Coming soon

Give your cloud browser a goal. We'll do the rest.

Agent Actions is our upcoming API for autonomous browser automation. Just pass in a natural language task and a URL. We'll execute the steps, stream the progress in real-time, and return the structured data you asked for.

API sketch, subject to change. Not a live contract yet.

Use Humanized REST today Read current docs
Draft REST request

One API call. Infinite possibilities.

Stop writing brittle CSS selectors. Just describe what you want the browser to accomplish, set your boundaries, and listen to the progress stream.

POST body
{
  "task": "Find order BC-1042 and return its delivery date.",
  "url": "https://example-shop.test/orders",
  "options": {
    "maxSteps": 12,
    "humanIntervention": "allowed",
    "secretValues": { "crmApiKey": "{{ vault ref }}" },
    "outputSchema": {
      "orderId": "string",
      "status": "string",
      "deliveryDate": "string"
    }
  }
}
Planned inputs

Set the goal, build the guardrails.

You control the execution. Inject secrets securely, define the exact JSON schema you want back, and decide if the agent is allowed to ask humans for help when it gets stuck.

task

Natural-language goal for the browser session to complete.

url | sessionId

Start from a URL or attach the task to an existing BrowserCity session.

options.maxSteps

Bound how many observe-plan-act loops the task can spend.

options.humanIntervention

Allow a human handoff event when the task needs approval, credentials, or judgment.

options.detectElements

Ask the agent to enrich observations with interactable element hints.

options.vision

Enable visual observations for pages where DOM text is not enough.

options.agent

Select the planned task-runner profile when multiple agents are available.

options.extendedSystemMessage

Add task-specific steering that should not live in the reusable prompt.

options.secretValues

Planned task-scoped secrets: redacted from stream events and cleaned up after completion.

options.outputSchema

Optional JSON Schema for structured final output.

SSE stream
event: agent_action.started
data: {"id":"act_01j...","status":"running"}

event: agent_action.step
data: {"index":1,"intent":"Open order lookup"}

event: agent_action.output
data: {"orderId":"BC-1042","deliveryDate":"2026-06-02"}

event: agent_action.completed
data: {"status":"completed","steps":8}
Draft stream events

Watch it think in real-time.

Every navigation, observation, and action is streamed back via Server-Sent Events (SSE). Hook directly into your UI or logging system without writing polling loops.

agent_action.startedagent_action.stepagent_action.observationagent_action.human_interventionagent_action.outputagent_action.completedagent_action.failed
Available now

Need browser actions today?

Use Humanized REST /v1/do/* for shipped browser actions like open, navigate, click, type, extract markdown, capture screenshots, and close sessions.

Explore Humanized REST
[ 06 / 06 ] — Get Started

Give your AI agents the web.

Start for free. No credit card required. Private sessions by default.

Start Free Read Docs Join Discord