Settings → Providers¶
Manages the local LLM endpoints that show up in the Agent pane's picker. See Local LLMs for end-to-end setup.
Listing endpoints¶
The Providers tab lists every configured endpoint with its kind, base URL, and a row of actions:
- Edit — opens the editor sheet pre-filled.
- Trash icon — removes the endpoint and clears its Keychain auth token (if any).
Empty state shows a hint pointing at Add.
If LM Studio's default server is already running on localhost:1234 and no
LM Studio endpoint exists yet, Loom shows an LM Studio server detected
callout. Click Add LM Studio to save the standard local endpoint in one
step. Windows also keeps an Add LM Studio shortcut in the endpoint toolbar
so the standard http://localhost:1234/v1 endpoint is one click away.
Adding an endpoint¶
Click Add to open the editor:
| Field | Notes |
|---|---|
| Display name | Free-form. Shown as the menu group header (Local · <name>). |
| Kind | Ollama, LM Studio, or OpenAI-compatible. Switching kinds swaps the default base URL hint. |
| Base URL | Full URL. Trailing slash is stripped. Ollama defaults to http://localhost:11434; LM Studio and OpenAI-compatible default to http://localhost:1234/v1. |
| Default model / Model | For Ollama and LM Studio, optional fallback if discovery fails. For OpenAI-compatible, required (the model id sent in the request body). |
| Requires auth token | Toggle. When on, reveals a SecureField for a bearer token. |
Test connection¶
Click Test connection before saving:
- Ollama: hits
GET <baseURL>/api/tags. Reports the number of models (or "No models / unreachable"). - LM Studio: hits
/api/v0/modelsfirst, then falls back toGET <baseURL>/models. Reports installed and loaded model counts. - OpenAI-compatible: hits
GET <baseURL>/models. Reports HTTP 200 (or the failure reason).
For LM Studio, the model menu lists loaded models first and includes available
context length, quantization, and architecture details from /api/v0/models.
On Windows, the Agent pane can also start the lms server and load the selected
model from the LM Studio runtime strip.
Test does not save the endpoint. You still have to click Save.
Editing¶
Selecting Edit on an existing row pre-fills the form, including the auth token (read from Keychain). Save overwrites the existing endpoint.
Removing¶
The trash icon:
- Removes the endpoint from
UserDefaults(loom.localEndpoints). - Deletes the matching Keychain item (account
local_endpoint_<UUID>). - Triggers an agent registry refresh so the Agent pane picker drops any descriptors tied to it.
Storage¶
- Endpoint metadata:
UserDefaultsunder keyloom.localEndpoints, JSON-encoded[LocalEndpoint]. - Auth tokens: macOS Keychain or Windows Credential Manager, scoped to the endpoint UUID.
See Keychain keys.
Auto-refresh on save¶
Saving (or removing) an endpoint triggers AgentRegistry.refresh(localEndpoints:). The Agent pane picker updates without an app restart.