Merge pull request #239 from RooVetGit/deepseek

Add DeepSeek to the list of providers

.changeset/modern-carrots-applaud.md

@@ -0,0 +1,5 @@
+---
+"roo-cline": patch
+---
+
+Add the DeepSeek provider

README.md

@@ -13,6 +13,7 @@ A fork of Cline, an autonomous coding agent, with some additional experimental f
 - Includes current time in the system prompt
 - Uses a file system watcher to more reliably watch for file system changes
 - Language selection for Cline's communication (English, Japanese, Spanish, French, German, and more)
+- Support for DeepSeek V3
 - Support for Meta 3, 3.1, and 3.2 models via AWS Bedrock
 - Support for listing models from OpenAI-compatible providers
 - Per-tool MCP auto-approval

src/api/index.ts

@@ -9,6 +9,7 @@ import { OllamaHandler } from "./providers/ollama"
 import { LmStudioHandler } from "./providers/lmstudio"
 import { GeminiHandler } from "./providers/gemini"
 import { OpenAiNativeHandler } from "./providers/openai-native"
+import { DeepSeekHandler } from "./providers/deepseek"
 import { ApiStream } from "./transform/stream"
 
 export interface SingleCompletionHandler {
@@ -41,6 +42,8 @@ export function buildApiHandler(configuration: ApiConfiguration): ApiHandler {
 			return new GeminiHandler(options)
 		case "openai-native":
 			return new OpenAiNativeHandler(options)
+		case "deepseek":
+			return new DeepSeekHandler(options)
 		default:
 			return new AnthropicHandler(options)
 	}

src/api/providers/__tests__/deepseek.test.ts

@@ -0,0 +1,167 @@
+import { DeepSeekHandler } from '../deepseek'
+import { ApiHandlerOptions } from '../../../shared/api'
+import OpenAI from 'openai'
+import { Anthropic } from '@anthropic-ai/sdk'
+
+// Mock dependencies
+jest.mock('openai')
+
+describe('DeepSeekHandler', () => {
+
+    const mockOptions: ApiHandlerOptions = {
+        deepSeekApiKey: 'test-key',
+        deepSeekModelId: 'deepseek-chat',
+    }
+
+    beforeEach(() => {
+        jest.clearAllMocks()
+    })
+
+    test('constructor initializes with correct options', () => {
+        const handler = new DeepSeekHandler(mockOptions)
+        expect(handler).toBeInstanceOf(DeepSeekHandler)
+        expect(OpenAI).toHaveBeenCalledWith({
+            baseURL: 'https://api.deepseek.com/v1',
+            apiKey: mockOptions.deepSeekApiKey,
+        })
+    })
+
+    test('getModel returns correct model info', () => {
+        const handler = new DeepSeekHandler(mockOptions)
+        const result = handler.getModel()
+        
+        expect(result).toEqual({
+            id: mockOptions.deepSeekModelId,
+            info: expect.objectContaining({
+                maxTokens: 8192,
+                contextWindow: 64000,
+                supportsPromptCache: false,
+                supportsImages: false,
+                inputPrice: 0.014,
+                outputPrice: 0.28,
+            })
+        })
+    })
+
+    test('getModel returns default model info when no model specified', () => {
+        const handler = new DeepSeekHandler({ deepSeekApiKey: 'test-key' })
+        const result = handler.getModel()
+        
+        expect(result.id).toBe('deepseek-chat')
+        expect(result.info.maxTokens).toBe(8192)
+    })
+
+    test('createMessage handles string content correctly', async () => {
+        const handler = new DeepSeekHandler(mockOptions)
+        const mockStream = {
+            async *[Symbol.asyncIterator]() {
+                yield {
+                    choices: [{
+                        delta: {
+                            content: 'test response'
+                        }
+                    }]
+                }
+            }
+        }
+
+        const mockCreate = jest.fn().mockResolvedValue(mockStream)
+        ;(OpenAI as jest.MockedClass<typeof OpenAI>).prototype.chat = {
+            completions: { create: mockCreate }
+        } as any
+
+        const systemPrompt = 'test system prompt'
+        const messages: Anthropic.Messages.MessageParam[] = [
+            { role: 'user', content: 'test message' }
+        ]
+
+        const generator = handler.createMessage(systemPrompt, messages)
+        const chunks = []
+        
+        for await (const chunk of generator) {
+            chunks.push(chunk)
+        }
+
+        expect(chunks).toHaveLength(1)
+        expect(chunks[0]).toEqual({
+            type: 'text',
+            text: 'test response'
+        })
+
+        expect(mockCreate).toHaveBeenCalledWith(expect.objectContaining({
+            model: mockOptions.deepSeekModelId,
+            messages: [
+                { role: 'system', content: systemPrompt },
+                { role: 'user', content: 'test message' }
+            ],
+            temperature: 0,
+            stream: true,
+            max_tokens: 8192,
+            stream_options: { include_usage: true }
+        }))
+    })
+
+    test('createMessage handles complex content correctly', async () => {
+        const handler = new DeepSeekHandler(mockOptions)
+        const mockStream = {
+            async *[Symbol.asyncIterator]() {
+                yield {
+                    choices: [{
+                        delta: {
+                            content: 'test response'
+                        }
+                    }]
+                }
+            }
+        }
+
+        const mockCreate = jest.fn().mockResolvedValue(mockStream)
+        ;(OpenAI as jest.MockedClass<typeof OpenAI>).prototype.chat = {
+            completions: { create: mockCreate }
+        } as any
+
+        const systemPrompt = 'test system prompt'
+        const messages: Anthropic.Messages.MessageParam[] = [
+            {
+                role: 'user',
+                content: [
+                    { type: 'text', text: 'part 1' },
+                    { type: 'text', text: 'part 2' }
+                ]
+            }
+        ]
+
+        const generator = handler.createMessage(systemPrompt, messages)
+        await generator.next()
+
+        expect(mockCreate).toHaveBeenCalledWith(expect.objectContaining({
+            messages: [
+                { role: 'system', content: systemPrompt },
+                {
+                    role: 'user',
+                    content: [
+                        { type: 'text', text: 'part 1' },
+                        { type: 'text', text: 'part 2' }
+                    ]
+                }
+            ]
+        }))
+    })
+
+    test('createMessage handles API errors', async () => {
+        const handler = new DeepSeekHandler(mockOptions)
+        const mockStream = {
+            async *[Symbol.asyncIterator]() {
+                throw new Error('API Error')
+            }
+        }
+
+        const mockCreate = jest.fn().mockResolvedValue(mockStream)
+        ;(OpenAI as jest.MockedClass<typeof OpenAI>).prototype.chat = {
+            completions: { create: mockCreate }
+        } as any
+
+        const generator = handler.createMessage('test', [])
+        await expect(generator.next()).rejects.toThrow('API Error')
+    })
+})

src/api/providers/deepseek.ts

@@ -0,0 +1,26 @@
+import { OpenAiHandler } from "./openai"
+import { ApiHandlerOptions, ModelInfo } from "../../shared/api"
+import { deepSeekModels, deepSeekDefaultModelId } from "../../shared/api"
+
+export class DeepSeekHandler extends OpenAiHandler {
+    constructor(options: ApiHandlerOptions) {
+        if (!options.deepSeekApiKey) {
+            throw new Error("DeepSeek API key is required. Please provide it in the settings.")
+        }
+        super({
+            ...options,
+            openAiApiKey: options.deepSeekApiKey,
+            openAiModelId: options.deepSeekModelId ?? deepSeekDefaultModelId,
+            openAiBaseUrl: options.deepSeekBaseUrl ?? "https://api.deepseek.com/v1",
+            includeMaxTokens: true
+        })
+    }
+
+    override getModel(): { id: string; info: ModelInfo } {
+        const modelId = this.options.deepSeekModelId ?? deepSeekDefaultModelId
+        return {
+            id: modelId,
+            info: deepSeekModels[modelId as keyof typeof deepSeekModels] || deepSeekModels[deepSeekDefaultModelId]
+        }
+    }
+}

src/api/providers/openai.ts

@@ -11,7 +11,7 @@ import { convertToOpenAiMessages } from "../transform/openai-format"
 import { ApiStream } from "../transform/stream"
 
 export class OpenAiHandler implements ApiHandler {
-	private options: ApiHandlerOptions
+	protected options: ApiHandlerOptions
 	private client: OpenAI
 
 	constructor(options: ApiHandlerOptions) {
@@ -38,12 +38,16 @@ export class OpenAiHandler implements ApiHandler {
 			{ role: "system", content: systemPrompt },
 			...convertToOpenAiMessages(messages),
 		]
+		const modelInfo = this.getModel().info
 		const requestOptions: OpenAI.Chat.ChatCompletionCreateParams = {
 			model: this.options.openAiModelId ?? "",
 			messages: openAiMessages,
 			temperature: 0,
 			stream: true,
 		}
+		if (this.options.includeMaxTokens) {
+			requestOptions.max_tokens = modelInfo.maxTokens
+		}
 
 		if (this.options.includeStreamOptions ?? true) {
 			requestOptions.stream_options = { include_usage: true }

src/core/webview/ClineProvider.ts

@@ -40,6 +40,7 @@ type SecretKey =
 	| "openAiApiKey"
 	| "geminiApiKey"
 	| "openAiNativeApiKey"
+	| "deepSeekApiKey"
 type GlobalStateKey =
 	| "apiProvider"
 	| "apiModelId"
@@ -443,6 +444,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 							await this.updateGlobalState("anthropicBaseUrl", anthropicBaseUrl)
 							await this.storeSecret("geminiApiKey", geminiApiKey)
 							await this.storeSecret("openAiNativeApiKey", openAiNativeApiKey)
+							await this.storeSecret("deepSeekApiKey", message.apiConfiguration.deepSeekApiKey)
 							await this.updateGlobalState("azureApiVersion", azureApiVersion)
 							await this.updateGlobalState("openRouterModelId", openRouterModelId)
 							await this.updateGlobalState("openRouterModelInfo", openRouterModelInfo)
@@ -1121,6 +1123,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 			anthropicBaseUrl,
 			geminiApiKey,
 			openAiNativeApiKey,
+			deepSeekApiKey,
 			azureApiVersion,
 			openRouterModelId,
 			openRouterModelInfo,
@@ -1163,6 +1166,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 			this.getGlobalState("anthropicBaseUrl") as Promise<string | undefined>,
 			this.getSecret("geminiApiKey") as Promise<string | undefined>,
 			this.getSecret("openAiNativeApiKey") as Promise<string | undefined>,
+			this.getSecret("deepSeekApiKey") as Promise<string | undefined>,
 			this.getGlobalState("azureApiVersion") as Promise<string | undefined>,
 			this.getGlobalState("openRouterModelId") as Promise<string | undefined>,
 			this.getGlobalState("openRouterModelInfo") as Promise<ModelInfo | undefined>,
@@ -1222,6 +1226,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 				anthropicBaseUrl,
 				geminiApiKey,
 				openAiNativeApiKey,
+				deepSeekApiKey,
 				azureApiVersion,
 				openRouterModelId,
 				openRouterModelInfo,
@@ -1344,6 +1349,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 			"openAiApiKey",
 			"geminiApiKey",
 			"openAiNativeApiKey",
+			"deepSeekApiKey",
 		]
 		for (const key of secretKeys) {
 			await this.storeSecret(key, undefined)

src/shared/api.ts

@@ -8,6 +8,7 @@ export type ApiProvider =
 	| "lmstudio"
 	| "gemini"
 	| "openai-native"
+	| "deepseek"
 
 export interface ApiHandlerOptions {
 	apiModelId?: string
@@ -38,6 +39,10 @@ export interface ApiHandlerOptions {
 	openRouterUseMiddleOutTransform?: boolean
 	includeStreamOptions?: boolean
 	setAzureApiVersion?: boolean
+	deepSeekBaseUrl?: string
+	deepSeekApiKey?: string
+	deepSeekModelId?: string
+	includeMaxTokens?: boolean
 }
 
 export type ApiConfiguration = ApiHandlerOptions & {
@@ -489,6 +494,22 @@ export const openAiNativeModels = {
 	},
 } as const satisfies Record<string, ModelInfo>
 
+// DeepSeek
+// https://platform.deepseek.com/docs/api
+export type DeepSeekModelId = keyof typeof deepSeekModels
+export const deepSeekDefaultModelId: DeepSeekModelId = "deepseek-chat"
+export const deepSeekModels = {
+	"deepseek-chat": {
+		maxTokens: 8192,
+		contextWindow: 64_000,
+		supportsImages: false,
+		supportsPromptCache: false,
+		inputPrice: 0.014,  // $0.014 per million tokens
+		outputPrice: 0.28,  // $0.28 per million tokens
+		description: `DeepSeek-V3 achieves a significant breakthrough in inference speed over previous models. It tops the leaderboard among open-source models and rivals the most advanced closed-source models globally.`,
+	},
+} as const satisfies Record<string, ModelInfo>
+
 // Azure OpenAI
 // https://learn.microsoft.com/en-us/azure/ai-services/openai/api-version-deprecation
 // https://learn.microsoft.com/en-us/azure/ai-services/openai/reference#api-specs

webview-ui/src/components/settings/ApiOptions.tsx

@@ -17,6 +17,8 @@ import {
 	azureOpenAiDefaultApiVersion,
 	bedrockDefaultModelId,
 	bedrockModels,
+	deepSeekDefaultModelId,
+	deepSeekModels,
 	geminiDefaultModelId,
 	geminiModels,
 	openAiModelInfoSaneDefaults,
@@ -130,10 +132,11 @@ const ApiOptions = ({ showModelOptions, apiErrorMessage, modelIdErrorMessage }:
 					<VSCodeOption value="openrouter">OpenRouter</VSCodeOption>
 					<VSCodeOption value="anthropic">Anthropic</VSCodeOption>
 					<VSCodeOption value="gemini">Google Gemini</VSCodeOption>
-					<VSCodeOption value="vertex">GCP Vertex AI</VSCodeOption>
+					<VSCodeOption value="deepseek">DeepSeek</VSCodeOption>
-					<VSCodeOption value="bedrock">AWS Bedrock</VSCodeOption>
 					<VSCodeOption value="openai-native">OpenAI</VSCodeOption>
 					<VSCodeOption value="openai">OpenAI Compatible</VSCodeOption>
+					<VSCodeOption value="vertex">GCP Vertex AI</VSCodeOption>
+					<VSCodeOption value="bedrock">AWS Bedrock</VSCodeOption>
 					<VSCodeOption value="lmstudio">LM Studio</VSCodeOption>
 					<VSCodeOption value="ollama">Ollama</VSCodeOption>
 				</VSCodeDropdown>
@@ -560,6 +563,34 @@ const ApiOptions = ({ showModelOptions, apiErrorMessage, modelIdErrorMessage }:
 				</div>
 			)}
 
+			{selectedProvider === "deepseek" && (
+				<div>
+					<VSCodeTextField
+						value={apiConfiguration?.deepSeekApiKey || ""}
+						style={{ width: "100%" }}
+						type="password"
+						onInput={handleInputChange("deepSeekApiKey")}
+						placeholder="Enter API Key...">
+						<span style={{ fontWeight: 500 }}>DeepSeek API Key</span>
+					</VSCodeTextField>
+					<p
+						style={{
+							fontSize: "12px",
+							marginTop: "5px",
+							color: "var(--vscode-descriptionForeground)",
+						}}>
+						This key is stored locally and only used to make API requests from this extension.
+						{!apiConfiguration?.deepSeekApiKey && (
+							<VSCodeLink
+								href="https://platform.deepseek.com/"
+								style={{ display: "inline", fontSize: "inherit" }}>
+								You can get a DeepSeek API key by signing up here.
+							</VSCodeLink>
+						)}
+					</p>
+				</div>
+			)}
+
 			{selectedProvider === "ollama" && (
 				<div>
 					<VSCodeTextField
@@ -652,6 +683,7 @@ const ApiOptions = ({ showModelOptions, apiErrorMessage, modelIdErrorMessage }:
 							{selectedProvider === "vertex" && createDropdown(vertexModels)}
 							{selectedProvider === "gemini" && createDropdown(geminiModels)}
 							{selectedProvider === "openai-native" && createDropdown(openAiNativeModels)}
+							{selectedProvider === "deepseek" && createDropdown(deepSeekModels)}
 						</div>
 
 						<ModelInfoView
@@ -836,6 +868,8 @@ export function normalizeApiConfiguration(apiConfiguration?: ApiConfiguration) {
 			return getProviderData(vertexModels, vertexDefaultModelId)
 		case "gemini":
 			return getProviderData(geminiModels, geminiDefaultModelId)
+		case "deepseek":
+			return getProviderData(deepSeekModels, deepSeekDefaultModelId)
 		case "openai-native":
 			return getProviderData(openAiNativeModels, openAiNativeDefaultModelId)
 		case "openrouter":