Prerequisites
- Python 3.12
- uv and Git, used by the install command below
- An S3 bucket for benchmark artifacts and agents
- AWS credentials with the required permissions
- A configured sandbox provider
- A Vals API key when using hosted mode
Install the CLI
valkyrie and valk invoke the same CLI.
Configure Valkyrie
Upload an agent
Push a local agent directory that contains a validcontract.yaml:
Start your first run
--connect streams progress until the run finishes, so you can watch the first run end to end. The command prints the run id; use it with any other run command. Add --model when the agent contract requires one.
Next steps
- Start a run for run scope, secrets, and repeated runs
- Monitor runs for streaming, status, and history
- Results and outputs for scores and agent files
valkyrie run startreference for every option and default