Skip to main content

Prerequisites

Python 3.12+

Helios requires Python 3.12 or later

Docker

Docker must be installed and running

Installation

Configure API Keys

Set your LLM provider API key:
export GEMINI_API_KEY=your-api-key
# or
export GOOGLE_API_KEY=your-api-key

Run Your First Task

1

Run a simple task

helios tasks/create-hello-file
This runs a headless task that creates a file in a Docker container.
2

Watch the result

The output shows the agent’s progress and verification result:
Task: create-hello-file
Model: gemini/gemini-2.5-computer-use-preview-10-2025

[Agent] Creating file...
[Tool] bash: echo "Hello World" > /home/hello.txt
[Verify] Running test.sh...
[Result] PASS (reward: 1.0)

Run a GUI Task

GUI tasks require the desktop Docker image:
1

Build the desktop image

docker build -t cua-desktop -f docker/Dockerfile.desktop .
2

Run with the web viewer

helios tasks/explore-desktop --watch
3

Open the viewer

Navigate to http://localhost:8080 to watch the agent work in real-time.
The --watch flag starts a web server that streams the agent’s activity, including VNC view of the desktop, tool calls, and LLM responses.

Choose a Model

Specify a different model with the -m flag:
# Use Claude Sonnet
helios tasks/create-hello-file -m claude-sonnet-4-20250514

# Use OpenAI
helios tasks/create-hello-file -m openai/computer-use-preview

# Use Bedrock
helios tasks/create-hello-file -m bedrock/global.anthropic.claude-opus-4-5-20251101-v1:0

Run Multiple Tasks

Run a batch of tasks in parallel:
# Run all tasks in a directory with 4 concurrent containers
helios batch tasks/ -n 4

# With a specific model
helios batch tasks/ -n 4 -m claude-sonnet-4-20250514

Interactive Mode

Pause and inspect the agent during execution:
helios tasks/create-hello-file -i
Press p to pause/resume execution.

Next Steps