1. Download and installation
Choose the package for your operating system and processor from the download page. These instructions apply to v3.2, Build 20261117.44.
macOS: Apple Silicon or Intel
- Download the matching Apple Silicon or Intel
.dmgfile. - Open the disk image and drag AI To Local into
/Applications. - Launch the app. It runs from the menu bar; closing its dashboard window hides the window but leaves the bridge running. Use the menu-bar app menu and choose Quit to stop it.
Windows 10 or 11, x64
- Portable package: download the x64
.zip, extract it to a folder you control, and launch the application from that folder. - Installer: download
Setup.zip, extract it, and run the included setup application. - The app stays available from the system tray. Closing its window does not stop the background bridge; choose Quit from the tray menu to stop it.
Windows packages include the runtime they need. You do not need to install Node.js or Python.
2. Activation and hardware binding
Open the local dashboard at http://127.0.0.1:18926. If it does not open automatically, paste that address into your browser.
- Enter your license key and email address in the activation panel.
- Enter the six-digit one-time code sent to your email to verify account ownership.
- After verification, the app creates a local Ed25519 device keypair and binds the license to this device.
- The app provisions the encrypted tunnel and reports when the local bridge is ready.
Email verification is required for initial activation. OAuth approvals for AI client connections are then handled in the local console and do not require additional verification emails.
3. Connecting web AI clients
MCP (Model Context Protocol) is a standard for AI applications to discover and call external tools. AI To Local v3.2 uses OAuth 2.1 for remote connections. Access tokens last six hours; refresh tokens use a 30-day sliding renewal window.
ChatGPT Web with Developer Mode
- In ChatGPT, enable Developer Mode in settings.
- Open Connected Apps and select Add MCP Server.
- Choose the OAuth connection flow and enter the MCP address shown by AI To Local.
- When the approval request appears, open the local console at
http://127.0.0.1:18926and approve it with one click. - Return to ChatGPT and finish the connection. Future tool calls use the approved connection without another verification email.
Local loopback and studio clients
Claude Desktop, Cursor, and Gemini for macOS can connect locally through 127.0.0.1:18927 using the local Bearer token shown in the console, or launch the app's MCP process in -mode stdio where the client supports standard input/output transport. Keep the token private and configure it only in the intended local client.
For Gemini.app confirmation settings and the exact stdio configuration, see the Gemini for macOS guide.
4. Concurrent agent slots and your own tunnel
V3.2 can serve multiple connected AI agents at the same time. Each agent gets an isolated folder whitelist, and file write locks prevent simultaneous edits from overwriting one another. The included encrypted cloud tunnel is used by default.
Power users can optionally attach their own Cloudflare Tunnel token and custom domain. This routes tunnel traffic through the user's Cloudflare account. The domain can be any Cloudflare-managed domain, including a low-cost $0.99/yr numeric .xyz domain.
5. Folder whitelist and least privilege
Add only the project folders that an AI client needs. The app uses Go os.Root operations and symlink verification to keep access inside authorized roots, alongside checks for sensitive credential locations.
- Choose a specific workspace such as
~/Projects/sample-apporC:\Users\You\Projects\sample-app. - Review each agent's folder access independently and remove folders when the work is complete.
- Never whitelist
/orC:\. Broad roots expose files unrelated to your project.
6. Terminal and high-risk controls
File access and command execution use separate controls in the local dashboard:
- File Tools: controls approved file reading, searching, and editing inside whitelisted folders.
- Terminal Session Tools: enables managed sessions through
start_process, useful for longer builds and tests. - High-Risk Shell: enables direct
execute_commandexecution and is off by default. Turn it on only when a task requires it, then turn it off again.
These switches are independent. A client can use approved file tools while terminal access remains disabled.
7. Troubleshooting
Access Denied: Path out of whitelist
The requested path is outside the folders assigned to that agent. Add the specific project folder in the console or ask the AI to work within an already approved folder. Do not solve this by granting access to a drive or filesystem root.
502 response or tunnel offline
Confirm the app is running in the menu bar or system tray, the device has internet access, and the dashboard reports the tunnel as connected. Restarting the bridge from its tray menu can re-establish the tunnel. Local loopback clients can still connect on the same computer when the cloud tunnel is offline.
Gemini confirmation popups
Gemini for macOS may ask for approval before running command tools. Review the one-click console option and configuration steps in the Gemini for macOS guide.
8. Stopping, unbinding, and uninstalling
Choose Quit from the menu-bar or system-tray app menu for an immediate stop. This closes active bridge and tunnel services.
If you are moving to another computer, release the current device binding from the local console or account controls before activating on the new device. On macOS, quit the app and move it from /Applications to Trash. On Windows, uninstall the installed edition through Settings > Apps; for the portable edition, quit it and delete its extracted folder. AI To Local does not install a persistent system service or leave a background process after quitting.