Repository navigation
[Feature Request] Multiple LM Studio Providers #42
Description
Hello,
I wanted to have multiple LM Studio providers, and I would think it could be possible if we could use for provider "lmstudio.*" instead of just the default one, and for each add the models discovered.
Like that we can have the use case :
{
"$schema": "https://opencode.ai/config.json",
"provider": {
"lmstudio-local": {
"name": "LM Studio (local)",
"npm": "@ai-sdk/openai-compatible",
"options": {
"baseURL": "http://127.0.0.1:1234/v1",
"includeUsage": true,
"apiKey": "{env:API_LOCAL_LMSTUDIO}",
},
},
"lmstudio-katana": {
"npm": "@ai-sdk/openai-compatible",
"name": "LM Studio (Remote)",
"options": {
"baseURL": "http://<remote>/v1",
"includeUsage": true,
"apiKey": "{env:API_LOCAL_LMSTUDIO_REMOTE}",
},
},
},
"disabled_providers": [],
}
EDIT: Just found the corresponding code
Activity
dgnsrekt commentedon Aug 15, 2026
Cross-linking #50, which is the same topology seen from the other side.
My LM Studio already federates two hosts behind one endpoint, and it reports the same
keyonce per host. Because the plugin builds its model map withObject.fromEntries, the duplicate records collapse and one host'sloaded_instancesare discarded — solimit.contextcan be taken from a host that is not the one serving the request.So multi-host setups are reaching the plugin today, just without a way to represent them. If this issue is resolved by giving routed instances distinct provider or model identities, that would subsume #50; if it is resolved another way, #50 still needs the per-key merge on its own.