OpenUI ObservabilityUser analytics
for AI agents

See what users saw and did, where your agent fell short, and what to build next.Free during early access

You’re on the waitlist.

See what users saw and did, where your agent fell short, and what to build next.Free during early access

You’re on the waitlist.

A successful trace
can still fail the user.

Tracing was built for services, where requests either succeed or fail. It shows what your agent did, not whether the user’s goal was met.

Execution : Successful

Traces can tell youWhat the agent did

The request succeeded
No errors occurred
The model and tools worked
The system behaved as expected

Outcome : Failed

But traces can’t tell youWhat the user experienced

Did the user get what they needed?
Did the experience actually work?
Why did the user retry or leave?
What needs to improve for the user?

Session replay

Understand every
user session

Replay the exact interface, responses, and interactions each user experienced.

Triage

Discover sessions
worth investigating

Surface failed journeys, unmet needs, and high-impact issues to understand where users struggle.

Annotations

Turn bad responses
into feedback

Mark the broken response, add context, and share the session with your team.

Evals

Prevent the same
issue twice

Turn production failures into evals that catch regressions before release.

Insights

Discover what to
build next

Aggregate demand across every session to see which needs recur and how often.

Data controls

Keep responses in the browser and capture only the metadata you need.

Your data stays yours

Your data stays private and is never used to train models. Deploy in your VPC or self-host.

Compliance

Find GDPR, SOC 2, and ISO 27001 evidence in the Trust centre.

Set up in two lines

Keep your agent framework and backend. The SDK captures events OpenUI already emits as it renders.

1
Generate a frontend API key
Create a publishable key in the OpenUI console. It’s safe to use in browser code.
2
Initialize the SDK
Add it once at your app entry point. No tracking code needed in your components.

// src/main.tsximport { createRoot } from "react-dom/client";import { App } from "./App";import * as Observability from "@openuidev/observability-cloud";
Observability.init({ apiKey: "pk-th-…" });
const root = document.getElementById("root");createRoot(root!).render(<App />);

Free during early access.

See your agent through your users’ eyes

You’re on the waitlist.

View pricing

Frequently asked questions

Have another question?
Join our Discord.

How is OpenUI Observability different from tracing tools?

Tracing tools show how your agent ran. OpenUI Observability shows what users experienced, including generated UI, interactions, errors, corrections, and fallbacks.

They work together. Export flagged responses as eval cases to tools such as Braintrust or LangSmith.

What does the SDK capture?

The SDK captures events from OpenUI when rendering settles, including timestamps, update counts, error counts, and error details such as code, source, and component.

Full capture includes the generated response and is enabled by default. Minimal capture excludes response content and error messages.

Can I control what data is sent?

Yes. Choose minimal capture to send metadata without generated responses. Use the beforeSend hook to edit or drop events before they’re sent to OpenUI Observability.

Is the frontend API key safe to expose?

Yes. Frontend API keys are publishable and designed for use in browser code, like an analytics key.

What does OpenUI Observability cost?

OpenUI Observability is free during early access. Pricing will be published before paid access begins.

Do I need OpenUI Gateway?

No. OpenUI Observability and OpenUI Gateway work independently.

We recommend using both. Gateway makes generated UI more reliable. Observability shows what users experienced and whether it worked.

OpenUI ObservabilityUser analytics
for AI agents