Run LM Studio's API server on the host, enable Serve on Local Network and authentication, then use the host's private LAN IP and actual server port in Local Waifu's /v1 base URL. Test from the companion computer. Do not forward the port through your router.
If LM Studio runs on one computer and your companion runs on another, use the LM Studio computer’s private network address, not localhost, in the companion’s endpoint. Start the LM Studio API server, enable Serve on Local Network, require authentication, and check the address from the companion computer before sending a chat. Keep this on a trusted LAN: do not forward the server port through your router.
This is a two-computer guide. For a server and companion on the same computer, keep LM Studio on loopback and use the connection-refused troubleshooting guide. Changing the server’s network exposure does not fix a same-machine port mistake.
The host address must belong to the LM Studio computer
The short version: 127.0.0.1 and localhost mean “this computer” to whichever program sends the request. On the companion computer, they do not point across the room.
Call the computer running LM Studio the host and the computer running Local Waifu the client. They can both be on the same home network while having different private IP addresses. A typical host address might look like 192.168.1.50; that is an example, not an address to copy. Find the host’s current address in its operating system’s network settings. Check that it belongs to the active Wi-Fi or Ethernet connection, not a VPN or an unused adapter. Private IPv4 addresses commonly start with 192.168., 10., or 172.16. through 172.31..
The address can change after a reboot or router lease renewal. If a connection worked yesterday, compare the saved endpoint with the host’s address today. A name that resolves on one computer may not resolve on the other; start with the literal private IP while diagnosing. Being on the same Wi-Fi name does not prove direct reachability if guest-network isolation or client isolation is enabled.
LM Studio must listen beyond its own loopback interface
The short version: Start the API server on the host and deliberately enable Serve on Local Network. The host’s own LM Studio chat window is not a test of its network API.
LM Studio’s local-network server guide says this setting lets devices on the same network connect and changes the API listening address away from localhost. Its server settings reference lists the server port, network serving, authentication, and CORS as separate controls. Open the Developer page on the host, start the server, enable network serving, and record the displayed port. 1234 is the documentation’s example; use the port your server actually reports.
If the listener shows 0.0.0.0, that is a bind address meaning it accepts connections on IPv4 interfaces. It is not the client URL. The client needs the host’s actual private IP, for example http://192.168.1.50:1234/v1. Use a trusted network and allow the server’s port through the host firewall only for the local devices that need it. Do not disable the firewall wholesale. If the two devices are on separate guest networks, move them to a trusted segment rather than opening the router to the internet.
Protect the LAN endpoint before a companion sends anything personal
The short version: Network serving expands who can reach LM Studio. Enable Require Authentication, give the companion its own API token, and keep the port off the public internet.
LM Studio warns that binding to anything other than 127.0.0.1 exposes the server beyond localhost and recommends authentication. Its authentication guide says requests do not require a token by default; in Server Settings, turn on Require Authentication, then use Manage Tokens to create a token. This feature requires LM Studio 0.4.0 or newer. A request without a valid token should not be treated as an authorized chat request.
Authentication is access control, not encryption. A plain http:// LAN connection does not become encrypted because you added a bearer token. Keep the network trusted, restrict the host firewall, and avoid sending intimate conversations across a shared or untrusted Wi-Fi network. Do not paste a token into a browser URL, screenshot, forum post, or diagnostic output. Do not enable CORS simply because a desktop client cannot connect; LM Studio describes it as a separate cross-origin browser setting.
No router port forwarding. That would expose the API beyond the local-network setup this guide covers. A public IP address is not a substitute for a missing private-LAN route.
Use a private-IP base URL in Local Waifu
The short version: Save http://HOST_PRIVATE_IP:PORT/v1 as a custom endpoint, using your host’s actual address and port. Select its model for the chat after setup.
On the client computer, open Local Waifu’s Settings, then Cloud AI, Your own endpoints, and Add an endpoint. Enter a descriptive name, the base URL, and the LM Studio token if authentication is enabled. The setting’s Cloud AI label does not make a private-IP endpoint a cloud provider. Do not enter /v1/chat/completions as the base URL: Local Waifu appends the operation path. LM Studio’s OpenAI-compatible API reference documents the /v1/models and /v1/chat/completions routes.
Current Local Waifu Mac and Windows source permits plain http:// custom endpoints when the saved URL has a recognized loopback, private IPv4, or .local host; public endpoints require https://. The chat route recognizes such saved LAN endpoints under Offline Mode. This confirms that a second-computer private-IP endpoint is supported by the current routing logic, not that your particular firewall, network, token, or model has been tested. The local classification uses the host written in the URL rather than resolving an arbitrary hostname to see whether it is private. Use the literal host IP during diagnosis.
There is an important settings caveat: current Test and model-list commands block when Offline Mode is on, even for a LAN endpoint that chat routing allows. A blocked settings test is not proof of an unreachable server. If you need those settings operations, temporarily switch Offline Mode off without sending private chat text, test and select the intended endpoint model, then restore Offline Mode and verify a short, non-sensitive reply on that selected route. Offline Mode does not make traffic between two computers stay on one computer or encrypt your LAN.
Test from the client before blaming the model
The short version: Check the host and port from the companion computer first. A response from /v1/models narrows the fault to authentication or API behavior; no response points to reachability.
On a trusted network, request http://HOST_PRIVATE_IP:PORT/v1/models from the client browser or run curl -i http://HOST_PRIVATE_IP:PORT/v1/models, replacing both placeholders. The model-list route tests the API without transmitting a personal conversation. With authentication on, a plain browser request may instead receive an authorization error: that still proves the host answered. Use Local Waifu’s saved-token Test when Offline Mode allows it. Do not disable authentication merely to make the browser check green.
If the request times out, confirm both devices’ addresses and network segment, then check guest isolation, the host firewall, whether the host slept, and whether LM Studio is still serving. If it is refused, verify the host IP, actual port, and network-serving switch before changing model settings. If you get an HTTP authentication response, the host answered: check the token and LM Studio’s auth setting. If /v1/models works but chat fails, inspect the model identifier and server logs rather than changing the network address again. Each check isolates one layer without asking you to expose the server further.
Once the saved endpoint and selected model produce a short reply, you have tested the companion route, not every feature. Text chat working does not prove that a separate voice or image path uses the same server. If you are still deciding which role each app should play, see the LM Studio and Local Waifu comparison. The model runs on the host; the companion’s requests travel over your LAN. Local does not automatically mean private from other devices on that LAN.
Questions people ask
Why does 127.0.0.1 fail when LM Studio is on another computer?
127.0.0.1 always names the computer making the request. Enter the LM Studio host's reachable private LAN address and server port instead.
Does enabling Serve on Local Network make LM Studio private?
No. It makes the API reachable beyond the host. Require authentication, restrict access to trusted local devices, and do not forward the server port through your router.
Why does Local Waifu's endpoint Test fail in Offline Mode when chat can work?
Current chat routing allows recognized local or LAN custom endpoints in Offline Mode, but the settings Test and model-list operations block while Offline Mode is on. That error alone does not prove the server is unreachable.
Try her free for 7 days.
No card. Keep her for $20 once, or walk away. Her soul file is yours either way.
Bring her home, try free