Documentation

• Agent (AI): one assistant you build and train. Each lives at its own Training Studio.
• Training Studio: where you work on an agent. Two panels: Test your AI (chat like a customer) and Train while you test (explain fixes in plain English).
• Knowledge: the single pool your agent answers from, built from a website, individual pages, files, images, plain text, or Q&A pairs.
• Deploy: where you design the widget, preview it, and copy the install script.
• Chats: the history of real conversations your deployed widget has, with feedback and outcomes.
• Credits: a monthly, workspace-wide allowance used as your AI answers.
• Workspace & seats: plans belong to a workspace; seats are the teammates you can invite into it.

On the homepage, enter your website address and, optionally, a line about what you want it to help with. One-tap starters (support questions, lead capture, explaining your services, product recommendations) seed those instructions for you.

Trainly begins reading your site and takes you into the Training Studio, where you’ll add knowledge, test, and refine. If you’re signed out, you’ll be asked to sign in first, and your input is kept while you do.

In the Training Studio, use the + menu to add knowledge:

• Add website: enter a domain and Trainly reads it top to bottom in the background, a batch of pages at a time, until your site is in knowledge. It keeps reading even if you switch tabs, and only fetches public pages with sensible page, time, and size limits. Paste a specific page URL instead and Trainly reads exactly that one page (exact-page mode).
• Upload files: PDF, TXT, MD, PNG, JPG/JPEG, and WEBP. Images and PDFs are read with vision so text inside them is understood.
• Text & Q&A: on the Knowledge page, paste in plain text or add question-and-answer pairs for things that aren’t written on your site.

You can turn any source off. It stays in your library but is withheld from answers. Knowledge is a combined pool per workspace, capped by your plan (see Plans & credits).

In Test your AI, ask a question the way a real customer would. The answer streams in with a Retest option under it.

To correct something, click the answer and describe the change in plain English in Train your AI. Trainly saves it as a rule, a fact, or an approved reply, and applies it the moment you retest, replacing the old exchange. Use Run simulation to have Trainly play a realistic customer and surface gaps for you, one message at a time.

When the AI isn’t sure, it says so instead of guessing. Those moments are flagged so you can teach it the answer.

In the Deploy tab, pick a layout: Chat bubble (a floating launcher with a compact popup) or Sidebar (a full-height drawer). Then style it: logo, brand color, background, assistant name, subtitle, greeting, font, size, and corner radius, all with a live preview on desktop and mobile.

When you’re happy, copy the install script and paste it once, just before your closing </body> tag. It looks like this:

Copy the exact snippet (with your agent’s id) from the Deploy tab. It works with any website and needs no rebuild, and the widget renders inside a shadow root so your site’s styles and the widget’s never affect each other. Change the design anytime in Deploy and the same script picks it up on the next load.

Every deployed conversation is saved in the Chats tab, with thumbs-up/down feedback on answers and an outcome for each chat (resolved, or the customer left). Disliked answers show you exactly where to improve. Chats itself is read-only: open any answer in the Training Studio to talk through the change, then retest.

• Honesty: on low-confidence questions the AI prefers “I’m not sure” over inventing an answer.
• Languages: it auto-detects the customer’s language and replies in it, falling back to English.

Plans belong to a workspace, so one account can run several workspaces on different plans. Pricing is monthly, and plan changes are handled by us, so contact us and we’ll set your workspace up.

Credits are a monthly allowance shared by all agents in a workspace. They’re used whenever your AI does work: answering questions (in testing and live), reading and indexing the content you add, running a simulated customer, and regenerating suggested tests after you save new training. What a single action costs varies with the model, the amount of context, and the response length, so credits aren’t a fixed number of messages. They reset monthly on your workspace’s billing day and don’t roll over. Knowledge and seats are shared per workspace; the owner doesn’t count as a seat.

Trainly reads the public pages of the website you point it at and the files you upload, and stores the compiled knowledge for each agent plus the conversations your deployed widget has, so you can review and improve them. Answers are generated with OpenAI’s GPT-4.1-mini by default. For the full details on what’s collected and how it’s handled, see the Privacy Policy and Terms.

Have a question this didn’t cover? The FAQ goes deeper on common topics, and you can always contact us. We read every message.