Skip to content

Quick Start

This guide walks you through collecting a demonstration, learning a policy, and evaluating the agent.

Prerequisites

1. Collect a Demonstration

Start capturing your screen and inputs:

openadapt capture start --name my-task

Now perform the task you want to automate:

  1. Click on applications
  2. Type text
  3. Navigate menus
  4. Complete your workflow

When finished, stop the capture:

# Press Ctrl+C in the terminal, or:
openadapt capture stop

2. View the Trajectory

Inspect what was captured:

openadapt capture view my-task

This opens a trajectory viewer showing:

  • Observations (screenshots) at each step
  • Actions (mouse and keyboard events)
  • Timing information

3. List Your Demonstrations

See all collected demonstrations:

openadapt capture list

Output:

NAME         EVENTS   DURATION   DATE
my-task      45       2m 30s     2026-01-16
login-demo   23       1m 15s     2026-01-15

4. Learn a Policy

Learn an agent policy from your demonstration trajectory:

openadapt train start --capture my-task --model qwen3vl-2b

Monitor policy learning progress:

openadapt train status

Policy learning creates a checkpoint file in training_output/.

5. Evaluate the Agent

Test your trained policy on a benchmark:

openadapt eval run --checkpoint training_output/model.pt --benchmark waa

Or run a mock evaluation to verify the setup:

openadapt eval mock --tasks 10

6. Evaluate an API Agent

Test API-based agents (Claude, GPT-4V):

# Set your API key
export ANTHROPIC_API_KEY=your-key-here

# Run evaluation
openadapt eval run --agent api-claude --benchmark waa

Complete Workflow Example

Here is a complete example demonstrating the full pipeline:

# 1. Install OpenAdapt
pip install openadapt[all]

# 2. Check system requirements
openadapt doctor

# 3. Collect a demonstration
openadapt capture start --name email-reply
# ... perform the task ...
# Press Ctrl+C to stop

# 4. View the trajectory
openadapt capture view email-reply

# 5. Learn a policy
openadapt train start --capture email-reply --model qwen3vl-2b

# 6. Wait for policy learning to complete
openadapt train status

# 7. Evaluate the agent
openadapt eval run --checkpoint training_output/model.pt --benchmark waa

Next Steps