Tool: Stagehand Agent — Zelaxy Docs

The Stagehand Agent is an autonomous browser automation agent built on Stagehand. Unlike the standard Stagehand tool (single actions), this block plans and executes multi-step tasks across pages, forms, and flows.

Overview

Property	Value
Type	`stagehand_agent`
Category	Tool — Web Scraping
Auth	Model provider API key + Browserbase server config

Configuration

Setting	Type	Description
Start URL	Short input	First URL to open
Task	Long input	Goal/instructions for the autonomous run
Provider	Dropdown	LLM provider (`anthropic`, `openai`, `google`)
Model	Dropdown	Agent model name
Agent Mode	Dropdown	`hybrid`, `dom`, or `cua` mode
Max Steps	Slider	Max number of steps before stop
Use Web Search	Switch	Enables agent web search capabilities
System Prompt (Optional)	Long input	Extra system-level guidance
Exclude Tools (Optional)	Short input	Comma-separated internal tool names to disable
Variables	Table	Secret/variable placeholders (e.g. `%username%`)
Model Provider API Key	Password	API key for the selected provider
Browserbase API Key (Optional)	Password	Per-run override for Browserbase API key
Browserbase Project ID (Optional)	Short input	Per-run override for Browserbase project
Output Schema	JSON schema	Optional strict structured output target

Behavior Notes

The agent preserves legacy secure placeholder handling, including %variable% substitution and [[ACTION:...]] secure directives.
When credentials are provided in variables, the runtime can attempt direct login before handing off to the autonomous agent loop.
If advanced agent options are rejected by the runtime, Zelaxy automatically falls back to a compatibility config to avoid hard failures.

Outputs

Field	Type	Description
`agentResult`	`json`	Final success/completion/message/actions payload
`structuredOutput`	`json`	Parsed JSON output matching the schema (if available)
`secureActions`	`json[]`	Executed secure directive results
`config`	`json`	Effective provider/model/mode/search settings used

Example: Price Comparison

Workflow:

[Starter: Product name] -> [Stagehand Agent] -> [Response]

Task: Search for "{{starter.input}}" on Amazon and Best Buy. Extract the product name, price, rating, and availability from the first result on each site.

The agent autonomously navigates both sites and can return structured comparison JSON when Output Schema is provided.

Use Variables for secrets:

Key	Value
`username`	`my.user@example.com`
`password`	`my-secret-password`

Task example:

Log in to the portal, open the billing page, and return invoice id, amount, and due date.
If needed, execute [[ACTION:type %password% into the password field and submit]].

Tips

Use high-level goals, but include explicit success criteria for more reliable completion.
Keep Max Steps bounded to avoid wandering behavior on dynamic sites.
Use Output Schema whenever downstream blocks need deterministic JSON fields.
Start with hybrid mode, then switch to dom or cua for site-specific reliability tuning.

Stagehand Agent