Connect LM Studio
LM Studio can remain your desktop model manager. AI Power Grid connects only to its local OpenAI-compatible server; the worker does not import or upload the model.
Install or Start LM Studio
Keep the desktop installation you already use. New operators can install LM Studio from lmstudio.ai/download, download a model inside LM Studio, and load it before enabling the local server. The Grid worker does not import, copy, or keep that model loaded.
Detect the Endpoint
Enable LM Studio’s local server. Its common base endpoint is
http://127.0.0.1:1234; the worker discovers models through /v1/models. If
you chose another port, enter the endpoint shown in LM Studio manually.
Compatible Models
The wizard lists the identifiers reported by LM Studio. Select one exact model and require the local generation test to complete. Detection proves that the server reports a model, not that its prompt template, context limit, or output path works.
Connect and Configure
- Download and start the verified text worker from aipowergrid.io/run.
- Select detected LM Studio, or enter the displayed base endpoint.
- Choose the exact model, run the local generation test, and review its context and output limits.
- Set concurrency and an optional local-time pause schedule.
- Approve the worker-scoped Console connection and wait for Online.
Operator Controls
Use one simultaneous Grid job initially. Desktop sleep, unloading the model, or stopping LM Studio makes the backend unavailable; the worker cannot keep that runtime alive for you.
Schedules and pause windows limit when the Grid may use the selected capacity. They do not inspect every unrelated request made by another local application.
Security Boundary
Keep the LM Studio server on loopback unless you intentionally secure an operator-controlled LAN endpoint. The worker connects outbound to the Grid, does not upload model files, and never requests a wallet private key. Community workers receive plaintext prompts and outputs.
Verify on the Grid
After the dashboard shows Online, verify the exact worker name or ID and advertised model on aipowergrid.io/run. A loaded model and local server do not prove that Core accepted the worker.
Troubleshooting
- Not detected: confirm the local server is enabled and enter its displayed base URL manually.
- Empty or malformed output: test the selected model in LM Studio and check its prompt template before retrying the worker test.
- Worker drops offline: prevent sleep, keep the model loaded, and inspect the local worker logs for the bounded connection error.