CMX documentation

Compressius Maximus

CMX is a local gateway for AI coding clients. It sits between a client such as Codex or OpenCode and the model provider that client already uses. In longer sessions, CMX can summarize eligible older context before forwarding a request. Your provider still handles the model request and its normal charges still apply.

The gateway listens on 127.0.0.1:17322. You can use it locally without a CMX account. Sign in to CMX only if you want an account-linked dashboard and detailed account statistics.

Get startedInstall, choose whether to connect detected clients, and check a real request.

Supported clientsConnect Codex, OpenCode, Claude Code, Aider, Antigravity (IDE & CLI), or Cline.

Model providersFind built-in upstream routes and learn how native credentials work.

Undo or uninstallRestore direct connections before removing CMX.

What to expect

• CMX only considers compression above its configured context threshold. A short session may pass through without changes.
• A model-generated summary can omit details. CMX preserves recent context, instructions, and protected tool exchanges, then checks each proposal against policy.
• Estimated token savings are not a billing guarantee. Compare actual charges in your model provider’s account.
• Prompts and responses go from your client through the local gateway to your chosen model provider. They are not sent to CMX Cloud. See Security and data handling for local storage details.

Start with the quickstart, then use the guide for your client. If a command differs in your installation, check cmx --version and cmx <command> --help.