fix: prevent duplicate LM Studio models with case-insensitive deduplication

- Keep both listDownloadedModels and listLoaded APIs to support JIT loading
- Implement case-insensitive deduplication to prevent duplicates
- When duplicates are found, prefer loaded model data for accurate runtime info
- Add test coverage for deduplication logic
- Addresses feedback about LM Studio's JIT Model Loading feature (v0.3.5+)

Fixes #6954

Author: daniel-lxs

src/api/providers/fetchers/__tests__/lmstudio.test.ts

@@ -143,6 +143,50 @@ describe("LMStudio Fetcher", () => {
 			expect(result).toEqual({ [mockRawModel.modelKey]: expectedParsedModel })
 		})
 
+		it("should deduplicate models when both downloaded and loaded", async () => {
+			const mockDownloadedModel: LLMInfo = {
+				type: "llm" as const,
+				modelKey: "mistralai/devstral-small-2505",
+				format: "safetensors",
+				displayName: "Devstral Small 2505",
+				path: "mistralai/devstral-small-2505",
+				sizeBytes: 13277565112,
+				architecture: "mistral",
+				vision: false,
+				trainedForToolUse: false,
+				maxContextLength: 131072,
+			}
+
+			const mockLoadedModel: LLMInstanceInfo = {
+				type: "llm",
+				modelKey: "devstral-small-2505", // Different key but should match case-insensitively
+				format: "safetensors",
+				displayName: "Devstral Small 2505",
+				path: "mistralai/devstral-small-2505",
+				sizeBytes: 13277565112,
+				architecture: "mistral",
+				identifier: "mistralai/devstral-small-2505",
+				instanceReference: "RAP5qbeHVjJgBiGFQ6STCuTJ",
+				vision: false,
+				trainedForToolUse: false,
+				maxContextLength: 131072,
+				contextLength: 7161, // Runtime context info
+			}
+
+			mockedAxios.get.mockResolvedValueOnce({ data: { status: "ok" } })
+			mockListDownloadedModels.mockResolvedValueOnce([mockDownloadedModel])
+			mockListLoaded.mockResolvedValueOnce([{ getModelInfo: vi.fn().mockResolvedValueOnce(mockLoadedModel) }])
+
+			const result = await getLMStudioModels(baseUrl)
+
+			// Should only have one model, with the loaded model's runtime info taking precedence
+			expect(Object.keys(result)).toHaveLength(1)
+
+			// The downloaded model's path should be the key, but with loaded model's data
+			const expectedParsedModel = parseLMStudioModel(mockLoadedModel)
+			expect(result[mockDownloadedModel.path]).toEqual(expectedParsedModel)
+		})
+
 		it("should use default baseUrl if an empty string is provided", async () => {
 			const defaultBaseUrl = "http://localhost:1234"
 			const defaultLmsUrl = "ws://localhost:1234"

src/api/providers/fetchers/lmstudio.ts

@@ -81,13 +81,37 @@ export async function getLMStudioModels(baseUrl = "http://localhost:1234"): Prom
 		} catch (error) {
 			console.warn("Failed to list downloaded models, falling back to loaded models only")
 		}
-		// We want to list loaded models *anyway* since they provide valuable extra info (context size)
+
+		// Get loaded models for their runtime info (context size)
 		const loadedModels = (await client.llm.listLoaded().then((models: LLM[]) => {
 			return Promise.all(models.map((m) => m.getModelInfo()))
 		})) as Array<LLMInstanceInfo>
 
+		// Deduplicate loaded models - only add if no existing key contains the model's identifier (case-insensitive)
 		for (const lmstudioModel of loadedModels) {
-			models[lmstudioModel.modelKey] = parseLMStudioModel(lmstudioModel)
+			const modelIdentifier = lmstudioModel.modelKey.toLowerCase()
+
+			// Check if any existing model key contains this loaded model's identifier
+			const isDuplicate = Object.keys(models).some(
+				(existingKey) =>
+					existingKey.toLowerCase().includes(modelIdentifier) ||
+					modelIdentifier.includes(existingKey.toLowerCase()),
+			)
+
+			if (!isDuplicate) {
+				// Use modelKey for loaded models to maintain consistency
+				models[lmstudioModel.modelKey] = parseLMStudioModel(lmstudioModel)
+			} else {
+				// If it's a duplicate, update the existing entry with loaded model info for better context data
+				const existingKey = Object.keys(models).find(
+					(key) => key.toLowerCase().includes(modelIdentifier) || modelIdentifier.includes(key.toLowerCase()),
+				)
+				if (existingKey) {
+					// Update with loaded model data which has more accurate runtime info
+					models[existingKey] = parseLMStudioModel(lmstudioModel)
+				}
+			}
+
 			modelsWithLoadedDetails.add(lmstudioModel.modelKey)
 		}
 	} catch (error) {