LM Studio adds iPhone-to-Mac connection for running local LLMs via Locally app

LM Studio adds iPhone-to-Mac connection for running local LLMs via Locally app

LM Studio has released LM Link, a feature that establishes an end-to-end encrypted connection between iPhones and Macs running the company's local LLM software.

The feature updates both LM Studio's Mac app and the Locally AI iOS app (which LM Studio acquired earlier in 2026). Users can now interact with any local model installed on their Mac directly from their iPhone, including Apple's foundation model used in Apple Intelligence.

Technical implementation

LM Link uses Tailscale mesh VPNs for connectivity. According to LM Studio, "Your devices are never exposed to the public internet, because LM Link runs on top of custom Tailscale mesh VPNs." The company states this is "an entirely separate and self-contained use of Tailscale VPN primitives," meaning it won't interfere with existing Tailscale VPN configurations.

Users must create an LM Studio account and sign in on both devices to activate the feature. Once connected, all data and communication between devices remains encrypted.

Performance and limitations

Performance depends on the Mac's hardware specifications, same as running models locally. Early testing reveals connection drops when the iPhone app runs in the background—for example, when switching to locate documents or perform web searches. LM Studio developers acknowledge this behavior is tied to how the secure connection is established and are working to improve reconnection latency and connection persistence.

The feature works with models like Google's Gemma 4 12B, released yesterday and designed to run on Macs with 16GB or more memory.

Pricing

LM Link will be free during the preview period. After that, LM Studio plans to offer both free and paid plans, though pricing details have not been disclosed.

What this means

This extends the utility of local LLM deployment by enabling mobile access without compromising privacy—a key reason developers choose local models. The Tailscale implementation suggests LM Studio is targeting users who prioritize data sovereignty. The current reconnection issues indicate the feature is in active development, typical for preview releases.