From a request
to a verified result.
Ganado Bridge connects an authorized AI client to tools on your computer. The initial product targets individual macOS users who already work with AI coding or operations assistants.
The three parts
AI client: interprets your request and chooses tool calls. Transport: carries authenticated messages. Local agent: reads files or starts processes using the OS permissions it already has. These are separate components; a working local test does not prove a working ChatGPT connection.
What the core does
| Capability | Behavior |
|---|---|
| File access | Absolute paths, bounded pages, explicit continuation offsets. |
| Checked writes | New files require an absent target; replacement uses the current SHA-256. |
| Literal edits | Exactly one occurrence must match. Ambiguous edits fail. |
| Processes | Start once, poll a session, see real exit codes and timeout state. |
| Remote host | Commands target an explicitly configured, already-authorized SSH alias. |
| Audit metadata | Operation name, status and timing; not command text or file contents. |
What it does not do
Bridge does not provide its own AI model, hosted computer, privilege elevation, unrestricted availability, persistent sessions across restart, or a graphical remote-desktop stream. Direct file tools act on the connected local machine; SSH is explicit in a process call.
The current ChatGPT route
The internal deployment uses an official private MCP tunnel. That is not a public product endpoint and does not mean a new user can install Bridge from the directory. The commercial launch needs independently tested setup, a supported distribution transport and reviewer-ready isolation.
Reference: OpenAI’s Secure MCP Tunnel documentation.