Add deepseek for agent and static model choice defaults

src/api/providers/deepseek.ts

@@ -1,24 +1,158 @@
-import { OpenAiHandler, OpenAiHandlerOptions } from "./openai"
-import { ModelInfo } from "../../shared/api"
-import { deepSeekModels, deepSeekDefaultModelId } from "../../shared/api"
-
-export class DeepSeekHandler extends OpenAiHandler {
-	constructor(options: OpenAiHandlerOptions) {
-		super({
-			...options,
-			openAiApiKey: options.deepSeekApiKey ?? "not-provided",
-			openAiModelId: options.apiModelId ?? deepSeekDefaultModelId,
-			openAiBaseUrl: options.deepSeekBaseUrl ?? "https://api.deepseek.com/v1",
-			openAiStreamingEnabled: true,
-			includeMaxTokens: true,
+import { Anthropic } from "@anthropic-ai/sdk"
+import { ApiHandlerOptions, ModelInfo, deepSeekModels, deepSeekDefaultModelId } from "../../shared/api"
+import { ApiHandler, SingleCompletionHandler } from "../index"
+import { convertToR1Format } from "../transform/r1-format"
+import { convertToOpenAiMessages } from "../transform/openai-format"
+import { ApiStream } from "../transform/stream"
+
+interface DeepSeekUsage {
+	prompt_tokens: number
+	completion_tokens: number
+	prompt_cache_miss_tokens?: number
+	prompt_cache_hit_tokens?: number
+}
+
+export class DeepSeekHandler implements ApiHandler, SingleCompletionHandler {
+	private options: ApiHandlerOptions
+
+	constructor(options: ApiHandlerOptions) {
+		if (!options.deepSeekApiKey) {
+			throw new Error("DeepSeek API key is required. Please provide it in the settings.")
+		}
+		this.options = options
+	}
+
+	private get baseUrl(): string {
+		return this.options.deepSeekBaseUrl ?? "https://api.deepseek.com/v1"
+	}
+
+	async *createMessage(systemPrompt: string, messages: Anthropic.Messages.MessageParam[]): ApiStream {
+		const modelInfo = this.getModel().info
+		const modelId = this.options.apiModelId ?? deepSeekDefaultModelId
+		const isReasoner = modelId.includes("deepseek-reasoner")
+
+		const systemMessage = { role: "system", content: systemPrompt }
+		const formattedMessages = isReasoner
+			? convertToR1Format([{ role: "user", content: systemPrompt }, ...messages])
+			: [systemMessage, ...convertToOpenAiMessages(messages)]
+
+		const response = await fetch(`${this.baseUrl}/chat/completions`, {
+			method: "POST",
+			headers: {
+				"Content-Type": "application/json",
+				Authorization: `Bearer ${this.options.deepSeekApiKey}`,
+			},
+			body: JSON.stringify({
+				model: modelId,
+				messages: formattedMessages,
+				temperature: 0,
+				stream: true,
+				max_tokens: modelInfo.maxTokens,
+			}),
 		})
+
+		if (!response.ok) {
+			throw new Error(`DeepSeek API error: ${response.statusText}`)
+		}
+
+		if (!response.body) {
+			throw new Error("No response body received from DeepSeek API")
+		}
+
+		const reader = response.body.getReader()
+		const decoder = new TextDecoder()
+		let buffer = ""
+
+		try {
+			while (true) {
+				const { done, value } = await reader.read()
+				if (done) break
+
+				buffer += decoder.decode(value, { stream: true })
+				const lines = buffer.split("\n")
+				buffer = lines.pop() || ""
+
+				for (const line of lines) {
+					if (line.trim() === "") continue
+					if (!line.startsWith("data: ")) continue
+
+					const data = line.slice(6)
+					if (data === "[DONE]") continue
+
+					try {
+						const chunk = JSON.parse(data)
+						const delta = chunk.choices[0]?.delta ?? {}
+
+						if (delta.content) {
+							yield {
+								type: "text",
+								text: delta.content,
+							}
+						}
+
+						if ("reasoning_content" in delta && delta.reasoning_content) {
+							yield {
+								type: "reasoning",
+								text: delta.reasoning_content,
+							}
+						}
+
+						if (chunk.usage) {
+							const usage = chunk.usage as DeepSeekUsage
+							let inputTokens = (usage.prompt_tokens || 0) - (usage.prompt_cache_hit_tokens || 0)
+							yield {
+								type: "usage",
+								inputTokens: inputTokens,
+								outputTokens: usage.completion_tokens || 0,
+								cacheReadTokens: usage.prompt_cache_hit_tokens || 0,
+								cacheWriteTokens: usage.prompt_cache_miss_tokens || 0,
+							}
+						}
+					} catch (error) {
+						console.error("Error parsing DeepSeek response:", error)
+					}
+				}
+			}
+		} finally {
+			reader.releaseLock()
+		}
 	}
 
-	override getModel(): { id: string; info: ModelInfo } {
+	getModel(): { id: string; info: ModelInfo } {
 		const modelId = this.options.apiModelId ?? deepSeekDefaultModelId
 		return {
 			id: modelId,
 			info: deepSeekModels[modelId as keyof typeof deepSeekModels] || deepSeekModels[deepSeekDefaultModelId],
 		}
 	}
+
+	async completePrompt(prompt: string): Promise<string> {
+		try {
+			const response = await fetch(`${this.baseUrl}/chat/completions`, {
+				method: "POST",
+				headers: {
+					"Content-Type": "application/json",
+					Authorization: `Bearer ${this.options.deepSeekApiKey}`,
+				},
+				body: JSON.stringify({
+					model: this.getModel().id,
+					messages: [{ role: "user", content: prompt }],
+					temperature: 0,
+					stream: false,
+				}),
+			})
+
+			if (!response.ok) {
+				throw new Error(`DeepSeek API error: ${response.statusText}`)
+			}
+
+			const data = await response.json()
+			return data.choices[0]?.message?.content || ""
+		} catch (error) {
+			if (error instanceof Error) {
+				throw new Error(`DeepSeek completion error: ${error.message}`)
+			}
+			throw error
+		}
+	}
 }

src/api/providers/pearai.ts

@@ -1,36 +1,109 @@
-import { OpenAiHandler } from "./openai"
 import * as vscode from "vscode"
-import { AnthropicModelId, ApiHandlerOptions, ModelInfo, PEARAI_URL } from "../../shared/api"
+import { ApiHandlerOptions, PEARAI_URL, ModelInfo } from "../../shared/api"
 import { AnthropicHandler } from "./anthropic"
+import { DeepSeekHandler } from "./deepseek"
+
+interface PearAiModelsResponse {
+	models: {
+		[key: string]: {
+			underlyingModel?: string
+			[key: string]: any
+		}
+	}
+	defaultModelId: string
+}
+
+export class PearAiHandler {
+	private handler!: AnthropicHandler | DeepSeekHandler
 
-export class PearAiHandler extends AnthropicHandler {
 	constructor(options: ApiHandlerOptions) {
 		if (!options.pearaiApiKey) {
 			vscode.window.showErrorMessage("PearAI API key not found.", "Login to PearAI").then(async (selection) => {
 				if (selection === "Login to PearAI") {
 					const extensionUrl = `${vscode.env.uriScheme}://pearai.pearai/auth`
 					const callbackUri = await vscode.env.asExternalUri(vscode.Uri.parse(extensionUrl))
-
 					vscode.env.openExternal(
 						await vscode.env.asExternalUri(
-							vscode.Uri.parse(
-								`https://trypear.ai/signin?callback=${callbackUri.toString()}`, // Change to localhost if running locally
-							),
+							vscode.Uri.parse(`https://trypear.ai/signin?callback=${callbackUri.toString()}`),
 						),
 					)
 				}
 			})
 			throw new Error("PearAI API key not found. Please login to PearAI.")
 		}
-		super({
+
+		// Initialize with a default handler synchronously
+		this.handler = new AnthropicHandler({
 			...options,
 			apiKey: options.pearaiApiKey,
 			anthropicBaseUrl: PEARAI_URL,
+			apiModelId: "claude-3-5-sonnet-20241022",
+		})
+
+		// Then try to initialize the correct handler asynchronously
+		this.initializeHandler(options).catch((error) => {
+			console.error("Failed to initialize PearAI handler:", error)
 		})
 	}
 
-	override getModel(): { id: AnthropicModelId; info: ModelInfo } {
-		const baseModel = super.getModel()
+	private async initializeHandler(options: ApiHandlerOptions): Promise<void> {
+		const modelId = options.apiModelId || "pearai-model"
+
+		if (modelId === "pearai-model") {
+			try {
+				const response = await fetch(`${PEARAI_URL}/getPearAIAgentModels`)
+				if (!response.ok) {
+					throw new Error(`Failed to fetch models: ${response.statusText}`)
+				}
+				const data = (await response.json()) as PearAiModelsResponse
+				const underlyingModel = data.models[modelId]?.underlyingModel || "claude-3-5-sonnet-20241022"
+				console.dir(underlyingModel)
+				if (underlyingModel.startsWith("deepseek")) {
+					this.handler = new DeepSeekHandler({
+						...options,
+						deepSeekApiKey: options.pearaiApiKey,
+						deepSeekBaseUrl: PEARAI_URL,
+						apiModelId: underlyingModel,
+					})
+				} else {
+					// Default to Claude
+					this.handler = new AnthropicHandler({
+						...options,
+						apiKey: options.pearaiApiKey,
+						anthropicBaseUrl: PEARAI_URL,
+						apiModelId: underlyingModel,
+					})
+				}
+			} catch (error) {
+				console.error("Error fetching PearAI models:", error)
+				// Default to Claude if there's an error
+				this.handler = new AnthropicHandler({
+					...options,
+					apiKey: options.pearaiApiKey,
+					anthropicBaseUrl: PEARAI_URL,
+					apiModelId: "claude-3-5-sonnet-20241022",
+				})
+			}
+		} else if (modelId.startsWith("claude")) {
+			this.handler = new AnthropicHandler({
+				...options,
+				apiKey: options.pearaiApiKey,
+				anthropicBaseUrl: PEARAI_URL,
+			})
+		} else if (modelId.startsWith("deepseek")) {
+			this.handler = new DeepSeekHandler({
+				...options,
+				deepSeekApiKey: options.pearaiApiKey,
+				deepSeekBaseUrl: PEARAI_URL,
+			})
+		} else {
+			throw new Error(`Unsupported model: ${modelId}`)
+		}
+	}
+
+	getModel(): { id: string; info: ModelInfo } {
+		console.dir(this.handler)
+		const baseModel = this.handler.getModel()
 		return {
 			id: baseModel.id,
 			info: {
@@ -42,4 +115,13 @@ export class PearAiHandler extends AnthropicHandler {
 			},
 		}
 	}
+
+	async *createMessage(systemPrompt: string, messages: any[]): AsyncGenerator<any> {
+		const generator = this.handler.createMessage(systemPrompt, messages)
+		yield* generator
+	}
+
+	async completePrompt(prompt: string): Promise<string> {
+		return this.handler.completePrompt(prompt)
+	}
 }

src/shared/api.ts

@@ -698,18 +698,22 @@ export const deepSeekModels = {
 		maxTokens: 8192,
 		contextWindow: 64_000,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 0.014, // $0.014 per million tokens
 		outputPrice: 0.28, // $0.28 per million tokens
+		cacheWritesPrice: 0.27, // $0.27 per million tokens (cache miss)
+		cacheReadsPrice: 0.07, // $0.07 per million tokens (cache hit)
 		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.`,
 	},
 	"deepseek-reasoner": {
 		maxTokens: 8192,
 		contextWindow: 64_000,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 0.55, // $0.55 per million tokens
 		outputPrice: 2.19, // $2.19 per million tokens
+		cacheWritesPrice: 0.55, // $0.55 per million tokens (cache miss)
+		cacheReadsPrice: 0.14, // $0.14 per million tokens (cache hit)
 		description: `DeepSeek-R1 achieves performance comparable to OpenAI-o1 across math, code, and reasoning tasks.`,
 	},
 } as const satisfies Record<string, ModelInfo>
@@ -791,3 +795,65 @@ export const unboundDefaultModelInfo: ModelInfo = {
 export const PEARAI_URL = "https://stingray-app-gb2an.ondigitalocean.app/pearai-server-api2/integrations/cline"
 // DEV:
 // export const PEARAI_URL = "http://localhost:8000/integrations/cline"
+
+// PearAI
+export type PearAiModelId = keyof typeof pearAiModels
+export const pearAiDefaultModelId: PearAiModelId = "pearai-model"
+export const pearAiModels = {
+	"pearai-model": {
+		maxTokens: 8192,
+		contextWindow: 64000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		inputPrice: 0.014,
+		outputPrice: 0.28,
+		cacheWritesPrice: 0.27,
+		cacheReadsPrice: 0.07,
+		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.",
+	},
+	"claude-3-5-sonnet-20241022": {
+		maxTokens: 8192,
+		contextWindow: 200000,
+		supportsImages: true,
+		supportsComputerUse: true,
+		supportsPromptCache: true,
+		inputPrice: 3.0,
+		outputPrice: 15.0,
+		cacheWritesPrice: 3.75,
+		cacheReadsPrice: 0.3,
+	},
+	"claude-3-5-haiku-20241022": {
+		maxTokens: 8192,
+		contextWindow: 200000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		inputPrice: 1.0,
+		outputPrice: 5.0,
+		cacheWritesPrice: 1.25,
+		cacheReadsPrice: 0.1,
+	},
+	"deepseek-chat": {
+		maxTokens: 8192,
+		contextWindow: 64000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		inputPrice: 0.014,
+		outputPrice: 0.28,
+		cacheWritesPrice: 0.27,
+		cacheReadsPrice: 0.07,
+		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.",
+	},
+	"deepseek-reasoner": {
+		maxTokens: 8192,
+		contextWindow: 64000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		inputPrice: 0.55,
+		outputPrice: 2.19,
+		cacheWritesPrice: 0.55,
+		cacheReadsPrice: 0.14,
+		description: "DeepSeek-R1 achieves performance comparable to OpenAI-o1 across math, code, and reasoning tasks.",
+	},
+} as const satisfies Record<string, ModelInfo>