feat: add user-configurable search score threshold slider for semantic search (#5027)

packages/types/src/codebase-index.ts

@@ -10,6 +10,7 @@ export const codebaseIndexConfigSchema = z.object({
 	codebaseIndexEmbedderProvider: z.enum(["openai", "ollama", "openai-compatible"]).optional(),
 	codebaseIndexEmbedderBaseUrl: z.string().optional(),
 	codebaseIndexEmbedderModelId: z.string().optional(),
+	codebaseIndexSearchMinScore: z.number().min(0).max(1).optional(),
 })
 
 export type CodebaseIndexConfig = z.infer<typeof codebaseIndexConfigSchema>

src/services/code-index/__tests__/config-manager.spec.ts

@@ -709,6 +709,153 @@ describe("CodeIndexConfigManager", () => {
 				const result = await configManager.loadConfiguration()
 				expect(result.requiresRestart).toBe(false)
 			})
+
+			describe("currentSearchMinScore priority system", () => {
+				it("should return user-configured score when set", async () => {
+					mockContextProxy.getGlobalState.mockReturnValue({
+						codebaseIndexEnabled: true,
+						codebaseIndexQdrantUrl: "http://qdrant.local",
+						codebaseIndexEmbedderProvider: "openai",
+						codebaseIndexEmbedderModelId: "text-embedding-3-small",
+						codebaseIndexSearchMinScore: 0.8, // User setting
+					})
+					mockContextProxy.getSecret.mockImplementation((key: string) => {
+						if (key === "codeIndexOpenAiKey") return "test-key"
+						return undefined
+					})
+
+					await configManager.loadConfiguration()
+					expect(configManager.currentSearchMinScore).toBe(0.8)
+				})
+
+				it("should fall back to model-specific threshold when user setting is undefined", async () => {
+					mockContextProxy.getGlobalState.mockReturnValue({
+						codebaseIndexEnabled: true,
+						codebaseIndexQdrantUrl: "http://qdrant.local",
+						codebaseIndexEmbedderProvider: "ollama",
+						codebaseIndexEmbedderModelId: "nomic-embed-code",
+						// No codebaseIndexSearchMinScore - user hasn't configured it
+					})
+
+					await configManager.loadConfiguration()
+					// nomic-embed-code has a specific threshold of 0.15
+					expect(configManager.currentSearchMinScore).toBe(0.15)
+				})
+
+				it("should fall back to default SEARCH_MIN_SCORE when neither user setting nor model threshold exists", async () => {
+					mockContextProxy.getGlobalState.mockReturnValue({
+						codebaseIndexEnabled: true,
+						codebaseIndexQdrantUrl: "http://qdrant.local",
+						codebaseIndexEmbedderProvider: "openai",
+						codebaseIndexEmbedderModelId: "unknown-model", // Model not in profiles
+						// No codebaseIndexSearchMinScore
+					})
+					mockContextProxy.getSecret.mockImplementation((key: string) => {
+						if (key === "codeIndexOpenAiKey") return "test-key"
+						return undefined
+					})
+
+					await configManager.loadConfiguration()
+					// Should fall back to default SEARCH_MIN_SCORE (0.4)
+					expect(configManager.currentSearchMinScore).toBe(0.4)
+				})
+
+				it("should respect user setting of 0 (edge case)", async () => {
+					mockContextProxy.getGlobalState.mockReturnValue({
+						codebaseIndexEnabled: true,
+						codebaseIndexQdrantUrl: "http://qdrant.local",
+						codebaseIndexEmbedderProvider: "ollama",
+						codebaseIndexEmbedderModelId: "nomic-embed-code",
+						codebaseIndexSearchMinScore: 0, // User explicitly sets 0
+					})
+
+					await configManager.loadConfiguration()
+					// Should return 0, not fall back to model threshold (0.15)
+					expect(configManager.currentSearchMinScore).toBe(0)
+				})
+
+				it("should use model-specific threshold with openai-compatible provider", async () => {
+					mockContextProxy.getGlobalState.mockImplementation((key: string) => {
+						if (key === "codebaseIndexConfig") {
+							return {
+								codebaseIndexEnabled: true,
+								codebaseIndexQdrantUrl: "http://qdrant.local",
+								codebaseIndexEmbedderProvider: "openai-compatible",
+								codebaseIndexEmbedderModelId: "nomic-embed-code",
+								// No codebaseIndexSearchMinScore
+							}
+						}
+						if (key === "codebaseIndexOpenAiCompatibleBaseUrl") return "https://api.example.com/v1"
+						return undefined
+					})
+					mockContextProxy.getSecret.mockImplementation((key: string) => {
+						if (key === "codebaseIndexOpenAiCompatibleApiKey") return "test-api-key"
+						return undefined
+					})
+
+					await configManager.loadConfiguration()
+					// openai-compatible provider also has nomic-embed-code with 0.15 threshold
+					expect(configManager.currentSearchMinScore).toBe(0.15)
+				})
+
+				it("should use default model ID when modelId is not specified", async () => {
+					mockContextProxy.getGlobalState.mockReturnValue({
+						codebaseIndexEnabled: true,
+						codebaseIndexQdrantUrl: "http://qdrant.local",
+						codebaseIndexEmbedderProvider: "openai",
+						// No modelId specified
+						// No codebaseIndexSearchMinScore
+					})
+					mockContextProxy.getSecret.mockImplementation((key: string) => {
+						if (key === "codeIndexOpenAiKey") return "test-key"
+						return undefined
+					})
+
+					await configManager.loadConfiguration()
+					// Should use default model (text-embedding-3-small) threshold (0.4)
+					expect(configManager.currentSearchMinScore).toBe(0.4)
+				})
+
+				it("should handle priority correctly: user > model > default", async () => {
+					// Test 1: User setting takes precedence
+					mockContextProxy.getGlobalState.mockReturnValue({
+						codebaseIndexEnabled: true,
+						codebaseIndexQdrantUrl: "http://qdrant.local",
+						codebaseIndexEmbedderProvider: "ollama",
+						codebaseIndexEmbedderModelId: "nomic-embed-code", // Has 0.15 threshold
+						codebaseIndexSearchMinScore: 0.9, // User overrides
+					})
+
+					await configManager.loadConfiguration()
+					expect(configManager.currentSearchMinScore).toBe(0.9) // User setting wins
+
+					// Test 2: Model threshold when no user setting
+					mockContextProxy.getGlobalState.mockReturnValue({
+						codebaseIndexEnabled: true,
+						codebaseIndexQdrantUrl: "http://qdrant.local",
+						codebaseIndexEmbedderProvider: "ollama",
+						codebaseIndexEmbedderModelId: "nomic-embed-code",
+						// No user setting
+					})
+
+					const newManager = new CodeIndexConfigManager(mockContextProxy)
+					await newManager.loadConfiguration()
+					expect(newManager.currentSearchMinScore).toBe(0.15) // Model threshold
+
+					// Test 3: Default when neither exists
+					mockContextProxy.getGlobalState.mockReturnValue({
+						codebaseIndexEnabled: true,
+						codebaseIndexQdrantUrl: "http://qdrant.local",
+						codebaseIndexEmbedderProvider: "openai",
+						codebaseIndexEmbedderModelId: "custom-unknown-model",
+						// No user setting, unknown model
+					})
+
+					const anotherManager = new CodeIndexConfigManager(mockContextProxy)
+					await anotherManager.loadConfiguration()
+					expect(anotherManager.currentSearchMinScore).toBe(0.4) // Default
+				})
+			})
 		})
 
 		describe("empty/missing API key handling", () => {

src/services/code-index/config-manager.ts

@@ -3,7 +3,7 @@ import { ContextProxy } from "../../core/config/ContextProxy"
 import { EmbedderProvider } from "./interfaces/manager"
 import { CodeIndexConfig, PreviousConfigSnapshot } from "./interfaces/config"
 import { SEARCH_MIN_SCORE } from "./constants"
-import { getDefaultModelId, getModelDimension } from "../../shared/embeddingModels"
+import { getDefaultModelId, getModelDimension, getModelScoreThreshold } from "../../shared/embeddingModels"
 
 /**
  * Manages configuration state and validation for the code indexing feature.
@@ -34,10 +34,10 @@ export class CodeIndexConfigManager {
 		const codebaseIndexConfig = this.contextProxy?.getGlobalState("codebaseIndexConfig") ?? {
 			codebaseIndexEnabled: false,
 			codebaseIndexQdrantUrl: "http://localhost:6333",
-			codebaseIndexSearchMinScore: 0.4,
 			codebaseIndexEmbedderProvider: "openai",
 			codebaseIndexEmbedderBaseUrl: "",
 			codebaseIndexEmbedderModelId: "",
+			codebaseIndexSearchMinScore: undefined,
 		}
 
 		const {
@@ -46,6 +46,7 @@ export class CodeIndexConfigManager {
 			codebaseIndexEmbedderProvider,
 			codebaseIndexEmbedderBaseUrl,
 			codebaseIndexEmbedderModelId,
+			codebaseIndexSearchMinScore,
 		} = codebaseIndexConfig
 
 		const openAiKey = this.contextProxy?.getSecret("codeIndexOpenAiKey") ?? ""
@@ -60,8 +61,8 @@ export class CodeIndexConfigManager {
 		this.isEnabled = codebaseIndexEnabled || false
 		this.qdrantUrl = codebaseIndexQdrantUrl
 		this.qdrantApiKey = qdrantApiKey ?? ""
+		this.searchMinScore = codebaseIndexSearchMinScore
 		this.openAiOptions = { openAiNativeApiKey: openAiKey }
-		this.searchMinScore = SEARCH_MIN_SCORE
 
 		// Set embedder provider with support for openai-compatible
 		if (codebaseIndexEmbedderProvider === "ollama") {
@@ -139,7 +140,7 @@ export class CodeIndexConfigManager {
 				openAiCompatibleOptions: this.openAiCompatibleOptions,
 				qdrantUrl: this.qdrantUrl,
 				qdrantApiKey: this.qdrantApiKey,
-				searchMinScore: this.searchMinScore,
+				searchMinScore: this.currentSearchMinScore,
 			},
 			requiresRestart,
 		}
@@ -294,7 +295,7 @@ export class CodeIndexConfigManager {
 			openAiCompatibleOptions: this.openAiCompatibleOptions,
 			qdrantUrl: this.qdrantUrl,
 			qdrantApiKey: this.qdrantApiKey,
-			searchMinScore: this.searchMinScore,
+			searchMinScore: this.currentSearchMinScore,
 		}
 	}
 
@@ -337,9 +338,18 @@ export class CodeIndexConfigManager {
 	}
 
 	/**
-	 * Gets the configured minimum search score.
+	 * Gets the configured minimum search score based on user setting, model-specific threshold, or fallback.
+	 * Priority: 1) User setting, 2) Model-specific threshold, 3) Default SEARCH_MIN_SCORE constant.
 	 */
-	public get currentSearchMinScore(): number | undefined {
-		return this.searchMinScore
+	public get currentSearchMinScore(): number {
+		// First check if user has configured a custom score threshold
+		if (this.searchMinScore !== undefined) {
+			return this.searchMinScore
+		}
+
+		// Fall back to model-specific threshold
+		const currentModelId = this.modelId ?? getDefaultModelId(this.embedderProvider)
+		const modelSpecificThreshold = getModelScoreThreshold(this.embedderProvider, currentModelId)
+		return modelSpecificThreshold ?? SEARCH_MIN_SCORE
 	}
 }

src/services/code-index/embedders/ollama.ts

@@ -1,5 +1,6 @@
 import { ApiHandlerOptions } from "../../../shared/api"
 import { EmbedderInfo, EmbeddingResponse, IEmbedder } from "../interfaces"
+import { getModelQueryPrefix } from "../../../shared/embeddingModels"
 import { t } from "../../../i18n"
 
 /**
@@ -25,6 +26,12 @@ export class CodeIndexOllamaEmbedder implements IEmbedder {
 		const modelToUse = model || this.defaultModelId
 		const url = `${this.baseUrl}/api/embed` // Endpoint as specified
 
+		// Apply model-specific query prefix if required
+		const queryPrefix = getModelQueryPrefix("ollama", modelToUse)
+		const processedTexts = queryPrefix
+			? texts.map((text) => (text.startsWith(queryPrefix) ? text : `${queryPrefix}${text}`))
+			: texts
+
 		try {
 			// Note: Standard Ollama API uses 'prompt' for single text, not 'input' for array.
 			// Implementing based on user's specific request structure.
@@ -35,7 +42,7 @@ export class CodeIndexOllamaEmbedder implements IEmbedder {
 				},
 				body: JSON.stringify({
 					model: modelToUse,
-					input: texts, // Using 'input' as requested
+					input: processedTexts, // Using 'input' as requested
 				}),
 			})
 

src/services/code-index/embedders/openai-compatible.ts

@@ -6,7 +6,7 @@ import {
 	MAX_BATCH_RETRIES as MAX_RETRIES,
 	INITIAL_RETRY_DELAY_MS as INITIAL_DELAY_MS,
 } from "../constants"
-import { getDefaultModelId } from "../../../shared/embeddingModels"
+import { getDefaultModelId, getModelQueryPrefix } from "../../../shared/embeddingModels"
 import { t } from "../../../i18n"
 
 interface EmbeddingItem {
@@ -59,9 +59,14 @@ export class OpenAICompatibleEmbedder implements IEmbedder {
 	 */
 	async createEmbeddings(texts: string[], model?: string): Promise<EmbeddingResponse> {
 		const modelToUse = model || this.defaultModelId
+
+		// Apply model-specific query prefix if required
+		const queryPrefix = getModelQueryPrefix("openai-compatible", modelToUse)
+		const processedTexts = queryPrefix ? texts.map((text) => `${queryPrefix}${text}`) : texts
+
 		const allEmbeddings: number[][] = []
 		const usage = { promptTokens: 0, totalTokens: 0 }
-		const remainingTexts = [...texts]
+		const remainingTexts = [...processedTexts]
 
 		while (remainingTexts.length > 0) {
 			const currentBatch: string[] = []

src/shared/embeddingModels.ts

@@ -6,6 +6,8 @@ export type EmbedderProvider = "openai" | "ollama" | "openai-compatible" // Add
 
 export interface EmbeddingModelProfile {
 	dimension: number
+	scoreThreshold?: number // Model-specific minimum score threshold for semantic search
+	queryPrefix?: string // Optional prefix required by the model for queries
 	// Add other model-specific properties if needed, e.g., context window size
 }
 
@@ -18,21 +20,31 @@ export type EmbeddingModelProfiles = {
 // Example profiles - expand this list as needed
 export const EMBEDDING_MODEL_PROFILES: EmbeddingModelProfiles = {
 	openai: {
-		"text-embedding-3-small": { dimension: 1536 },
-		"text-embedding-3-large": { dimension: 3072 },
-		"text-embedding-ada-002": { dimension: 1536 },
+		"text-embedding-3-small": { dimension: 1536, scoreThreshold: 0.4 },
+		"text-embedding-3-large": { dimension: 3072, scoreThreshold: 0.4 },
+		"text-embedding-ada-002": { dimension: 1536, scoreThreshold: 0.4 },
 	},
 	ollama: {
-		"nomic-embed-text": { dimension: 768 },
-		"mxbai-embed-large": { dimension: 1024 },
-		"all-minilm": { dimension: 384 },
+		"nomic-embed-text": { dimension: 768, scoreThreshold: 0.4 },
+		"nomic-embed-code": {
+			dimension: 3584,
+			scoreThreshold: 0.15,
+			queryPrefix: "Represent this query for searching relevant code: ",
+		},
+		"mxbai-embed-large": { dimension: 1024, scoreThreshold: 0.4 },
+		"all-minilm": { dimension: 384, scoreThreshold: 0.4 },
 		// Add default Ollama model if applicable, e.g.:
 		// 'default': { dimension: 768 } // Assuming a default dimension
 	},
 	"openai-compatible": {
-		"text-embedding-3-small": { dimension: 1536 },
-		"text-embedding-3-large": { dimension: 3072 },
-		"text-embedding-ada-002": { dimension: 1536 },
+		"text-embedding-3-small": { dimension: 1536, scoreThreshold: 0.4 },
+		"text-embedding-3-large": { dimension: 3072, scoreThreshold: 0.4 },
+		"text-embedding-ada-002": { dimension: 1536, scoreThreshold: 0.4 },
+		"nomic-embed-code": {
+			dimension: 3584,
+			scoreThreshold: 0.15,
+			queryPrefix: "Represent this query for searching relevant code: ",
+		},
 	},
 }
 
@@ -59,6 +71,38 @@ export function getModelDimension(provider: EmbedderProvider, modelId: string):
 	return modelProfile.dimension
 }
 
+/**
+ * Retrieves the score threshold for a given provider and model ID.
+ * @param provider The embedder provider (e.g., "openai").
+ * @param modelId The specific model ID (e.g., "text-embedding-3-small").
+ * @returns The score threshold or undefined if the model is not found.
+ */
+export function getModelScoreThreshold(provider: EmbedderProvider, modelId: string): number | undefined {
+	const providerProfiles = EMBEDDING_MODEL_PROFILES[provider]
+	if (!providerProfiles) {
+		return undefined
+	}
+
+	const modelProfile = providerProfiles[modelId]
+	return modelProfile?.scoreThreshold
+}
+
+/**
+ * Retrieves the query prefix for a given provider and model ID.
+ * @param provider The embedder provider (e.g., "openai").
+ * @param modelId The specific model ID (e.g., "nomic-embed-code").
+ * @returns The query prefix or undefined if the model doesn't require one.
+ */
+export function getModelQueryPrefix(provider: EmbedderProvider, modelId: string): string | undefined {
+	const providerProfiles = EMBEDDING_MODEL_PROFILES[provider]
+	if (!providerProfiles) {
+		return undefined
+	}
+
+	const modelProfile = providerProfiles[modelId]
+	return modelProfile?.queryPrefix
+}
+
 /**
  * Gets the default *specific* embedding model ID based on the provider.
  * Does not include the provider prefix.