Frigade
Product growth and onboarding for modern software companies
Start new thread

Frigade Assist API - Give your AI agent guides that show users where to click

by
•
Most in-app AI agents answer "how do I do this?" with a wall of text, because they can't see the screen. The Frigade Assist API is one tool call your existing agent makes. Frigade draws a step-by-step guide right on the page, hands the agent a plain answer when it can't, and tells it what the user is looking at. It knows your product because browser agents used it first, and it relearns every release. Works with the Vercel AI SDK or anything that can call a tool.

Add a comment

Replies

Best
Maker
📌
Hey Product Hunt, Christian here, CTO and co-founder of Frigade. Frigade started as a drop-in AI assistant that lives inside your product and learns it by using it. Over the last year, more and more of the teams we talked to had already shipped their own in-app AI assistants - however, they kept hitting the same wall. While most agents do a decent job making tool calls and answering basic product questions, they have zero context on the user's state in the app - so when a user asks "how do I do X?" they get back a long list of bullets regurgitated from an outdated Help Center. So we pulled the part of Frigade that draws guides on the page out into its own API. Your agent registers one tool. When a user asks how to do something, Frigade renders a step-by-step guide on top of your UI. If it can't build one, it hands your agent a plain text answer to relay, and if it can't help at all, it says so. It also tells the agent what the user is currently looking at, so "what does this error mean?" actually has a "this". The part I'm most proud of is how it knows the product. You give it a test account, and our browser-based agent logs in and builds its own map of how the app works. That re-runs on a schedule, so when you move a button or launch a new feature, we're always up to date. Would love to hear how you're connecting your agent to the rest of your app!

How well does the Assist API handle changes in the UI after a product release? Does it automatically adapt when buttons or flow moves?

Maker

 yes, it automatically detects updates in the UI on a schedule. See more here

Hey Christian! love the idea and wish you all the best here! Congrats on the lunch

Maker

 thank you!

When making things like this, the biggest headache is that the selector breaks when the UI changes—how do you handle that?

The detail I like most is the browser agent that logs into a test account and remaps your product on a schedule — that fixes the stale-help-center problem at the root. Curious how the map handles things the test account can't see, like feature flags or higher plan tiers — do you run multiple accounts, or accept some blind spots? Congrats on the launch, Christian and team.

the failure i would want to know about is the stale guide.

pointing at a button is a claim about the ui, and uis move. when a text answer goes stale you can usually tell. a stale pointer just confidently highlights the wrong thing and the user does what they were told.

the show me where framing is right, for what it is worth. reading instructions and then hunting for the element is where people give up.

so does assist know when its own guide has drifted. deriving the target live from the dom every time is a completely different product from a stored selector, and i would say which one right on the page, because anyone who has built this internally asks inside ten seconds.

The part I like most is that the agent can guide users directly on the screen instead of just explaining steps. That could make onboarding much easier. 👀

I really like the idea of combining AI support with actual in-app guidance. Sometimes a written answer is technically correct but still leaves you wondering where to click.

Giving the agent a tool call that draws the actual on-page guide instead of dumping a wall of text is a smart way to close the "how do I do this" gap. How does it decide when to fall back to a plain text answer versus drawing the step-by-step overlay - is that the agent's call or does Frigade make that decision based on whether it has a matching workflow recorded?