LM Studio released a native iPhone app called Locally on June 4, 2026, paired with LM Link, a remote-access technology that lets you chat with large language models (LLMs) running on your Mac directly from your phone, with end-to-end encryption and no cloud intermediary. The company acquired the Locally AI app in April 2026 and rebuilt it as the official mobile front end for its local AI platform. This represents a third path in mobile AI that sits between running tiny models natively on your phone and sending all your data to cloud services like ChatGPT or Claude.

What Makes LM Link Different From Cloud AI?

Most mobile AI coverage presents a false choice: either send your prompts to a cloud service or run a small model directly on your phone's chip. LM Link introduces a different approach. Your Mac becomes a private inference server that you own and control, while your iPhone serves as an encrypted terminal into it. Nothing routes through a cloud relay, and chat history stays on each device. The only data that reaches LM Studio's servers is a device discovery list used to find your machine.

This design choice addresses a fundamental hardware limitation. High-end iPhones like those with the A17 or A18 Pro chip carry roughly 8 gigabytes of unified memory and between 50 to 90 gigabytes per second of memory bandwidth. A Mac Studio with an M4 Ultra reaches up to 800 gigabytes per second. Since generating tokens from large language models is bandwidth-bound, that gap explains why native iPhone inference is limited to models under 3 billion parameters. Larger models simply cannot feed data to the phone's processor fast enough to feel responsive.

How Does the Privacy Architecture Work?

LM Link uses end-to-end encryption over custom Tailscale mesh VPNs, a technology built on WireGuard. The phone-to-Mac connection uses ChaCha20-Poly1305 for encryption and Curve25519 for key exchange, both industry-standard modern cryptographic primitives. Devices are never exposed to the public internet. The coordination server only exchanges small encryption keys; actual data traffic flows directly between your devices.

LM Link layers its own self-contained Tailscale usage via tsnet, a userspace Go library that embeds Tailscale's VPN primitives. This means LM Link will not interfere with an existing Tailscale VPN setup on the same machine. The architecture ensures that your conversations remain private to your devices; there is no server in the middle accumulating your chat history.

What Are the Key Features and Limitations?

• Supported Devices: Locally is available for iPhone and iPad only during the preview period; Android has not been announced. LM Link is free during its request-gated preview, with free and paid tiers planned at general availability.
• Model Compatibility: LM Link works with any model already installed in LM Studio on your Mac, including Apple's built-in foundation models. Performance depends entirely on your Mac's hardware, not the iPhone's capabilities.
• Developer Integration: Because LM Link exposes LM Studio's local API, which is compatible with OpenAI's API format, developer tools that already point at that endpoint can reach it remotely, turning your Mac into a private cloud you control.

How to Set Up LM Link on Your Devices

• Create an Account: Sign up for an LM Studio account and sign in on both your Mac and iPhone to enable the connection.
• Activate LM Link: Once signed in on both devices, activate LM Link in the Locally app to establish the end-to-end encrypted connection between your phone and Mac.
• Start Chatting: Begin interacting with your local models from your iPhone; the connection will handle routing your prompts to your Mac and returning responses securely.
• Monitor Connection Stability: Be aware that the connection may drop if the Locally app spends time in the background, though LM Studio's developers are working to improve reconnection latency and keep the connection alive longer.

One user who tested LM Link for several days noted that the experience has been strong for workflows that rely on local models. The ability to access those models from an iPhone while on the go adds flexibility, though the connection can drop briefly when switching between apps. The developers acknowledged this is a byproduct of how the secure connection is established and indicated they are actively working to improve it.

Why Does This Matter for Privacy-Conscious Users?

LM Link represents a shift in how people can interact with AI without surrendering their data to cloud providers. Your Mac becomes the compute engine, and the phone becomes the interface. Because inference runs on your desktop hardware, not on a third-party server, you maintain control over which models run and how your data is processed. The end-to-end encryption ensures that even LM Studio cannot see your conversations.

This approach also works with headless server builds. LM Studio ships a command-line version called llmster for server environments without a graphical interface, and LM Link can connect to those instances too. In practice, the "Mac" on the other end of the link can be a headless Linux server or a cloud VM running LM Studio's inference stack, not only a desktop with a monitor.

The release comes after a busy spring of incremental updates to LM Studio's inference engine. Version 0.4.13 updated the MLX engine to version 1.8.1; version 0.4.14 shipped MTP Speculative Decoding as stable; and version 0.4.15 added CUDA tensor parallelism for multi-GPU loading and fixed a prompt-cache bug. Locally and LM Link represent the headline feature of version 0.4.16, sitting on top of a steadily maturing inference stack.

"By bringing Locally AI into the LM family, we are doubling down on our mission of making AI accessible and useful to you, across your devices, wherever you go," stated the LM Studio team.

LM Studio Team

The acquisition of Locally AI in April 2026 and the rapid integration of its creator, Adrien Grondin, into LM Studio's mobile development efforts shows the company's commitment to expanding beyond the desktop. LM Link is not the first attempt at connecting a phone to a local LLM instance; community apps such as "Off Grid" and "LM Mini" had already solved this problem for early adopters. What LM Studio adds is an official, team-maintained, end-to-end encrypted implementation with no third-party relay in the path.