cloudrundocs

Concepts

The core ideas in cloudrun — organizations, projects, engines, agents, and connectors.

Concepts

Five nouns cover cloudrun.

Organizations

Everything lives under an organization — your workspace. Members have roles (admin or member), and API keys are scoped to the org. Manage both under Settings.

Projects

A project is one deployment: its own isolated machine, its own agents, its own connected tools. Projects are the unit of separation — a project is "the whole thing" for a given team or product, not a folder.

Engines

An engine is an agent runtime a project runs. Engines come in kinds, and each kind brings its own substrate, draws its own node on the canvas, and answers its own API namespace — cloudrun doesn't assume what a runtime looks like. Every project starts with one engine and you can add more.

One kind ships today: Hermes. The contract kinds implement — and what a Hermes engine specifically offers — lives in Engines.

Agents

An agent is an individual worker living on an engine — the nodes you see on the canvas. What an agent owns is its engine kind's business: on Hermes, agents are profiles with their own instructions, skills, tools, channels, automations, and memory (the full list). One engine can host many agents, each fully independent.

Connectors

Connectors are one-click links to third-party tools. Connect a service once at the agent level and the agent can act through it — you grant access, cloudrun holds the connection, and no credentials get pasted into prompts.

The canvas

The project view is a canvas: every engine and agent is a node you can arrange. Clicking a node opens its panels. It's one view over the fleet rather than a dashboard per agent — positions are yours and persist.

On this page