Feature request: "LM Link" without Tailscale. I.e. Ability to point one LM Studio to use the server of another.

[mrcodetastic]

Firstly, awesome product. Please don't go down the enshitification path with this.

Can you please implement an advanced users options such that a user could have LM Studio to use another LM Studio's 'Local Server'?

It's great that LM Studio has always had a 'Local Server' capability that can be exposed to all network interfaces, but I've always found it odd, that one couldn't simply "point" a LM Studio instance to a "Local Server" elsewhere.

Appreciate there's now LM Link, but this is unnecessary when:

• Users might already have devices on the same LAN
• Users might already be using ZeroTier or equivalents to have a private network.

Also, TailScale and ZeroTier can have issues when people are also running their own VPNs for privacy reasons.

[AndyP2]

This would have been my preference too, rather than having to use lm link and running into all the problems I documented in #1639. Ideally we could run lms as a service on the machine with the graphics cards in it, the connect to it with lm studio on either that machine or another on the local network.

[HavenCTO]

I built a workaround for this. Works great with lms running as a daemon on the compute machine. https://github.com/HavenCTO/lmstudio-bridge

Avoids Tailscale using libp2p via the IPFS p2p tunnel feature, which handles NAT traversal, just connect using the remote's PeerID.

Of course this requires a separate ChatGPT-like client since lmster is only the backend, so not the full native GUI solution. But it solves the headless + NAT traversal problem today.

[mrcodetastic]

I built a workaround for this. Works great with lms running as a daemon on the compute machine. https://github.com/HavenCTO/lmstudio-bridge

Nice work.

My temporary "fix" was to run LibreChat on the remote LM Studio device with the powerful GPU.

danny-avila/LibreChat#12234

[AndyP2]

My workaround is much more primitive: I remote in to the GPU machine and make adjustments on LM studio, which prompts to stop the daemon when it starts. When I'm finished (or after a reboot), an NSSM service runs lms daemon up again.
I also have OpenWebUI running in a Docker container from my Ollama experiments and could wire that up too, but my main focus is getting Continue working in VS Code.