Getting started

Prerequisites

Prepare GitHub and the local runtime before connecting a repository.

Accounts and access

You need a GitHub account that can read the target repository and create or use a GitHub Project. Repository connection and workspace administration require an owner or admin role in Orchestrator.

Model-provider access is configured later on the runner. You can use OpenAI Codex OAuth or an API key, Anthropic setup-token OAuth or an API key, and an OpenRouter API key.

Local tools

The official CLI runs on macOS and Linux. Install Homebrew, Docker, kind, and kubectl. Docker must be running before the CLI prepares the local runtime.

brew install kind kubectl
brew install interesta-ai/tap/orchestrator
orchestrator --version

Plan local capacity

Agent workloads run in a local kind cluster. Available CPU, memory, and disk affect how many jobs can run concurrently and how quickly repository images are prepared.

Frequently asked questions

Does the CLI support Windows?

The current documented release targets macOS and Linux through Homebrew.

Why are Docker, kind, and kubectl required?

The runner uses a local Kubernetes cluster to isolate each agent job and manage its workspace, sidecar, credentials, and cleanup.

Must Docker be running during setup?

Yes. The install command pulls missing website-specified runtime images, loads them into kind, and prepares the cluster.

Do I need all three model providers?

No. Configure only the providers you intend to select for agents.