Operative

Computer use agent powered by Claude. Provides a Docker container environment where Claude interacts with a virtual desktop via screenshot, mouse, and keyboard tools, with a Streamlit UI for human interaction.

Image: ghcr.io/hanzoai/operative:latest Repo: github.com/hanzoai/operative

Note: Computer use is a beta feature. It poses unique risks distinct from standard API features. To minimize risks:

1. Use a dedicated virtual machine or container with minimal privileges.
2. Avoid giving the model access to sensitive data such as account login information.
3. Limit internet access to an allowlist of domains.
4. Require human confirmation for decisions with real-world consequences.

Instructions found in web content or images may override user instructions. Isolate the agent from sensitive data and actions.

What It Does

• Runs a full Linux desktop (X11/Xvfb + VNC) inside Docker
• Claude controls the desktop via screenshot/mouse/keyboard tool calls
• Supports Anthropic API (direct), AWS Bedrock, and Google Vertex
• Streamlit UI for sending instructions and watching the agent work
• Built-in bash and file editing tools alongside computer control

Quick Start

Anthropic API

export ANTHROPIC_API_KEY=%your_api_key%
docker run \
    -e ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY \
    -v $HOME/.anthropic:/home/operative/.anthropic \
    -p 5900:5900 \
    -p 8501:8501 \
    -p 6080:6080 \
    -p 8080:8080 \
    -it ghcr.io/hanzoai/operative:latest

Bedrock

export AWS_PROFILE=<your_aws_profile>
docker run \
    -e API_PROVIDER=bedrock \
    -e AWS_PROFILE=$AWS_PROFILE \
    -e AWS_REGION=us-west-2 \
    -v $HOME/.aws:/home/operative/.aws \
    -v $HOME/.anthropic:/home/operative/.anthropic \
    -p 5900:5900 -p 8501:8501 -p 6080:6080 -p 8080:8080 \
    -it ghcr.io/hanzoai/operative:latest

Vertex

gcloud auth application-default login
export VERTEX_REGION=%your_vertex_region%
export VERTEX_PROJECT_ID=%your_vertex_project_id%
docker run \
    -e API_PROVIDER=vertex \
    -e CLOUD_ML_REGION=$VERTEX_REGION \
    -e ANTHROPIC_VERTEX_PROJECT_ID=$VERTEX_PROJECT_ID \
    -v $HOME/.config/gcloud/application_default_credentials.json:/home/operative/.config/gcloud/application_default_credentials.json \
    -p 5900:5900 -p 8501:8501 -p 6080:6080 -p 8080:8080 \
    -it ghcr.io/hanzoai/operative:latest

Accessing the Interface

Once the container is running, open http://localhost:8080 for the combined agent chat + desktop view.

The container stores settings (API key, custom system prompt) in ~/.anthropic/. Mount this directory to persist settings between runs.

Screen Size

Set resolution with WIDTH and HEIGHT environment variables:

docker run \
    -e ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY \
    -e WIDTH=1920 -e HEIGHT=1080 \
    -v $HOME/.anthropic:/home/operative/.anthropic \
    -p 5900:5900 -p 8501:8501 -p 6080:6080 -p 8080:8080 \
    -it ghcr.io/hanzoai/operative:latest

XGA resolution (1024x768) is recommended. For higher resolutions, scale screenshots down to XGA and map coordinates back proportionally. For smaller displays, add black padding to reach 1024x768.

Development

./setup.sh  # configure venv, install dev dependencies, pre-commit hooks
docker build . -t operative:local
export ANTHROPIC_API_KEY=%your_api_key%
docker run \
    -e ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY \
    -v $(pwd)/operative:/home/operative/ \
    -v $HOME/.anthropic:/home/operative/.anthropic \
    -p 5900:5900 -p 8501:8501 -p 6080:6080 -p 8080:8080 \
    -it operative:local

The docker run command mounts the repo inside the container with auto-reloading enabled.

Credits

Based on Anthropic's computer use demo. See upstream repository for original implementation.

License

MIT