Skip to main content

Overview

Browserbase integrates with Prime Intellect through BrowserEnv to provide the infrastructure for evaluating and training state-of-the-art browser agents. BrowserEnv connects Prime’s evaluation and RL training pipelines to Browserbase’s cloud browsers, giving your agents real browser sessions to interact with — no local browser setup required. Use it to benchmark agent performance on web tasks, generate reward signals for reinforcement learning, and train models that can navigate, extract, and act on live web pages. How it works:
BrowserEnv supports two modes:

DOM Mode

The agent receives structured DOM content (HTML/accessibility tree) and issues high-level instructions. Recommended default for most tasks.

CUA Mode

The agent receives screenshots and uses coordinate-based actions (click, type, scroll). Best for pixel-grounded or visual tasks.

When You’d Use This

  • Benchmark browser agents — Run standardized browser tasks against your model and collect reward signals using prime eval.
  • Train browser-capable models with RL — Use Browserbase sessions as the environment in Prime’s Hosted Training or self-managed prime-rl pipelines.
  • Run browser tasks in Prime’s environment workflow — Plug BrowserEnv into any verifiers-compatible evaluation or training loop.

How It Fits Together

ComponentRole
BrowserbaseCloud browser infrastructure — creates and manages browser sessions via API
verifiers BrowserEnvEnvironment abstraction that wraps Browserbase sessions for evaluation and training
Prime CLI / Hosted TrainingOrchestrates evaluation runs (prime eval) and RL training loops
Browserbase handles the browsers. BrowserEnv translates between Browserbase sessions and the verifiers environment interface. Prime orchestrates the workflow.

Guides

Evaluate Browser Agents

Run browser agent evaluations with prime eval.

Train Browser Agents

Wire BrowserEnv into Prime RL training workflows.