Remote Mac

2026 M6 Mac mini Running Ollama: How to Set Up Tailscale Remote Access?

Set up Tailscale on the M6 Mac mini and each client, then reach Ollama through the Mac’s tailnet address; do not expose its API directly to the public internet. This is the right setup when every device can join the private network and the Mac is awake with Ollama running.

Who this guide is for: Individual developers calling Ollama on a Mac mini from a laptop or phone while away from home.
HomeLab users who want private remote connectivity without router port forwarding.
Small-team technical leads who need to grant access to selected members and document the service’s availability limits.

Last updated October 20, 2026. Apple’s M6 announcement and macOS version information, plus current Tailscale and Ollama documentation, are the references for this guide. Confirm the macOS version actually installed on the device; a version mentioned in earlier discussion is not evidence that it shipped with, or is installed on, this Mac.

Start by defining what “remote access” means

The target is a request from an authorized client to an Ollama service on the Mac, carried over a private tailnet. That is different from publishing an API on a public IP, and it depends on three separate conditions:

  • The Mac is reachable on the tailnet. Tailscale connects participating devices; it does not start a sleeping Mac.
  • Ollama is listening on an address the client can reach. A Mac can appear online in Tailscale while the application remains bound to its local loopback address.
  • The service process is running. A valid network route cannot restart Ollama if the app has stopped or failed to launch.

Can Tailscale replace router port forwarding? For devices that can run the Tailscale client, it can provide the private network path without forwarding Ollama’s port from the router. That does not mean the service is automatically safe: the Mac’s listener, tailnet access policy, and Ollama process still need to be checked. A subnet router is intended for reaching devices on a network that do not run the client themselves, not as a default requirement for connecting one Mac and one remote laptop.

Before changing anything, confirm these prerequisites:

  • Ollama starts and responds locally on the Mac.
  • You can sign in to the Tailscale account or tailnet that will authorize the devices.
  • The Mac has a working local network connection.
  • The remote client can install and use the Tailscale app, or an administrator has deliberately chosen another network design.
  • Someone responsible for the service can check the Mac locally if it becomes unreachable.

Enroll the Mac and remote clients in the same private network

Install Tailscale on the Mac using the official macOS installation instructions. Install the appropriate client on each laptop or phone that needs access. Follow the sign-in and approval flow shown for the actual app and account: device approval may be required, and the available screens can differ by client form.

After sign-in, check that the Mac and intended client appear in the same tailnet and show as connected. If the devices belong to separate tailnets, do not assume that sharing a device is the same as joining one common network. Tailscale documents device sharing as a separate option; use it only when that is the intended administrative arrangement and the receiving user is authorized.

Find the Mac’s Tailscale address in the device list or through the client’s device details. Use that address from the remote client, not the Mac’s home LAN address and not its public IP. The Tailscale address and DNS documentation explains how these addresses identify tailnet devices. Keep the address in your service notes alongside the device name so that an operator can distinguish it from a router-assigned local address.

Does an outside client need a subnet router? Usually not when both the M6 Mac mini and the client can run Tailscale and have been authorized. Consider a subnet router only when the requirement is to reach a separate LAN device that cannot run Tailscale itself, and review the router’s advertised routes and access rules before enabling them.

Configure Ollama to accept requests beyond the Mac itself

Ollama’s official FAQ describes its default local listening behavior and the OLLAMA_HOST setting for changing the listening address. A service bound only to loopback is available to processes on the Mac, not directly to a remote tailnet client.

First establish the local baseline: open Ollama on the Mac and make a local request. Then choose a listening address deliberately. Binding to the Mac’s Tailscale address can limit the service to that interface if the address is available when Ollama starts and the application accepts that binding in the installed version. Verify the result after restarting the app; do not assume the setting took effect merely because a command completed.

If a broader bind address is required, Ollama’s documentation gives OLLAMA_HOST as the configuration mechanism. On macOS, its FAQ describes setting environment variables for the app through launchctl. For example, an administrator can set the intended value and then quit and reopen the Ollama app:

launchctl setenv OLLAMA_HOST "0.0.0.0:11434"

The value above listens on all interfaces; it is not a security control. The documented default service port is 11434, so verify the actual listener and do not assume a changed environment variable altered the running process. If broad binding is necessary, use the macOS firewall and tailnet access controls to limit who can reach it. Avoid router port forwarding and do not publish this endpoint as an unauthenticated public API.

Important: A Tailscale connection identifies a route between devices; changing Ollama’s bind address does not add user authentication to the API. Keep access limited to trusted tailnet members and review the applicable Tailscale access-control setup.

The service’s HTTP interface is documented in the Ollama API introduction. A successful connection to the Mac is not proof that the API is restricted to the right people. If a team needs to share access, confirm whether to add members to the tailnet or use controlled device sharing, then review the Tailscale sharing guidance.

Test the route, port, API, and model request separately

Run tests in order. That makes it easier to locate the failing layer instead of changing firewall or sleep settings when the problem is actually an Ollama listener.

  • Check tailnet connectivity. From the remote device, confirm that the Mac appears connected in the Tailscale client. If it does not, resolve sign-in, approval, or device connectivity before testing Ollama.
  • Check the target address. Use the Mac’s tailnet address or its tailnet DNS name, if configured. Do not substitute the home LAN address when testing over the internet.
  • Check the listener from the Mac. Confirm Ollama is running and listening on the expected interface and port. If it is still bound only to loopback, the remote device cannot reach it through the tailnet.
  • Check the API from the remote client. Request the tags endpoint using the Mac’s tailnet address: curl http://<MAC_TAILNET_ADDRESS>:11434/api/tags. The API documentation describes the endpoint; use the response to distinguish an HTTP service response from a network timeout.
  • Check an actual model request. Once the API responds, submit a small request using the client or API workflow the team plans to use. Confirm that the response comes from the intended Mac and model, not a local service on the client.

What if Tailscale connects but Ollama does not? Check the Ollama process, the bound address, the firewall, and the tailnet policy. A connected tailnet proves only that the network path is available; it does not prove that an application is listening or that its access policy permits the request.

Record the client device, target address, Ollama status, listener address, and result of each test. That record gives the next operator a way to separate an enrollment issue from an application issue without making unrelated system changes.

Confirm access while the Mac is awake, then document its sleep boundary

Test remote requests while the Mac is awake and Ollama is running. Then separately observe what happens when the Mac sleeps, and record the actual result for the installed macOS version, power configuration, and network environment. Do not treat successful access while awake as evidence that Tailscale can wake the host.

Can Tailscale keep Ollama reachable after the Mac sleeps? No. Tailscale provides network connectivity between devices; it does not guarantee that a sleeping Mac will respond or that the Ollama process will remain active. If the Mac is asleep, first verify its local power and wake behavior rather than treating the problem as a failed tailnet route. This guide does not prescribe sleep-disabling commands: that is a separate power-management decision and should be based on the device’s actual operating requirements.

Use this handover checklist before giving access to another person:

  • [ ] Confirm the Mac and each approved client are enrolled in the intended tailnet or explicitly shared under a controlled arrangement.
  • [ ] Record the Mac’s Tailscale address or DNS name and the Ollama listener address.
  • [ ] Verify that Ollama answers locally and through the authorized remote client.
  • [ ] Confirm that no router rule exposes the Ollama API to the public internet.
  • [ ] Review tailnet membership and device approvals; remove access when it is no longer required.
  • [ ] Record the responsible operator, who can change access, and how to check the Mac locally if it goes offline.
  • [ ] Note that sleeping, a stopped app, and a disconnected network can each make the service unavailable for different reasons.

Choose the access pattern that matches the devices

Access pattern Use it when Main trade-off
Tailscale on the Mac and every client Each authorized device can install the client and join the intended tailnet Simple device-to-device access; every client still needs authorization
Controlled sharing between tailnets A specific external user needs access without joining the owner’s tailnet Sharing needs deliberate review and does not replace service-level checks
Subnet router A required LAN device cannot run Tailscale itself Adds routing and policy decisions; unnecessary for a Mac and client that can both run the app
Public port forwarding Avoid for an Ollama endpoint without a separately designed authentication and security layer Enlarges public exposure and is not needed for private tailnet access

Decide whether to keep the host local or use a remote Mac environment

A Mac mini at home keeps the hardware under local control, but its availability depends on local power, home internet, the Mac’s sleep state, and someone being able to troubleshoot it. The operator also owns the work of updates, access reviews, and service restarts. Those are real costs even when no new hardware purchase is involved.

A rented Mac environment can be more convenient for temporary development or a test service when maintaining a home host is the bigger burden. It also introduces a provider dependency and recurring rental costs, so it is not automatically the better choice for continuous workloads or software that needs a physical local connection. For a short-lived remote environment, KuzCloud rental may offer a simpler path than maintaining a Mac mini at home; review the available KuzCloud plans before deciding, and compare the required system access and operating responsibilities rather than assuming the environments are interchangeable. The KuzCloud console is available for reviewing account-side options, while KuzCloud help can clarify service questions.

If the home Mac remains the right fit, the operational standard is straightforward: restrict the tailnet to approved devices, verify Ollama’s listener, test the API from an authorized client, and document that sleep can interrupt access. If the work is temporary and local hardware upkeep is the main obstacle, compare a remote Mac environment against those responsibilities before committing to a long-running home deployment.

Run Your Workloads on a Dedicated Remote Mac

Provision a dedicated Mac mini M4 and receive SSH or VNC credentials, usually within five minutes.

Install your own tools with full root access on physical Apple Silicon hardware without virtualization overhead.