Skip to main content
Webcompute browser work moving through managed agent infrastructure Webcompute gives agents and applications a managed browser runtime. Start with a delegated browser task, switch to exact Playwright when code should own the steps, and inspect the live browser, recordings, events, downloads, and artifacts when evidence matters.

​
Start from the path that matches the job

If you need toStart withWhat you get
Prove a browser task from the terminalweb agentA delegated run, final answer, optional JSON, and live inspection while the run is active.
Embed delegated browser work in an appSDK web.agent()Model-backed browsing with policy, approvals, structured output, streaming, and artifacts.
Own exact browser stepsSDK browser runtime or CLI browser commandsDeterministic Playwright control against the same managed browser runtime.
Give a coding agent a browserMCP runtime integrationManaged sessions, bounded observations, status, and cleanup.
Capture one URLQuick actionsScrape, screenshot, or PDF without owning browser lifecycle.
Attach an existing automation stackCDP URLProtocol-level interop when your framework must drive the browser directly.
web agent \
  --url https://www.sec.gov/edgar/search/ \
  --allow-domain sec.gov \
  "Find Apple's latest 10-Q filing. Return filing date, accession number, filing URL, and a one-sentence summary."
Example result: 
Apple Inc. latest 10-Q filing

Form: 10-Q
Filing date: <date from SEC EDGAR>
Accession: <accession number>
URL: https://www.sec.gov/Archives/...
Summary: ...
Evidence: SEC EDGAR company filing page
Webcompute Debug UI showing a live managed browser on example.com, recording status, browser readiness, active page metadata, and the event timeline The same runtime also runs in inspection mode: managed browser, active page, recording state, runtime readiness, and reviewable evidence.

Run your first browser agent

Install the CLI, sign in, configure a model, and run a real browser task.

Choose your path

Pick CLI agent, SDK agent, direct Playwright, MCP, REST, CDP, or quick actions.

Understand returned evidence

Compare final answers, structured output, Playwright results, observations, artifacts, Debug UI, and errors.

Build an agent workflow

Embed web.agent() in product code with domain policy, approvals, and structured output.

Debug a live run

Inspect live sessions, recordings, artifacts, status, blockers, and errors.

Connect a coding agent

Give Codex, Claude Code, Cursor, OpenCode, or another MCP host the right runtime contract.

​
One runtime, multiple control modes

1

Delegate the goal

Use web agent or SDK web.agent() when a configured model should decide the browser steps.
2

Control exact steps

Use direct Playwright control when your app or coding agent should own deterministic browser steps.
3

Inspect what happened

Use Debug UI, recordings, observations, status, events, files, downloads, and screenshots to understand the run.
4

Add production guardrails

Add policy, approvals, secrets, retries, timeouts, resource handling, and cleanup as the workflow moves toward production.

​
What Webcompute returns

Browser work should not disappear into a black box. Depending on the surface, a run can return a final answer, structured output, Playwright return values, compact runtime status, bounded observations, logs, files, downloads, recordings, Debug UI URLs, CDP URLs, and error codes. Use the high-level agent path when you want the browser task done. Use the runtime path when you need to own each step. For concrete envelopes by surface, read what Webcompute returns.