Simple, predictable pricing

Switchyard doesn't include AI compute costs. You provide your own API keys (e.g., Anthropic, OpenAI) which are stored encrypted on your relay server. This gives you full control over AI costs, model selection, and provider relationships. The platform fee covers orchestration, pipeline management, and the dashboard.

Cursor and Copilot are single-agent IDE assistants. Switchyard is a multi-agent orchestration platform. Instead of one AI helping you type code faster, Switchyard decomposes complex tasks into subtasks, assigns them to specialized agents (coders, testers, reviewers, security analysts), and merges output through a structured pipeline with human review gates.

Correct. This is an architectural guarantee, not a policy. The Switchyard Relay runs on your infrastructure and manages all code, API keys, and execution traces locally. The control plane receives only metadata: task status, step counts, token usage, and costs. There is no code path that could send source code to our servers even if we wanted to.

Switchyard currently uses Claude Code CLI as the primary agent runtime. Support for additional providers (OpenAI Codex, local models via Ollama) is on the roadmap. Since you bring your own API keys, you control which provider and model tier each pipeline stage uses.

One command: curl -fsSL https://switchyard.cloud/install | sh. The installer detects your OS, checks for Docker, runs an interactive setup wizard, and starts the relay daemon. Total time is about 60 seconds. The relay runs as a Docker container with a persistent data volume.

The relay (data plane) is fully self-hosted on your infrastructure. The control plane (dashboard, task orchestration, analytics) is hosted by Switchyard on Cloudflare. A fully self-hosted control plane option is being evaluated for enterprise customers with strict requirements.

You can archive projects to free up slots, or upgrade to the Team plan for up to 50 projects. Archived projects retain all history and can be restored at any time.