Session replay for browser agents

Tracevell traces LLM calls via auto-instrumentation, and records browser sessions via native integrations with common browser automation frameworks:

In the Tracevell trace view, you can scrub the recording and correlate moments in the browser with the spans that produced them.

Getting started

If you don’t have a project API key, you can get one by signing up on Tracevell or spinning up a self-hosted instance and getting a key from the project settings.

Follow the integration guide for your framework to enable tracing + session replay:

Privacy: masking inputs

If your agent interacts with sensitive forms, configure session replay input masking so typed values aren’t recorded.

TypeScript
Python

import { Laminar } from '@lmnr-ai/lmnr';

Laminar.initialize({
  projectApiKey: process.env.LMNR_PROJECT_API_KEY,
  sessionRecordingOptions: {
    maskInputOptions: {
      text: true,
      textarea: true,
      email: true,
      tel: true,
      number: true,
    },
  },
});

See also: SessionRecordingOptions and Laminar.initialize.

from lmnr import Laminar

Laminar.initialize(
    project_api_key="...",
    session_recording_options={
        "mask_input_options": {
            "text": True,
            "textarea": True,
            "email": True,
            "tel": True,
            "number": True,
        }
    },
)

See also: SessionRecordingOptions and Laminar.initialize.

What it looks like

When you open a trace, you’ll see a session replay panel below the timeline. Scrubbing the recording keeps the trace timeline in sync, so you can jump between what happened in the browser and the spans that produced it.

This works similarly for other supported browser-agent frameworks (Stagehand, Playwright, Puppeteer, Skyvern) when instrumented via their Tracevell integrations.

Overview

Tracing

Signals

Evaluations

Debugger

Datasets

Platform

Session replay for browser agents

Overview