Quick start

Get started with OpenSteer in minutes with SDK, CUA agent, or CLI examples

Get up and running with OpenSteer in three different ways: SDK, CUA Agent, or CLI.

SDK

1. Install OpenSteer

   Install the SDK package in your project:

   npm

   npm install opensteer
   
2. Create your first script

   Create a new file quickstart.ts with this complete working example:

   import { Opensteer } from "opensteer";
   
   const opensteer = new Opensteer({ name: "quickstart" });
   
   try {
     await opensteer.launch();
     await opensteer.goto("https://example.com");
   
     await opensteer.snapshot({ mode: "action" });
     await opensteer.click({ description: "main call to action" });
   
     await opensteer.snapshot({ mode: "extraction" });
     const data = await opensteer.extract({
       description: "hero section",
       schema: { title: "string", href: "string" },
     });
   
     console.log(data);
   } finally {
     await opensteer.close();
   }
   
3. Run your script

   Execute the script:

   node quickstart.ts
   

   The script will:

   1. Launch a browser
   2. Navigate to example.com
   3. Take an action snapshot and click an element
   4. Take an extraction snapshot and extract structured data
   5. Close the browser
4. Understand the key concepts

   Namespace (name): Selectors are persisted in .opensteer/selectors/quickstart/ for deterministic replay.

   Snapshots: mode: "action" optimizes HTML for interactions, mode: "extraction" optimizes for data extraction.

   Descriptions: Natural language element targeting like "main call to action" uses AI to find elements and persists selectors.

   Schema extraction: Define the structure you want with { title: "string", href: "string" } and OpenSteer extracts it.

5. Next steps

   Explore more capabilities:

   Actions

   Click, input, hover, select, scroll, and more

   Extraction

   Extract structured data with typed schemas

   Browser automation

   Complete guide to browser operations

   Cloud mode

   Run in the cloud instead of locally

CUA Agent

1. Install OpenSteer

   npm install opensteer
   
2. Set up API key

   Set your model provider API key:

   export OPENAI_API_KEY="your-api-key"
   # or ANTHROPIC_API_KEY, GOOGLE_GENERATIVE_AI_API_KEY
   
3. Create CUA agent script

   Create agent.ts:

   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);
     console.log("Actions taken:", result.actions.length);
   } finally {
     await opensteer.close();
   }
   
4. Run the agent

   node agent.ts
   

   The agent will:

   1. Navigate to Hacker News
   2. Identify the top story link
   3. Click on it
   4. Report back with the result
5. Configure agent behavior

   Customize the agent with options:

   const result = await agent.execute({
     instruction: "Find and summarize the top 3 posts",
     maxSteps: 30,           // Allow more steps for complex tasks
     highlightCursor: true    // Visualize agent actions
   });
   

CUA agents perform autonomous browser actions. Always test with safe websites and appropriate maxSteps limits.

Supported providers: OpenAI (openai/computer-use-preview), Anthropic (anthropic/claude-3-7-sonnet-20250219), Google (google/gemini-2.0-flash-exp)

CLI

1. Install globally

   npm i -g opensteer
   opensteer skills install
   
2. Open a browser session

   opensteer open https://example.com --session demo --name quickstart
   

   This opens a browser, navigates to the URL, and creates a session named demo with namespace quickstart.

3. Take an action snapshot

   opensteer snapshot action --session demo
   

   This returns cleaned HTML with element counters for targeting.

4. Interact with elements

   opensteer click --description "main call to action" --session demo
   

   Click elements using natural language descriptions.

5. Extract data

   First, take an extraction snapshot:

   opensteer snapshot extraction --session demo
   

   Then extract structured data:

   opensteer extract '{"title":"string","href":"string"}' --description "hero section" --session demo
   
6. Close the session

   opensteer close --session demo
   

For CI/CD environments, set OPENSTEER_SESSION or OPENSTEER_CLIENT_ID environment variables instead of using --session flag.

Best practices for AI agents

If you’re building AI agents that generate browser automation scripts:

1. Use OpenSteer APIs instead of raw Playwright calls for persistent, replayable scripts
2. Keep namespace consistent: SDK name must match CLI --name
3. Take snapshots before actions: snapshot({ mode: "action" }) before interactions, snapshot({ mode: "extraction" }) before data extraction
4. Prefer description targeting: More maintainable than CSS selectors
5. Always cleanup: Wrap in try/finally and call close()

See AI Agents guide for complete integration patterns and Skills for installing the OpenSteer skill.

Additional examples

Basic usage

Navigate, snapshot, and interact with pages

Data extraction

Extract structured data from websites

Form filling

Automate form submissions

AI integration

Build AI agents with OpenSteer

Next steps

API Reference

Explore the complete API documentation

CLI Reference

Learn all CLI commands and options

Selectors

Understand selector persistence and replay

Cloud integration

Run browser automation in the cloud