LM Studio adds iPhone-to-Mac Link for local LLMs

LM Studio has brought its LM Link feature to the iPhone, letting users tap language models running on a more powerful machine they own, such as a Mac, from their phone as if the models were local, 9to5Mac reported on June 4. LM Link itself is a cross-device feature LM Studio introduced earlier in 2026 in partnership with Tailscale; the new development extends it to mobile.

According to LM Studio's official description, once devices running LM Studio (or its headless "llmster" build) are authenticated, they automatically discover each other and establish an end-to-end encrypted connection regardless of network, "without exposure to the public Internet." Models hosted on a powerful device appear alongside local models in the model loader, so a lightweight client can use large open-weight models hosted elsewhere. LM Studio says chats remain on the local device and "nothing gets uploaded to LM Studio's backend servers apart from your device list," which is used for discovery.

LM Studio says LM Link runs "on top of custom Tailscale mesh VPNs." In a February 25 blog post, Tailscale said the feature is built on tsnet, "a userspace Go program that works without touching kernel sockets or routing tables," providing short-lived onboarding and encrypted peer-to-peer traffic. Tailscale's Kevin Purdy wrote that prompt and response data "is never seen, either by Tailscale or LM Studio's backend service." LM Studio's FAQ states LM Link is "an entirely separate and self-contained use of Tailscale VPN primitives" that "coexists with other uses of Tailscale on your machine or network, with no interference or interplay."

Running LLMs locally keeps data on hardware the user controls, but ties inference to whichever machine holds the model. A vendor-supplied, encrypted bridge lets a phone or laptop reach a desktop or GPU rig without exposing ports to the internet or hand-rolling a tunnel. LM Studio notes remote models also work with tools that already point at its local server (localhost:1234), such as Codex, Claude Code, and OpenCode, so existing local-inference workflows can use remote hardware unchanged. Tailscale frames the same mechanism for teams sharing GPU-backed models, edge devices needing more compute, and regulated industries that want to avoid public model exposure. Community projects like Off Grid and LM Mini, documented in a dev.to guide and the App Store, show developers were already assembling comparable mobile-to-local setups.

LM Link is in Preview and "free for the duration of the Preview period," with both free and paid plans planned at general availability, per LM Studio, which says it is rolling out access in batches. Observers should watch for the paid-plan pricing, published detail on LM Link's key management and trust model, and how iPhone latency and throughput compare with desktop-to-desktop links, especially for large models on lower-RAM hardware. LM Studio says the free tier can be used at home or at work, with enterprise deployment available on request.