feat: Add custom model dimension support and improve codebase indexing

- Add custom embedding dimension configuration for unknown models
- Fix Ollama API implementation to use correct prompt-based structure
- Enhance OpenAI embedder with base URL and dimension support
- Improve UI with provider-specific settings and better model management
- Add comprehensive VSCode settings for codebase indexing feature

**Core Changes:**
- Add `codebaseIndexEmbedderDimension` field to schemas and config
- Implement proper dimension validation and fallback logic
- Support manual dimension override for custom/unknown models

**API Fixes:**
- Fix Ollama embedder to use individual `prompt` requests instead of batch `input`
- Correct response parsing to use `embedding` (singular) field
- Add OpenAI base URL support for proxies and Azure endpoints
- Implement dimension parameter for OpenAI embedding requests

**UI Improvements:**
- Reorganize settings with provider-specific sections
- Replace model dropdown with flexible text input
- Add dimension input field with validation
- Improve placeholder text and help descriptions
- Add OpenAI base URL configuration field

**Configuration:**
- Add VSCode settings for all codebase indexing options
- Enhance config manager with proper dimension handling
- Improve service factory with better error messages
- Add comprehensive validation for custom models

Author: DanielusG

src/core/webview/webviewMessageHandler.ts

@@ -1327,6 +1327,7 @@ export const webviewMessageHandler = async (provider: ClineProvider, message: We
 				codebaseIndexEmbedderProvider: "openai",
 				codebaseIndexEmbedderBaseUrl: "",
 				codebaseIndexEmbedderModelId: "",
+				codebaseIndexEmbedderDimension: null,
 			}
 			await updateGlobalState("codebaseIndexConfig", codebaseIndexConfig)
 

src/exports/roo-code.d.ts

@@ -86,6 +86,7 @@ type GlobalSettings = {
 				codebaseIndexEmbedderProvider?: ("openai" | "ollama") | undefined
 				codebaseIndexEmbedderBaseUrl?: string | undefined
 				codebaseIndexEmbedderModelId?: string | undefined
+				codebaseIndexEmbedderDimension?: (number | null) | undefined
 		  }
 		| undefined
 	alwaysAllowWrite?: boolean | undefined
@@ -865,6 +866,7 @@ type IpcMessage =
 											codebaseIndexEmbedderProvider?: ("openai" | "ollama") | undefined
 											codebaseIndexEmbedderBaseUrl?: string | undefined
 											codebaseIndexEmbedderModelId?: string | undefined
+											codebaseIndexEmbedderDimension?: (number | null) | undefined
 									  }
 									| undefined
 								alwaysAllowWrite?: boolean | undefined
@@ -1377,6 +1379,7 @@ type TaskCommand =
 								codebaseIndexEmbedderProvider?: ("openai" | "ollama") | undefined
 								codebaseIndexEmbedderBaseUrl?: string | undefined
 								codebaseIndexEmbedderModelId?: string | undefined
+								codebaseIndexEmbedderDimension?: (number | null) | undefined
 						  }
 						| undefined
 					alwaysAllowWrite?: boolean | undefined

src/exports/types.ts

@@ -86,6 +86,7 @@ type GlobalSettings = {
 				codebaseIndexEmbedderProvider?: ("openai" | "ollama") | undefined
 				codebaseIndexEmbedderBaseUrl?: string | undefined
 				codebaseIndexEmbedderModelId?: string | undefined
+				codebaseIndexEmbedderDimension?: (number | null) | undefined
 		  }
 		| undefined
 	alwaysAllowWrite?: boolean | undefined
@@ -879,6 +880,7 @@ type IpcMessage =
 											codebaseIndexEmbedderProvider?: ("openai" | "ollama") | undefined
 											codebaseIndexEmbedderBaseUrl?: string | undefined
 											codebaseIndexEmbedderModelId?: string | undefined
+											codebaseIndexEmbedderDimension?: (number | null) | undefined
 									  }
 									| undefined
 								alwaysAllowWrite?: boolean | undefined
@@ -1393,6 +1395,7 @@ type TaskCommand =
 								codebaseIndexEmbedderProvider?: ("openai" | "ollama") | undefined
 								codebaseIndexEmbedderBaseUrl?: string | undefined
 								codebaseIndexEmbedderModelId?: string | undefined
+								codebaseIndexEmbedderDimension?: (number | null) | undefined
 						  }
 						| undefined
 					alwaysAllowWrite?: boolean | undefined

src/package.json

@@ -313,6 +313,57 @@
 					"type": "string",
 					"default": "",
 					"description": "%settings.customStoragePath.description%"
+				},
+				"roo-cline.codebaseIndexEnabled": {
+					"type": "boolean",
+					"default": false,
+					"description": "Enable codebase indexing for semantic search.",
+					"scope": "resource"
+				},
+				"roo-cline.codebaseIndexEmbedderProvider": {
+					"type": "string",
+					"enum": [
+						"openai",
+						"ollama"
+					],
+					"default": "openai",
+					"description": "Select the embedding provider (OpenAI or Ollama).",
+					"scope": "resource"
+				},
+				"roo-cline.codebaseIndexEmbedderBaseUrl": {
+					"type": "string",
+					"default": "",
+					"description": "Optional: Base URL for the selected embedding provider API (e.g., for proxies, Azure OpenAI, or self-hosted Ollama).",
+					"scope": "resource"
+				},
+				"roo-cline.codebaseIndexEmbedderModelId": {
+					"type": "string",
+					"default": "",
+					"description": "Optional: Specify a custom embedding model ID. Leave empty to use the provider's default (e.g., text-embedding-3-small for OpenAI).",
+					"scope": "resource"
+				},
+				"roo-cline.codebaseIndexEmbedderDimension": {
+					"type": [
+						"number",
+						"null"
+					],
+					"default": null,
+					"description": "Optional: Specify the embedding dimension for your custom model ID if it's not automatically recognized. Required only if using an unknown model ID.",
+					"scope": "resource"
+				},
+				"roo-cline.codebaseIndexQdrantUrl": {
+					"type": "string",
+					"default": "http://localhost:6333",
+					"description": "URL for the Qdrant vector database instance.",
+					"scope": "resource"
+				},
+				"roo-cline.codebaseIndexSearchMinScore": {
+					"type": "number",
+					"default": 0.4,
+					"minimum": 0,
+					"maximum": 1,
+					"description": "Minimum similarity score (0-1) for codebase search results.",
+					"scope": "resource"
 				}
 			}
 		}

src/schemas/index.ts

@@ -232,6 +232,7 @@ export const codebaseIndexConfigSchema = z.object({
 	codebaseIndexEmbedderProvider: z.enum(["openai", "ollama"]).optional(),
 	codebaseIndexEmbedderBaseUrl: z.string().optional(),
 	codebaseIndexEmbedderModelId: z.string().optional(),
+	codebaseIndexEmbedderDimension: z.number().nullish(),
 })
 
 export type CodebaseIndexConfig = z.infer<typeof codebaseIndexConfigSchema>
@@ -244,7 +245,7 @@ export const codebaseIndexModelsSchema = z.object({
 export type CodebaseIndexModels = z.infer<typeof codebaseIndexModelsSchema>
 
 export const codebaseIndexProviderSchema = z.object({
-  codeIndexOpenAiKey: z.string().optional(),
+	codeIndexOpenAiKey: z.string().optional(),
 	codeIndexQdrantApiKey: z.string().optional(),
 })
 
@@ -661,7 +662,7 @@ export const providerSettingsSchema = z.object({
 	...groqSchema.shape,
 	...chutesSchema.shape,
 	...litellmSchema.shape,
-  ...codebaseIndexProviderSchema.shape
+	...codebaseIndexProviderSchema.shape,
 })
 
 export type ProviderSettings = z.infer<typeof providerSettingsSchema>

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

@@ -17,6 +17,7 @@ export class CodeIndexConfigManager {
 	private qdrantUrl?: string = "http://localhost:6333"
 	private qdrantApiKey?: string
 	private searchMinScore?: number
+	private dimension?: number // Added for custom model dimension
 
 	constructor(private readonly contextProxy: ContextProxy) {}
 
@@ -35,6 +36,7 @@ export class CodeIndexConfigManager {
 			qdrantUrl?: string
 			qdrantApiKey?: string
 			searchMinScore?: number
+			dimension?: number // Add dimension to return type
 		}
 		requiresRestart: boolean
 	}> {
@@ -47,6 +49,7 @@ export class CodeIndexConfigManager {
 			ollamaBaseUrl: this.ollamaOptions?.ollamaBaseUrl,
 			qdrantUrl: this.qdrantUrl,
 			qdrantApiKey: this.qdrantApiKey,
+			dimension: this.dimension, // Added dimension to snapshot
 		}
 
 		let codebaseIndexConfig = this.contextProxy?.getGlobalState("codebaseIndexConfig") ?? {
@@ -56,31 +59,42 @@ export class CodeIndexConfigManager {
 			codebaseIndexEmbedderProvider: "openai",
 			codebaseIndexEmbedderBaseUrl: "",
 			codebaseIndexEmbedderModelId: "",
+			codebaseIndexEmbedderDimension: null,
 		}
 
+		// Destructure known properties, access dimension defensively
 		const {
 			codebaseIndexEnabled,
 			codebaseIndexQdrantUrl,
 			codebaseIndexEmbedderProvider,
 			codebaseIndexEmbedderBaseUrl,
 			codebaseIndexEmbedderModelId,
 		} = codebaseIndexConfig
+		const codebaseIndexEmbedderDimension = (codebaseIndexConfig as any)?.codebaseIndexEmbedderDimension
 
 		const openAiKey = this.contextProxy?.getSecret("codeIndexOpenAiKey") ?? ""
 		const qdrantApiKey = this.contextProxy?.getSecret("codeIndexQdrantApiKey") ?? ""
 
 		this.isEnabled = codebaseIndexEnabled || false
 		this.qdrantUrl = codebaseIndexQdrantUrl
 		this.qdrantApiKey = qdrantApiKey ?? ""
-		this.openAiOptions = { openAiNativeApiKey: openAiKey }
 		this.searchMinScore = SEARCH_MIN_SCORE
 
 		this.embedderProvider = codebaseIndexEmbedderProvider === "ollama" ? "ollama" : "openai"
+		this.openAiOptions = {
+			openAiNativeApiKey: openAiKey,
+			openAiBaseUrl: this.embedderProvider === "openai" ? codebaseIndexEmbedderBaseUrl : undefined,
+		}
 		this.modelId = codebaseIndexEmbedderModelId || undefined
 
 		this.ollamaOptions = {
 			ollamaBaseUrl: codebaseIndexEmbedderBaseUrl,
 		}
+		// Parse and store dimension, ensuring it's a positive number or undefined
+		this.dimension =
+			typeof codebaseIndexEmbedderDimension === "number" && codebaseIndexEmbedderDimension > 0
+				? codebaseIndexEmbedderDimension
+				: undefined
 
 		return {
 			configSnapshot: previousConfigSnapshot,
@@ -94,6 +108,7 @@ export class CodeIndexConfigManager {
 				qdrantUrl: this.qdrantUrl,
 				qdrantApiKey: this.qdrantApiKey,
 				searchMinScore: this.searchMinScore,
+				dimension: this.dimension,
 			},
 			requiresRestart: this._didConfigChangeRequireRestart(previousConfigSnapshot),
 		}
@@ -103,7 +118,6 @@ export class CodeIndexConfigManager {
 	 * Checks if the service is properly configured based on the embedder type.
 	 */
 	public isConfigured(): boolean {
-
 		if (this.embedderProvider === "openai") {
 			const openAiKey = this.openAiOptions?.openAiNativeApiKey
 			const qdrantUrl = this.qdrantUrl
@@ -160,6 +174,9 @@ export class CodeIndexConfigManager {
 			if (prev.qdrantUrl !== this.qdrantUrl || prev.qdrantApiKey !== this.qdrantApiKey) {
 				return true
 			}
+
+			// Check dimension change
+			if (prev.dimension !== this.dimension) return true
 		}
 
 		return false
@@ -179,6 +196,7 @@ export class CodeIndexConfigManager {
 			qdrantUrl: this.qdrantUrl,
 			qdrantApiKey: this.qdrantApiKey,
 			searchMinScore: this.searchMinScore,
+			dimension: this.dimension,
 		}
 	}
 

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

@@ -22,50 +22,55 @@ export class CodeIndexOllamaEmbedder implements IEmbedder {
 	 */
 	async createEmbeddings(texts: string[], model?: string): Promise<EmbeddingResponse> {
 		const modelToUse = model || this.defaultModelId
-		const url = `${this.baseUrl}/api/embed` // Endpoint as specified
+		const url = `${this.baseUrl}/api/embed`
 
 		try {
-			// Note: Standard Ollama API uses 'prompt' for single text, not 'input' for array.
-			// Implementing based on user's specific request structure.
-			const response = await fetch(url, {
-				method: "POST",
-				headers: {
-					"Content-Type": "application/json",
-				},
-				body: JSON.stringify({
+			// Ollama API processes one text at a time using 'prompt' field
+			const embeddings: number[][] = []
+
+			for (const text of texts) {
+				const requestBody = {
 					model: modelToUse,
-					input: texts, // Using 'input' as requested
-				}),
-			})
+					prompt: text,
+				}
+
+				const response = await fetch(url, {
+					method: "POST",
+					headers: {
+						"Content-Type": "application/json",
+					},
+					body: JSON.stringify(requestBody),
+				})
 
-			if (!response.ok) {
-				let errorBody = "Could not read error body"
-				try {
-					errorBody = await response.text()
-				} catch (e) {
-					// Ignore error reading body
+				if (!response.ok) {
+					let errorBody = "Could not read error body"
+					try {
+						errorBody = await response.text()
+					} catch (e) {
+						// Ignore error reading body
+					}
+					throw new Error(
+						`Ollama API request failed with status ${response.status} ${response.statusText}: ${errorBody}`,
+					)
 				}
-				throw new Error(
-					`Ollama API request failed with status ${response.status} ${response.statusText}: ${errorBody}`,
-				)
-			}
 
-			const data = await response.json()
+				const data = await response.json()
+
+				// Extract embedding using 'embedding' key (singular)
+				const embedding = data.embedding
+				if (!embedding || !Array.isArray(embedding)) {
+					throw new Error(
+						'Invalid response structure from Ollama API: "embedding" array not found or not an array.',
+					)
+				}
 
-			// Extract embeddings using 'embeddings' key as requested
-			const embeddings = data.embeddings
-			if (!embeddings || !Array.isArray(embeddings)) {
-				throw new Error(
-					'Invalid response structure from Ollama API: "embeddings" array not found or not an array.',
-				)
+				embeddings.push(embedding)
 			}
 
 			return {
 				embeddings: embeddings,
 			}
 		} catch (error: any) {
-			// Log the original error for debugging purposes
-			console.error("Ollama embedding failed:", error)
 			// Re-throw a more specific error for the caller
 			throw new Error(`Ollama embedding failed: ${error.message}`)
 		}

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

@@ -15,15 +15,21 @@ import {
 export class OpenAiEmbedder extends OpenAiNativeHandler implements IEmbedder {
 	private embeddingsClient: OpenAI
 	private readonly defaultModelId: string
+	private readonly configuredDimension?: number
 
 	/**
 	 * Creates a new OpenAI embedder
-	 * @param options API handler options
+	 * @param options API handler options including optional model ID and dimension
 	 */
-	constructor(options: ApiHandlerOptions & { openAiEmbeddingModelId?: string }) {
+	constructor(options: ApiHandlerOptions & { openAiEmbeddingModelId?: string; embeddingDimension?: number }) {
 		super(options)
 		const apiKey = this.options.openAiNativeApiKey ?? "not-provided"
-		this.embeddingsClient = new OpenAI({ apiKey })
+		const baseURL = this.options.openAiBaseUrl
+		this.configuredDimension = options.embeddingDimension
+		this.embeddingsClient = new OpenAI({
+			apiKey,
+			...(baseURL && { baseURL }),
+		})
 		this.defaultModelId = options.openAiEmbeddingModelId || "text-embedding-3-small"
 	}
 
@@ -98,10 +104,15 @@ export class OpenAiEmbedder extends OpenAiNativeHandler implements IEmbedder {
 	): Promise<{ embeddings: number[][]; usage: { promptTokens: number; totalTokens: number } }> {
 		for (let attempts = 0; attempts < MAX_RETRIES; attempts++) {
 			try {
-				const response = await this.embeddingsClient.embeddings.create({
+				const params: OpenAI.Embeddings.EmbeddingCreateParams = {
 					input: batchTexts,
 					model: model,
-				})
+				}
+				if (this.configuredDimension) {
+					params.dimensions = this.configuredDimension
+				}
+
+				const response = await this.embeddingsClient.embeddings.create(params)
 
 				return {
 					embeddings: response.data.map((item) => item.embedding),

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

@@ -14,6 +14,7 @@ export interface CodeIndexConfig {
 	qdrantUrl?: string
 	qdrantApiKey?: string
 	searchMinScore?: number
+	dimension?: number
 }
 
 /**
@@ -28,4 +29,5 @@ export type PreviousConfigSnapshot = {
 	ollamaBaseUrl?: string
 	qdrantUrl?: string
 	qdrantApiKey?: string
+	dimension?: number
 }