Train, build, and deploy
browser agentsāfast
The easiest way to build reliable AI agents that actually understand the web.
Most browser agents wipe out before they catch a wave
Agents wipe out on real sitesāDOM changes break everything like a rogue wave.
Training is slow and brittleāyou're paddling against the current with no good datasets.
Debugging is a nightmareāblack box failures leave you lost at sea with zero visibility.
Everything you need to ride the web
SDK + datasets + eval harness + monitoring. The complete lineup for browser agents.
š SDK
Your surfboardāone SDK to ride any browser framework. Works with Playwright, Puppeteer, or bring your own board.
š Training Data
Pre-charted surf spotsāclean datasets of real tasks, ready to train your agents on the perfect waves.
š Eval Harness
Know your lineupābenchmark agent performance and compare different approaches to find your sweet spot.
ā” Monitoring
Live surf reportāreal-time dashboards for success rates, costs, and failures. Catch issues before they wipe you out.
Drop in with one line
No complex setup. No brittle selectors. Just agents that flow.
import { SurfsAgent } from '@surfs/agent'
// Grab your board and hit the water š
const agent = new SurfsAgent({
model: "gpt-4",
trackMetrics: true
})
// Catch the waveāautomatic instrumentation built in
await agent.ride({
task: "Find and add the cheapest red t-shirt to cart",
url: "https://shop.example.com"
})
// Full visibilityātraces, replays, metrics all logged šFull observability built-in. Ride smoothly, no manual logging required.
Built for the big waves
Real-time surf reports, auto-recovery from wipeouts, and instant replay debugging.
run_a3f9e2b8 Ā· 47.3s Ā· 12 steps
goto("https://shop.example.com")click("text=Search")fill("input[name=q]")click("text=Add to cart")Latest Resources
Stay updated with the latest insights, guides, and best practices for AI-powered development and testing.
Hybrid OnāDevice + Cloud LLM for Agentic Browsers: WebGPU/WASM Inference, Speculative Plans, and PolicyāGated Escalation
Design a splitābrain agentic browser: a compact WebGPU/WASM model handles perception and affordances onādevice; a cloud LLM performs hard planning. Stream tool calls, cache and patch plans, and enforce privacy via policyāgated escalation with attested telemetry.
Federated Learning for Browser Agents: OnāDevice FineāTuning with Differential Privacy and WebGPU
A practical, end-to-end design for federated learning in the browser: fine-tune task adapters on-device with WebGPU, enforce DP-SGD and per-user privacy budgets, coordinate FL rounds via Service Workers, defend against dropouts and adversaries, and ship robust updates without exfiltrating PII.
Speculative Execution for Browser Agents: Parallel Plans, Prefetch, and Rollback with CDP and Deterministic Snapshots
Design a speculative AI browser agent that forks branches in parallel tabs, prefetches and shares cache, then commits the best path. Covers CDP multi-sessions, snapshot-based rollback, write-gating, branch predictors, cost control, and evaluation.
Looking for PR-focused browser testing?
Try DebuggAI PR Copilot