Work with local files
Read or edit files in explicitly approved folders. Real data, real project context, right where it lives.
“Read the README in my approved project folder.”
A task, in plain language.Local Computer is a secure connection layer between your AI and the work already on your machine. Not another cloud desktop. Not another place to move your files.
Read or edit files in explicitly approved folders. Real data, real project context, right where it lives.
Run local commands and inspect durable tasks, subject to your machine's actual operating-system permissions.
See the authorized capabilities and tools on each connected device. No fixed integration for every use case.
Identity, client access, and workspace permissions are separate. Your agent can operate after you authorize the exact path — not because a website says it's connected.
See how permissions work ↗Download a verified Staging candidate for macOS Intel or Linux x64. Your Runtime operates on that machine.
Authorize device registration and ChatGPT OAuth, then approve the specific workspace or capability you need.
Ask it to inspect status, open an approved folder, and return an actual file. Revoke access when appropriate.
Current Staging candidates, with exact source identities and SHA-256 checksums. One latest download per platform.
Both builds have passed existing-device Runtime / real MCP file-read checks. Neither is a trusted public release or a claim of independent fresh-device installation acceptance. Windows is deferred.