, and watch your readiness score climb. See Knowledge Gaps in Agent Analytics for the full workflow. *** Next: [Publishing & Sharing](https://docs.supademo.com/ai-demo-agent/publishing-and-sharing) — go live and share your agent.