Introduction
CUA Agent Quickstart with OpenSteer
Learn how to execute CUA-style agent workflows in OpenSteer using reproducible browser automation and extraction primitives.
Quick Answer
Use the CUA quickstart to execute agent-style browser tasks on top of OpenSteer primitives while keeping replay and extraction consistency.
Quick Start: CUA Agent
import { Opensteer } from "opensteer";
const opensteer = new Opensteer({ model: "openai/computer-use-preview" });
try {
await opensteer.launch();
const agent = opensteer.agent({ mode: "cua" });
const result = await agent.execute({
instruction: "Go to Hacker News and open the top story.",
maxSteps: 20,
});
console.log(result.message);
} finally {
await opensteer.close();
}
9) Use CUA Agent
const opensteer = new Opensteer({
model: 'openai/computer-use-preview',
})
await opensteer.launch()
const agent = opensteer.agent({
mode: 'cua',
})
const result = await agent.execute({
instruction: 'Go to docs and summarize the first section',
maxSteps: 20,
highlightCursor: true,
})
console.log(result.message)
await opensteer.close()
V1 CUA providers: openai, anthropic, google.
Agent
agent(config: OpensteerAgentConfig): OpensteerAgentInstance
Create a Computer Use Agent (CUA) instance with ergonomic defaults.
const agent = opensteer.agent({
mode: 'cua',
model: 'openai/computer-use-preview',
})
const result = await agent.execute({
instruction: 'Open docs and summarize the first section',
maxSteps: 20,
highlightCursor: true,
})
- V1 supports
mode: 'cua'only. - Supported providers:
openai,anthropic,google. modelsupports either:"provider/model"string{ modelName: "provider/model", apiKey?, baseUrl?, organization?, thinkingBudget?, environment? }
agent.execute(instructionOrOptions): Promise<OpensteerAgentResult>
instructionOrOptions can be:
stringinstruction{ instruction: string, maxSteps?: number, highlightCursor?: boolean }
highlightCursor overrides the instance cursor setting for that execution:
true: force cursor preview on for this runfalse: force cursor preview off for this run- omitted: use instance cursor setting
Returns:
success,completed,message- executed
actions - token/inference
usage - resolved
providerandmodel
Covered Features
sdk:agentsdk:agent.execute
