Every eChat agent starts the same way: a short setup wizard that walks you from nothing to a working AI moderator you can talk to. It takes you through four steps in order - Instructions (your webinar, a Role, and a starting knowledge-source URL), Training, Product, and Agent - and by the end your agent isn't just configured, it's active and ready to answer questions in your webinar's live chat.
You can open the wizard from inside a specific eWebinar webinar's settings, or from your eChat Your Agents list. Either way the steps are the same; the only difference is whether eChat needs to ask which webinar the agent belongs to. Nothing you set here is permanent - every choice in the wizard can be changed later from the agent's own editor.
Step 1: Instructions
What it is
The first step gathers the three things eChat needs to get started: which webinar the agent is for, what job it's doing, and one URL to start learning from.
How it works
Your webinar. Today, eChat's job is to answer questions inside an eWebinar webinar's live chat, so every agent needs to be linked to at least one webinar - you can't continue until one is selected. If you opened setup from inside a specific webinar's settings, that webinar is already chosen for you and you won't be asked. If you started from your Your Agents list, pick the webinar here.
Role. Choose what job this agent is doing: Sales, Customer Success, or Support (selected by default). Whichever you pick, eChat pre-writes a starting set of instructions in the agent's Instructions that match that goal, so you're not starting from a blank page - a Sales agent's default instructions nudge it toward next steps, a Support agent's toward clear troubleshooting. If you've already edited the instructions and then switch roles, eChat asks before replacing your edits.
- Sales - a webinar whose job is to answer product questions, handle objections, and move attendees toward a purchase or a booked call
- Customer Success - a webinar aimed at existing customers, focused on adoption, getting more value from the product, or renewing
- Support - a webinar or training session where attendees mostly need help troubleshooting or learning how to do something
Knowledge source. Paste in a single page - your main website, a product page, or a docs page. This is the same mechanism as the Web Pages training source you'll see in the next step; entering it here just gets the crawl started earlier, so the agent already has something to work with by the time you reach Training.
Best practices
Match the role to what this specific webinar is for, not what your team internally calls it. A demo that's really meant to close deals should get Sales even if you think of it as onboarding. If none of the three fits exactly, pick the closest one - the seeded instructions are a starting draft, not a constraint. For the URL, use the page that most closely matches what this webinar covers, not necessarily your homepage - a product page or a feature page usually teaches the agent more relevant detail.
One agent isn't limited to a single webinar: you can link the same agent to several webinars later - different session times, or a sales version and an onboarding version of the same product - and eChat automatically uses the right one, because each message tells the agent which webinar the visitor is watching. Keep one agent scoped to one product or topic, and create a separate agent per product if you sell more than one. See deploying an agent to your webinars for how linking works after setup.
Step 2: Training
What it is
The wizard walks you through five training sources in a fixed order, each with its own screen and a Continue to … button to move on. You can skip any of them and come back later - setup is meant to get your agent started, not finished.
How it works
- Web Pages - crawls a site you point it at, and re-crawls on a schedule you choose (weekly by default, every two weeks, monthly, or once). If you already entered a starting URL, that crawl is already running.
- Video Transcript - eWebinar-linked agents only. Syncs your webinar's transcript automatically. Under Answer behavior you choose whether the agent only answers what's played so far (Covered so far, the default) or draws on the whole transcript regardless of where the visitor is in the video (Answer anytime).
- Q&A - write question-and-answer pairs by hand, import a CSV, Excel, or JSON file, or import your eWebinar Saved Replies. Q&A is the most authoritative source - it overrides anything the agent would otherwise say.
- Conversations - import past conversations, either a file export or straight from your eWebinar chat archive. Conversations aren't training data: eChat turns them into suggested Q&A pairs, and only the ones you approve become training.
- Documents - upload PDFs, Word (.docx) files, text, Markdown, or CSVs, or add a text document directly in the dashboard.
Best practices
Don't try to fill in every source during setup - add whatever you have handy and move on. Suggestions from Saved Replies and Conversations need your review before they're used to answer visitors, so plan to check back in Training after setup rather than assuming it's done. For a deeper look at each source, see training your agent.
Step 3: Product
What it is
The Product step holds the product name, the agent's language, and an AI-generated summary of what the agent is talking about, built from whatever it has crawled or uploaded so far in setup.
How it works
Once at least one training source is ready, click Generate Product Summary. eChat reads through the training material gathered up to this point and writes a plain-English description of the product or topic. It's a quick sanity check that the agent picked up the right thing before you finish setup, and you can edit the text if something's off.
Best practices
If the summary reads generic or slightly off, that's usually a sign the training added so far is thin - go back a step and add a more specific page, or add a document, rather than pushing ahead. The summary reflects whatever training exists at that moment; you can regenerate it later as Video Transcript syncing and any additional pages finish processing.
Step 4: Agent, and going live
What it is
The last step, Agent, is where you give your agent an identity - a name and an avatar - and review its model, grounding, and escalation settings.
How it works
Enter a name and upload an avatar image - this is what visitors see attached to the agent's messages in chat. Click Complete setup to activate the agent. If you started setup from eWebinar, you're taken back to eWebinar; otherwise eChat opens Playground, where you can start talking to the agent immediately, even while training - crawls, transcript syncing, and so on - keeps processing in the background.
Best practices
Name and avatar it the way visitors would expect a host or team member to look, not a generic "Bot" - it sits in a live chat next to your human moderators. Use Playground right away to sanity-check tone against the role you picked earlier; if something's off, it's much faster to fix Instructions now than after visitors start seeing it. See the Agent Settings Feature Guide for everything on the Agent screen.
Setup is meant to get your agent talking, not finished. Come back to Training anytime to add more pages, approve suggested Q&A, or upload documents - each source has its own help article covering how to get the most out of it.
Trying it in Playground
What it is
Playground is your private test chat with the agent - the same grounding and training it will use live, without attendees watching. It's where you check tone, catch weak answers, and follow citations back into Training when something looks off.
How it works
Ask questions the way attendees would. When the agent grounds a reply on your training material, Playground (and other test modes) can show citations for the evidence it used. Citations are labeled by source type: Web pages, Q&A, Conversations, Documents, or Transcript.
Each citation - and the top retrieved evidence in test mode - can include a View in Training data link. Clicking it opens the matching Training tab, keeps your Playground conversation open, and scrolls to highlight that item so you can see exactly what the agent drew on:
- Web pages - opens the Web Pages source and expands or filters the URL list to the cited page
- Q&A - pins a published pair, or highlights one still in Needs review
- Conversations, Documents, and Transcript - open that Training tab and highlight the matching item
These links are for you in Playground and test mode only - real webinar attendees never see them. Occasional synthetic or system context may not have a link to follow.
A citation into Conversations opens that review UI so you can inspect the source of a suggestion - conversations themselves are not training data. Only approved Q&A becomes training.
Best practices
When a Playground answer feels off, follow View in Training data before rewriting Instructions - often the fix is approving, editing, or adding a Q&A pair, or trimming a Web Pages crawl, rather than changing the agent's personality. See Web Pages and Q&A for how those Training tabs work once you land there.