Agnost AI Launch HN Demo

Hey folks, I am Shubham, one of the co founders of Agnost AI.

This is a quick demo for our launch HN.

Agnost AI is product analytics for teams building chat and voice agents.

The core insight is that users do not give feedback directly but rather leave it inside the conversations that they have with your AI agents.

Here in the dashboard that you can see, The core insight here is that Auto Subtitles is the largest cluster forming .

So you can go to intents, we can see this new cluster that's getting created, we can click on this, we can find out what conversations getting are getting matched to it, we can see what the users are talking to it where already it happens.

You can also see these user metadata fields that you can pass in the SDK that is the name, what plan are they on, what region are they on, etc.

something that a lot of our users like which we recently shipped is the ability to add enriched fields like we can enrich users on your behalf based on the data that you pass and the public information available online.

Another thing is violations where you can actually see when the agent misbehaves in production and you can set rules for it so that we can flag it instantly.

And you can get Slack alerts or emails, whatever you want.

Like one of the most used ones is this one where you want to verify each and every link that goes out and is not an internal or an invalid link.

and you can, as I said, you can configure alerts like a daily Slack summary, get Slack notifications, or even email the team or somebody relevant for specific use cases.

Creating alerts is pretty simple, you just have to write that in natural language and we take care of it.

One thing that we're also testing now is this auto improve feature where we try to show suggestions that could change in your prompts or tool call description that can actually improve the agent.

we are very specific of opening Pull request which you can actually, once you connect to GitHub repository you can configure the product specific prompt and the dev specific prompt separately.

But we're very hesitant because we do not want to open any slop PRs.

We're trying to be as specific as we can.

So that's why you'll see a very small number of PRs but those will be usually high confidence.

So yeah, that's the product.

You can apply multiple filters, you can yeah just copy the everything as a, as a markdown.

We are still building the Ask AI so yeah, would love your feedback and know what do you hate about this.

Thank you.