Overview

Proliferate runs coding agents like Claude Code, Codex, OpenCode, and Cursor inside isolated workspaces, then gives you a real diff to review before anything ships. This page is the map: what the pieces are and how they talk to each other. The next two pages go deeper on the components themselves and the architecture that connects them.

The short version

Every task happens in a workspace: a repository checkout, worktree, or cloud sandbox that an agent can read, edit, and run commands in. You open a workspace, pick an agent and a model, and send a prompt. The agent works, streams its progress back to you, and leaves behind a branch you can review like any other diff.

Three layers make that possible:

Client (Desktop or web)
  the app you look at: chats, terminals, files, git review, settings

Cloud control plane (Server)
  the source of truth: accounts, organizations, workspaces, credentials,
  billing, integration connections

Runtime (AnyHarness)
  the process that actually drives the agent, locally on your machine or
  inside a cloud sandbox

The client never talks to a model provider or a third-party tool directly. Third-party tool calls always go through a Proliferate-owned gateway, so a connected app's credentials never reach the agent. Model calls either use your own agent login or key directly, the same as running that agent outside Proliferate, or route through a managed gateway if you turn that on. See Architecture for exactly how that separation works.

Local or cloud, chosen up front

When you create a workspace you choose where it runs: a local checkout, a new worktree on your machine, or a cloud sandbox. Whichever you pick, the workspace behaves the same way (chat, terminal, files, git review) because it's the same runtime underneath. What changes is where that runtime process lives.

Because everything is git-backed, moving work from one target to the other is a normal git operation, not a special feature: push the branch from one workspace, then open or continue a workspace on that branch somewhere else. Read Local & cloud for the full picture.

Agents run through their native harness

Proliferate doesn't reimplement Claude Code, Codex, or any other agent. It launches each one through its own CLI or protocol, so the auth, tools, models, and permissions you already know keep working, and new agent features show up without waiting on Proliferate to add them. See Agents for the list of supported harnesses.

Where to go next