Merge pull request #3415 from Kilo-Org/catrielmuller/autocomplete-non-fim

Autocomplete Continue Experimental FIM Endpoint

Author: catrielmuller

src/esbuild.mjs

@@ -64,6 +64,9 @@ async function main() {
 					// Copy walkthrough files to dist directory
 					copyPaths([["walkthrough", "walkthrough"]], srcDir, distDir)
 
+					// Copy tree-sitter files to dist directory
+					copyPaths([["services/continuedev/tree-sitter", "tree-sitter"]], srcDir, distDir)
+
 					// Copy JSDOM xhr-sync-worker.js to fix runtime resolution
 					const jsdomWorkerDest = path.join(distDir, "xhr-sync-worker.js")
 

src/services/autocomplete/AutocompleteModel.ts

@@ -1,5 +1,4 @@
 import {
-	AUTOCOMPLETE_PROVIDER_MODELS,
 	AutocompleteProviderKey,
 	defaultProviderUsabilityChecker,
 	getKiloBaseUriFromToken,
@@ -14,7 +13,15 @@ import { ApiStreamChunk } from "../../api/transform/stream"
 import { ILLM, LLMOptions } from "../continuedev/core/index.js"
 import { DEFAULT_AUTOCOMPLETE_OPTS } from "../continuedev/core/util/parameters.js"
 import Mistral from "../continuedev/core/llm/llms/Mistral"
-import { OpenAI } from "../continuedev/core/llm/llms/OpenAI"
+import OpenRouter from "../continuedev/core/llm/llms/OpenRouter"
+import KiloCode from "../continuedev/core/llm/llms/KiloCode"
+
+export const AUTOCOMPLETE_PROVIDER_MODELS = {
+	mistral: "codestral-2501",
+	kilocode: "codestral-2501",
+	openrouter: "mistralai/codestral-2501",
+	bedrock: "mistral.codestral-2501-v1:0",
+} as const
 
 export class AutocompleteModel {
 	private apiHandler: ApiHandler | null = null
@@ -118,6 +125,16 @@ export class AutocompleteModel {
 					useCache: false, // Disable caching for autocomplete
 				},
 				uniqueId: `autocomplete-${provider}-${Date.now()}`,
+				// Add env for KiloCode metadata (organizationId and tester suppression)
+				env: config.organizationId
+					? {
+							kilocodeOrganizationId: config.organizationId,
+							// Add tester suppression if configured
+							...(this.profile?.kilocodeTesterWarningsDisabledUntil && {
+								kilocodeTesterWarningsDisabledUntil: this.profile.kilocodeTesterWarningsDisabledUntil,
+							}),
+						}
+					: undefined,
 			}
 
 			// Create appropriate LLM instance based on provider
@@ -131,7 +148,12 @@ export class AutocompleteModel {
 	/**
 	 * Extracts provider-specific configuration (API key, base URL, model) from this.profile
 	 */
-	private extractProviderConfig(): { apiKey: string; apiBase: string; model: string } | null {
+	private extractProviderConfig(): {
+		apiKey: string
+		apiBase: string
+		model: string
+		organizationId?: string
+	} | null {
 		if (!this.profile?.apiProvider) {
 			return null
 		}
@@ -158,8 +180,9 @@ export class AutocompleteModel {
 				}
 				return {
 					apiKey: this.profile.kilocodeToken,
-					apiBase: `${getKiloBaseUriFromToken(this.profile.kilocodeToken)}/openrouter/api/v1`,
+					apiBase: `${getKiloBaseUriFromToken(this.profile.kilocodeToken)}/api/openrouter/v1`,
 					model,
+					organizationId: this.profile.kilocodeOrganizationId,
 				}
 
 			case "openrouter":
@@ -194,9 +217,12 @@ export class AutocompleteModel {
 				return new Mistral(options)
 
 			case "kilocode":
+				// Use dedicated KiloCode class with custom headers and routing
+				return new KiloCode(options)
+
 			case "openrouter":
-				// Both use OpenAI-compatible API
-				return new OpenAI(options)
+				// Use standard OpenRouter
+				return new OpenRouter(options)
 
 			case "bedrock":
 				// Bedrock would need a custom implementation

src/services/autocomplete/AutocompleteProvider.ts

@@ -13,6 +13,7 @@ import { ClineProvider } from "../../core/webview/ClineProvider"
 import { MinimalConfigProvider } from "../continuedev/core/autocomplete/MinimalConfig"
 import { VsCodeIde } from "../continuedev/core/vscode-test-harness/src/VSCodeIde"
 import { ContinueCompletionProvider } from "../continuedev/core/vscode-test-harness/src/autocomplete/completionProvider"
+import OpenRouter from "../continuedev/core/llm/llms/OpenRouter"
 
 export class AutocompleteProvider {
 	private static instance: AutocompleteProvider | null = null

src/services/autocomplete/__tests__/AutocompleteModel.spec.ts

@@ -31,7 +31,7 @@ describe("AutocompleteModel", () => {
 				expect(result).toBeDefined()
 				expect(Mistral).toHaveBeenCalledWith(
 					expect.objectContaining({
-						model: "codestral-latest",
+						model: "codestral-2501",
 						apiKey: "test-mistral-key",
 						apiBase: "https://custom.mistral.ai/v1/",
 						contextLength: 32000,
@@ -78,7 +78,7 @@ describe("AutocompleteModel", () => {
 				expect(result).toBeDefined()
 				expect(OpenAI).toHaveBeenCalledWith(
 					expect.objectContaining({
-						model: "mistralai/codestral-2508",
+						model: "codestral-2501",
 						apiKey: "test-kilocode-token",
 					}),
 				)
@@ -109,7 +109,7 @@ describe("AutocompleteModel", () => {
 				expect(result).toBeDefined()
 				expect(OpenAI).toHaveBeenCalledWith(
 					expect.objectContaining({
-						model: "mistralai/codestral-2508",
+						model: "mistralai/codestral-2501",
 						apiKey: "test-openrouter-key",
 						apiBase: "https://custom.openrouter.ai/api/v1",
 					}),

src/services/continuedev/core/autocomplete/CompletionProvider.ts

@@ -55,7 +55,6 @@ export class CompletionProvider {
 			return undefined
 		}
 
-
 		// Temporary fix for JetBrains autocomplete bug as described in https://github.com/continuedev/continue/pull/3022
 		if (llm.model === undefined && llm.completionOptions?.model !== undefined) {
 			llm.model = llm.completionOptions.model
@@ -76,7 +75,6 @@ export class CompletionProvider {
 			llm.useLegacyCompletionsEndpoint = true
 		}
 
-		console.log('using LLM', llm)
 		return llm
 	}
 
@@ -123,7 +121,7 @@ export class CompletionProvider {
 
 		// Enable static contextualization if defined.
 		if (config?.experimental?.enableStaticContextualization) {
-			options.experimental_enableStaticContextualization = true
+			options.experimental_enableStaticContextualization = false
 		}
 
 		return options
@@ -138,7 +136,6 @@ export class CompletionProvider {
 			// Create abort signal if not given
 			if (!token) {
 				const controller = this.loggingService.createAbortController(input.completionId)
-				console.log('creating abort token because none given')
 				token = controller.signal
 			}
 			const startTime = Date.now()
@@ -214,12 +211,9 @@ export class CompletionProvider {
 
 				// Don't postprocess if aborted
 				if (token.aborted) {
-					console.log('aborted')
 					return undefined
 				}
 
-				console.log('raw completion', completion)
-
 				const processedCompletion = helper.options.transform
 					? postprocessCompletion({
 							completion,

src/services/continuedev/core/autocomplete/templating/AutocompleteTemplate.ts

@@ -373,15 +373,16 @@ export function getTemplateForModel(model: string): AutocompleteTemplate {
 		return codegeexFimTemplate
 	}
 
-	if (
-		lowerCaseModel.includes("gpt") ||
-		lowerCaseModel.includes("davinci-002") ||
-		lowerCaseModel.includes("claude") ||
-		lowerCaseModel.includes("granite3") ||
-		lowerCaseModel.includes("granite-3")
-	) {
-		return holeFillerTemplate
-	}
-
-	return stableCodeFimTemplate
+	// if (
+	// 	lowerCaseModel.includes("gpt") ||
+	// 	lowerCaseModel.includes("davinci-002") ||
+	// 	lowerCaseModel.includes("claude") ||
+	// 	lowerCaseModel.includes("granite3") ||
+	// 	lowerCaseModel.includes("granite-3")
+	// ) {
+	// Default Fallback Mode
+	return holeFillerTemplate
+	//}
+
+	//return stableCodeFimTemplate
 }

src/services/continuedev/core/index.d.ts

@@ -333,7 +333,6 @@ export interface ILLMLogger {
 
 export interface LLMOptions {
 	model: string
-
 	title?: string
 	uniqueId?: string
 	autocompleteOptions?: Partial<TabAutocompleteOptions>
@@ -346,6 +345,7 @@ export interface LLMOptions {
 	useLegacyCompletionsEndpoint?: boolean
 	capabilities?: ModelCapability
 	env?: Record<string, string | number | boolean>
+	promptTemplates?: Partial<Record<keyof PromptTemplates, PromptTemplate>>
 }
 
 type RequireAtLeastOne<T, Keys extends keyof T = keyof T> = Pick<T, Exclude<keyof T, Keys>> &
@@ -511,6 +511,9 @@ export interface BaseCompletionOptions {
 	maxTokens?: number
 	raw?: boolean
 	stream?: boolean
+	reasoning?: boolean
+	reasoningBudgetTokens?: number
+	promptCaching?: boolean
 }
 
 export interface ModelCapability {
@@ -617,18 +620,18 @@ export class SignatureInformation {
 	 * The label of this signature. Will be shown in
 	 * the UI.
 	 */
-	label: strin
-  /**
-   * The parameters of this signature.
-   */
-  parameters: ParameterInformation[];
-
-  /**
-   * The index of the active parameter.
-   *
-   * If provided, this is used in place of {@linkcode SignatureHelp.activeParameter}.
-   */
-  activeParameter?: number;g
+	label: string
+	/**
+	 * The parameters of this signature.
+	 */
+	parameters: ParameterInformation[]
+
+	/**
+	 * The index of the active parameter.
+	 *
+	 * If provided, this is used in place of {@linkcode SignatureHelp.activeParameter}.
+	 */
+	activeParameter?: number
 }
 
 export type ConfigMergeType = "merge" | "overwrite"

src/services/continuedev/core/llm/index.ts

@@ -15,7 +15,7 @@ import {
 	TabAutocompleteOptions,
 	Usage,
 } from "../index.js"
-import type { ILLMInteractionLog, ILLMLogger } from "../index.js"
+import type { CacheBehavior, ILLMInteractionLog, ILLMLogger } from "../index.js"
 import { mergeJson } from "../util/merge.js"
 import { renderChatMessage } from "../util/messageContent.js"
 import { TokensBatchingService } from "../util/TokensBatchingService.js"
@@ -86,6 +86,7 @@ export abstract class BaseLLM implements ILLM {
 	llmRequestHook?: (model: string, prompt: string) => any
 	apiKey?: string
 	apiBase?: string
+	cacheBehavior?: CacheBehavior
 	capabilities?: ModelCapability
 
 	lastRequestId: string | undefined