mirror of
https://github.com/CherryHQ/cherry-studio.git
synced 2025-12-26 03:31:24 +08:00
feat: support both function call and system prompt for MCP tools (#5499)
* feat: support both function call and system prompt for MCP tools - Add support for using both function call and system prompt to implement MCP tool calls - Refactor tool handling logic to be more flexible and maintainable - Improve code readability with better variable naming and comments - Fix potential issues with tool call implementation * fix: Add tool_calls in OpenAI streaming logic * refactor: enhance OpenAICompatibleProvider and BaseOpenAiProvider structure * feat: add tool call setting to SettingsTab component * fix: enhance tool call handling in OpenAICompatibleProvider * fix: enhance content handling in GeminiProvider for nonstreaming response * refactor: improve tool property filtering logic in OpenAIProvider and mcp-tools utility * fix: resolve eslint errors * fix: add history for function call message in GeminiProvider * refactor: unify MCP tool response handling across providers for consistency * refactor: update mcp tools conversion logic in OpenAICompatibleProvider and OpenAIProvider * refactor: enhance AihubmixProvider and BaseProvider with MCP tool handling methods * refactor: introduce SYSTEM_PROMPT_THRESHOLD constant in BaseProvider for improved readability * refactor: rename tool_call to enable_tool_use for clarity and consistency across the application * refactor: remove unnecessary onChunk call in processStream for cleaner code * fix: add toolCallId to response structure and enhance content handling in AnthropicProvider * fix: respond image data to llm while using function call * fix: add reasoning handling in OpenAICompatibleProvider for improved response processing --------- Co-authored-by: kanweiwei <kanweiwei@nutstore.net> Co-authored-by: jay <sevenjay@users.noreply.github.com>
This commit is contained in:
Camol
GitHub
parent
c0cb1693da
commit
ce8b85020b
@ -705,6 +705,7 @@
|
||||
"rerank_model_tooltip": "Click the Manage button in Settings -> Model Services to add.",
|
||||
"search": "Search models...",
|
||||
"stream_output": "Stream output",
|
||||
"enable_tool_use": "Enable Tool Use",
|
||||
"type": {
|
||||
"embedding": "Embedding",
|
||||
"free": "Free",
|
||||
|
||||
@ -705,6 +705,7 @@
|
||||
"rerank_model_tooltip": "設定->モデルサービスに移動し、管理ボタンをクリックして追加します。",
|
||||
"search": "モデルを検索...",
|
||||
"stream_output": "ストリーム出力",
|
||||
"enable_tool_use": "ツール呼び出し",
|
||||
"type": {
|
||||
"embedding": "埋め込み",
|
||||
"free": "無料",
|
||||
|
||||
@ -705,6 +705,7 @@
|
||||
"rerank_model_tooltip": "В настройках -> Служба модели нажмите кнопку \"Управление\", чтобы добавить.",
|
||||
"search": "Поиск моделей...",
|
||||
"stream_output": "Потоковый вывод",
|
||||
"enable_tool_use": "Вызов инструмента",
|
||||
"type": {
|
||||
"embedding": "Встраиваемые",
|
||||
"free": "Бесплатные",
|
||||
|
||||
@ -705,6 +705,7 @@
|
||||
"rerank_model_tooltip": "在设置->模型服务中点击管理按钮添加",
|
||||
"search": "搜索模型...",
|
||||
"stream_output": "流式输出",
|
||||
"enable_tool_use": "工具调用",
|
||||
"type": {
|
||||
"embedding": "嵌入",
|
||||
"free": "免费",
|
||||
|
||||
@ -705,6 +705,7 @@
|
||||
"rerank_model_tooltip": "在設定->模型服務中點擊管理按鈕添加",
|
||||
"search": "搜尋模型...",
|
||||
"stream_output": "串流輸出",
|
||||
"enable_tool_use": "工具調用",
|
||||
"type": {
|
||||
"embedding": "嵌入",
|
||||
"free": "免費",
|
||||
|
||||
@ -67,7 +67,7 @@ const MessageTools: FC<Props> = ({ blocks }) => {
|
||||
const isDone = status === 'done'
|
||||
const hasError = isDone && response?.isError === true
|
||||
const result = {
|
||||
params: tool.inputSchema,
|
||||
params: toolResponse.arguments,
|
||||
response: toolResponse.response
|
||||
}
|
||||
|
||||
|
||||
@ -70,6 +70,7 @@ const SettingsTab: FC<Props> = (props) => {
|
||||
const [maxTokens, setMaxTokens] = useState(assistant?.settings?.maxTokens ?? 0)
|
||||
const [fontSizeValue, setFontSizeValue] = useState(fontSize)
|
||||
const [streamOutput, setStreamOutput] = useState(assistant?.settings?.streamOutput ?? true)
|
||||
const [enableToolUse, setEnableToolUse] = useState(assistant?.settings?.enableToolUse ?? false)
|
||||
const { t } = useTranslation()
|
||||
|
||||
const dispatch = useAppDispatch()
|
||||
@ -222,6 +223,18 @@ const SettingsTab: FC<Props> = (props) => {
|
||||
/>
|
||||
</SettingRow>
|
||||
<SettingDivider />
|
||||
<SettingRow>
|
||||
<SettingRowTitleSmall>{t('models.enable_tool_use')}</SettingRowTitleSmall>
|
||||
<Switch
|
||||
size="small"
|
||||
checked={enableToolUse}
|
||||
onChange={(checked) => {
|
||||
setEnableToolUse(checked)
|
||||
updateAssistantSettings({ enableToolUse: checked })
|
||||
}}
|
||||
/>
|
||||
</SettingRow>
|
||||
<SettingDivider />
|
||||
<Row align="middle" justify="space-between" style={{ marginBottom: 10 }}>
|
||||
<HStack alignItems="center">
|
||||
<Label>{t('chat.settings.max_tokens')}</Label>
|
||||
|
||||
@ -24,6 +24,7 @@ const AssistantModelSettings: FC<Props> = ({ assistant, updateAssistant, updateA
|
||||
const [enableMaxTokens, setEnableMaxTokens] = useState(assistant?.settings?.enableMaxTokens ?? false)
|
||||
const [maxTokens, setMaxTokens] = useState(assistant?.settings?.maxTokens ?? 0)
|
||||
const [streamOutput, setStreamOutput] = useState(assistant?.settings?.streamOutput ?? true)
|
||||
const [enableToolUse, setEnableToolUse] = useState(assistant?.settings?.enableToolUse ?? false)
|
||||
const [defaultModel, setDefaultModel] = useState(assistant?.defaultModel)
|
||||
const [topP, setTopP] = useState(assistant?.settings?.topP ?? 1)
|
||||
const [customParameters, setCustomParameters] = useState<AssistantSettingCustomParameters[]>(
|
||||
@ -377,6 +378,18 @@ const AssistantModelSettings: FC<Props> = ({ assistant, updateAssistant, updateA
|
||||
/>
|
||||
</SettingRow>
|
||||
<Divider style={{ margin: '10px 0' }} />
|
||||
<SettingRow style={{ minHeight: 30 }}>
|
||||
<Label>{t('models.enable_tool_use')}</Label>
|
||||
<Switch
|
||||
size="small"
|
||||
checked={enableToolUse}
|
||||
onChange={(checked) => {
|
||||
setEnableToolUse(checked)
|
||||
updateAssistantSettings({ enableToolUse: checked })
|
||||
}}
|
||||
/>
|
||||
</SettingRow>
|
||||
<Divider style={{ margin: '10px 0' }} />
|
||||
<SettingRow style={{ minHeight: 30 }}>
|
||||
<Label>{t('models.custom_parameters')}</Label>
|
||||
<Button icon={<PlusOutlined />} onClick={onAddCustomParameter}>
|
||||
|
||||
@ -1,6 +1,6 @@
|
||||
import { isOpenAILLMModel } from '@renderer/config/models'
|
||||
import { getDefaultModel } from '@renderer/services/AssistantService'
|
||||
import { Assistant, Model, Provider, Suggestion } from '@renderer/types'
|
||||
import { Assistant, MCPCallToolResponse, MCPTool, MCPToolResponse, Model, Provider, Suggestion } from '@renderer/types'
|
||||
import { Message } from '@renderer/types/newMessage'
|
||||
import OpenAI from 'openai'
|
||||
|
||||
@ -18,6 +18,7 @@ import OpenAIProvider from './OpenAIProvider'
|
||||
export default class AihubmixProvider extends BaseProvider {
|
||||
private providers: Map<string, BaseProvider> = new Map()
|
||||
private defaultProvider: BaseProvider
|
||||
private currentProvider: BaseProvider
|
||||
|
||||
constructor(provider: Provider) {
|
||||
super(provider)
|
||||
@ -30,6 +31,7 @@ export default class AihubmixProvider extends BaseProvider {
|
||||
|
||||
// 设置默认提供商
|
||||
this.defaultProvider = this.providers.get('default')!
|
||||
this.currentProvider = this.defaultProvider
|
||||
}
|
||||
|
||||
/**
|
||||
@ -70,7 +72,8 @@ export default class AihubmixProvider extends BaseProvider {
|
||||
|
||||
public async completions(params: CompletionsParams): Promise<void> {
|
||||
const model = params.assistant.model
|
||||
return this.getProvider(model!).completions(params)
|
||||
this.currentProvider = this.getProvider(model!)
|
||||
return this.currentProvider.completions(params)
|
||||
}
|
||||
|
||||
public async translate(
|
||||
@ -100,4 +103,12 @@ export default class AihubmixProvider extends BaseProvider {
|
||||
public async getEmbeddingDimensions(model: Model): Promise<number> {
|
||||
return this.getProvider(model).getEmbeddingDimensions(model)
|
||||
}
|
||||
|
||||
public convertMcpTools<T>(mcpTools: MCPTool[]) {
|
||||
return this.currentProvider.convertMcpTools(mcpTools) as T[]
|
||||
}
|
||||
|
||||
public mcpToolCallResponseToMessage(mcpToolResponse: MCPToolResponse, resp: MCPCallToolResponse, model: Model) {
|
||||
return this.currentProvider.mcpToolCallResponseToMessage(mcpToolResponse, resp, model)
|
||||
}
|
||||
}
|
||||
|
||||
@ -1,15 +1,19 @@
|
||||
import Anthropic from '@anthropic-ai/sdk'
|
||||
import {
|
||||
Base64ImageSource,
|
||||
ImageBlockParam,
|
||||
MessageCreateParamsNonStreaming,
|
||||
MessageParam,
|
||||
TextBlockParam,
|
||||
ToolResultBlockParam,
|
||||
ToolUnion,
|
||||
ToolUseBlock,
|
||||
WebSearchResultBlock,
|
||||
WebSearchTool20250305,
|
||||
WebSearchToolResultError
|
||||
} from '@anthropic-ai/sdk/resources'
|
||||
import { DEFAULT_MAX_TOKENS } from '@renderer/config/constant'
|
||||
import { isReasoningModel, isVisionModel, isWebSearchModel } from '@renderer/config/models'
|
||||
import { isReasoningModel, isWebSearchModel } from '@renderer/config/models'
|
||||
import { getStoreSetting } from '@renderer/hooks/useSettings'
|
||||
import i18n from '@renderer/i18n'
|
||||
import { getAssistantSettings, getDefaultModel, getTopNamingModel } from '@renderer/services/AssistantService'
|
||||
@ -23,16 +27,24 @@ import {
|
||||
Assistant,
|
||||
EFFORT_RATIO,
|
||||
FileTypes,
|
||||
MCPCallToolResponse,
|
||||
MCPTool,
|
||||
MCPToolResponse,
|
||||
Model,
|
||||
Provider,
|
||||
Suggestion,
|
||||
ToolCallResponse,
|
||||
WebSearchSource
|
||||
} from '@renderer/types'
|
||||
import { ChunkType } from '@renderer/types/chunk'
|
||||
import type { Message } from '@renderer/types/newMessage'
|
||||
import { removeSpecialCharactersForTopicName } from '@renderer/utils'
|
||||
import { mcpToolCallResponseToAnthropicMessage, parseAndCallTools } from '@renderer/utils/mcp-tools'
|
||||
import {
|
||||
anthropicToolUseToMcpTool,
|
||||
mcpToolCallResponseToAnthropicMessage,
|
||||
mcpToolsToAnthropicTools,
|
||||
parseAndCallTools
|
||||
} from '@renderer/utils/mcp-tools'
|
||||
import { findFileBlocks, findImageBlocks, getMainTextContent } from '@renderer/utils/messageUtils/find'
|
||||
import { buildSystemPrompt } from '@renderer/utils/prompt'
|
||||
import { first, flatten, sum, takeRight } from 'lodash'
|
||||
@ -199,7 +211,7 @@ export default class AnthropicProvider extends BaseProvider {
|
||||
public async completions({ messages, assistant, mcpTools, onChunk, onFilterMessages }: CompletionsParams) {
|
||||
const defaultModel = getDefaultModel()
|
||||
const model = assistant.model || defaultModel
|
||||
const { contextCount, maxTokens, streamOutput } = getAssistantSettings(assistant)
|
||||
const { contextCount, maxTokens, streamOutput, enableToolUse } = getAssistantSettings(assistant)
|
||||
|
||||
const userMessagesParams: MessageParam[] = []
|
||||
|
||||
@ -215,10 +227,16 @@ export default class AnthropicProvider extends BaseProvider {
|
||||
|
||||
const userMessages = flatten(userMessagesParams)
|
||||
const lastUserMessage = _messages.findLast((m) => m.role === 'user')
|
||||
// const tools = mcpTools ? mcpToolsToAnthropicTools(mcpTools) : undefined
|
||||
|
||||
let systemPrompt = assistant.prompt
|
||||
if (mcpTools && mcpTools.length > 0) {
|
||||
|
||||
const { tools } = this.setupToolsConfig<ToolUnion>({
|
||||
model,
|
||||
mcpTools,
|
||||
enableToolUse
|
||||
})
|
||||
|
||||
if (this.useSystemPromptForTools && mcpTools && mcpTools.length) {
|
||||
systemPrompt = buildSystemPrompt(systemPrompt, mcpTools)
|
||||
}
|
||||
|
||||
@ -232,8 +250,6 @@ export default class AnthropicProvider extends BaseProvider {
|
||||
|
||||
const isEnabledBuiltinWebSearch = assistant.enableWebSearch
|
||||
|
||||
const tools: ToolUnion[] = []
|
||||
|
||||
if (isEnabledBuiltinWebSearch) {
|
||||
const webSearchTool = await this.getWebSearchParams(model)
|
||||
if (webSearchTool) {
|
||||
@ -244,7 +260,6 @@ export default class AnthropicProvider extends BaseProvider {
|
||||
const body: MessageCreateParamsNonStreaming = {
|
||||
model: model.id,
|
||||
messages: userMessages,
|
||||
// tools: isEmpty(tools) ? undefined : tools,
|
||||
max_tokens: maxTokens || DEFAULT_MAX_TOKENS,
|
||||
temperature: this.getTemperature(assistant, model),
|
||||
top_p: this.getTopP(assistant, model),
|
||||
@ -303,7 +318,7 @@ export default class AnthropicProvider extends BaseProvider {
|
||||
const processStream = (body: MessageCreateParamsNonStreaming, idx: number) => {
|
||||
return new Promise<void>((resolve, reject) => {
|
||||
// 等待接口返回流
|
||||
onChunk({ type: ChunkType.LLM_RESPONSE_CREATED })
|
||||
const toolCalls: ToolUseBlock[] = []
|
||||
let hasThinkingContent = false
|
||||
this.sdk.messages
|
||||
.stream({ ...body, stream: true }, { signal, timeout: 5 * 60 * 1000 })
|
||||
@ -380,30 +395,70 @@ export default class AnthropicProvider extends BaseProvider {
|
||||
})
|
||||
thinking_content += thinking
|
||||
})
|
||||
.on('contentBlock', (content) => {
|
||||
if (content.type === 'tool_use') {
|
||||
toolCalls.push(content)
|
||||
}
|
||||
})
|
||||
.on('finalMessage', async (message) => {
|
||||
const toolResults: Awaited<ReturnType<typeof parseAndCallTools>> = []
|
||||
// tool call
|
||||
if (toolCalls.length > 0) {
|
||||
const mcpToolResponses = toolCalls
|
||||
.map((toolCall) => {
|
||||
const mcpTool = anthropicToolUseToMcpTool(mcpTools, toolCall)
|
||||
if (!mcpTool) {
|
||||
return undefined
|
||||
}
|
||||
return {
|
||||
id: toolCall.id,
|
||||
toolCallId: toolCall.id,
|
||||
tool: mcpTool,
|
||||
arguments: toolCall.input as Record<string, unknown>,
|
||||
status: 'pending'
|
||||
} as ToolCallResponse
|
||||
})
|
||||
.filter((t) => typeof t !== 'undefined')
|
||||
toolResults.push(
|
||||
...(await parseAndCallTools(
|
||||
mcpToolResponses,
|
||||
toolResponses,
|
||||
onChunk,
|
||||
this.mcpToolCallResponseToMessage,
|
||||
model,
|
||||
mcpTools
|
||||
))
|
||||
)
|
||||
}
|
||||
|
||||
// tool use
|
||||
const content = message.content[0]
|
||||
if (content && content.type === 'text') {
|
||||
onChunk({ type: ChunkType.TEXT_COMPLETE, text: content.text })
|
||||
const toolResults = await parseAndCallTools(
|
||||
content.text,
|
||||
toolResponses,
|
||||
onChunk,
|
||||
idx,
|
||||
mcpToolCallResponseToAnthropicMessage,
|
||||
mcpTools,
|
||||
isVisionModel(model)
|
||||
toolResults.push(
|
||||
...(await parseAndCallTools(
|
||||
content.text,
|
||||
toolResponses,
|
||||
onChunk,
|
||||
this.mcpToolCallResponseToMessage,
|
||||
model,
|
||||
mcpTools
|
||||
))
|
||||
)
|
||||
if (toolResults.length > 0) {
|
||||
userMessages.push({
|
||||
role: message.role,
|
||||
content: message.content
|
||||
})
|
||||
}
|
||||
|
||||
toolResults.forEach((ts) => userMessages.push(ts as MessageParam))
|
||||
const newBody = body
|
||||
newBody.messages = userMessages
|
||||
await processStream(newBody, idx + 1)
|
||||
}
|
||||
userMessages.push({
|
||||
role: message.role,
|
||||
content: message.content
|
||||
})
|
||||
|
||||
if (toolResults.length > 0) {
|
||||
toolResults.forEach((ts) => userMessages.push(ts as MessageParam))
|
||||
const newBody = body
|
||||
newBody.messages = userMessages
|
||||
|
||||
onChunk({ type: ChunkType.LLM_RESPONSE_CREATED })
|
||||
await processStream(newBody, idx + 1)
|
||||
}
|
||||
|
||||
const time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
@ -434,7 +489,7 @@ export default class AnthropicProvider extends BaseProvider {
|
||||
})
|
||||
})
|
||||
}
|
||||
|
||||
onChunk({ type: ChunkType.LLM_RESPONSE_CREATED })
|
||||
await processStream(body, 0).finally(cleanup)
|
||||
}
|
||||
|
||||
@ -683,4 +738,47 @@ export default class AnthropicProvider extends BaseProvider {
|
||||
public async getEmbeddingDimensions(): Promise<number> {
|
||||
return 0
|
||||
}
|
||||
|
||||
public convertMcpTools<T>(mcpTools: MCPTool[]): T[] {
|
||||
return mcpToolsToAnthropicTools(mcpTools) as T[]
|
||||
}
|
||||
|
||||
public mcpToolCallResponseToMessage = (mcpToolResponse: MCPToolResponse, resp: MCPCallToolResponse, model: Model) => {
|
||||
if ('toolUseId' in mcpToolResponse && mcpToolResponse.toolUseId) {
|
||||
return mcpToolCallResponseToAnthropicMessage(mcpToolResponse, resp, model)
|
||||
} else if ('toolCallId' in mcpToolResponse) {
|
||||
return {
|
||||
role: 'user',
|
||||
content: [
|
||||
{
|
||||
type: 'tool_result',
|
||||
tool_use_id: mcpToolResponse.toolCallId!,
|
||||
content: resp.content
|
||||
.map((item) => {
|
||||
if (item.type === 'text') {
|
||||
return {
|
||||
type: 'text',
|
||||
text: item.text || ''
|
||||
} satisfies TextBlockParam
|
||||
}
|
||||
if (item.type === 'image') {
|
||||
return {
|
||||
type: 'image',
|
||||
source: {
|
||||
data: item.data || '',
|
||||
media_type: (item.mimeType || 'image/png') as Base64ImageSource['media_type'],
|
||||
type: 'base64'
|
||||
}
|
||||
} satisfies ImageBlockParam
|
||||
}
|
||||
return
|
||||
})
|
||||
.filter((n) => typeof n !== 'undefined'),
|
||||
is_error: resp.isError
|
||||
} satisfies ToolResultBlockParam
|
||||
]
|
||||
}
|
||||
}
|
||||
return
|
||||
}
|
||||
}
|
||||
|
||||
@ -1,9 +1,13 @@
|
||||
import { isFunctionCallingModel } from '@renderer/config/models'
|
||||
import { REFERENCE_PROMPT } from '@renderer/config/prompts'
|
||||
import { getLMStudioKeepAliveTime } from '@renderer/hooks/useLMStudio'
|
||||
import type {
|
||||
Assistant,
|
||||
GenerateImageParams,
|
||||
KnowledgeReference,
|
||||
MCPCallToolResponse,
|
||||
MCPTool,
|
||||
MCPToolResponse,
|
||||
Model,
|
||||
Provider,
|
||||
Suggestion,
|
||||
@ -22,10 +26,15 @@ import type OpenAI from 'openai'
|
||||
import type { CompletionsParams } from '.'
|
||||
|
||||
export default abstract class BaseProvider {
|
||||
// Threshold for determining whether to use system prompt for tools
|
||||
private static readonly SYSTEM_PROMPT_THRESHOLD: number = 128
|
||||
|
||||
protected provider: Provider
|
||||
protected host: string
|
||||
protected apiKey: string
|
||||
|
||||
protected useSystemPromptForTools: boolean = true
|
||||
|
||||
constructor(provider: Provider) {
|
||||
this.provider = provider
|
||||
this.host = this.getBaseURL()
|
||||
@ -47,6 +56,12 @@ export default abstract class BaseProvider {
|
||||
abstract generateImage(params: GenerateImageParams): Promise<string[]>
|
||||
abstract generateImageByChat({ messages, assistant, onChunk, onFilterMessages }: CompletionsParams): Promise<void>
|
||||
abstract getEmbeddingDimensions(model: Model): Promise<number>
|
||||
public abstract convertMcpTools<T>(mcpTools: MCPTool[]): T[]
|
||||
public abstract mcpToolCallResponseToMessage(
|
||||
mcpToolResponse: MCPToolResponse,
|
||||
resp: MCPCallToolResponse,
|
||||
model: Model
|
||||
): any
|
||||
|
||||
public getBaseURL(): string {
|
||||
const host = this.provider.apiHost
|
||||
@ -229,4 +244,31 @@ export default abstract class BaseProvider {
|
||||
cleanup
|
||||
}
|
||||
}
|
||||
|
||||
// Setup tools configuration based on provided parameters
|
||||
protected setupToolsConfig<T>(params: { mcpTools?: MCPTool[]; model: Model; enableToolUse?: boolean }): {
|
||||
tools: T[]
|
||||
} {
|
||||
const { mcpTools, model, enableToolUse } = params
|
||||
let tools: T[] = []
|
||||
|
||||
// If there are no tools, return an empty array
|
||||
if (!mcpTools?.length) {
|
||||
return { tools }
|
||||
}
|
||||
|
||||
// If the number of tools exceeds the threshold, use the system prompt
|
||||
if (mcpTools.length > BaseProvider.SYSTEM_PROMPT_THRESHOLD) {
|
||||
this.useSystemPromptForTools = true
|
||||
return { tools }
|
||||
}
|
||||
|
||||
// If the model supports function calling and tool usage is enabled
|
||||
if (isFunctionCallingModel(model) && enableToolUse) {
|
||||
tools = this.convertMcpTools<T>(mcpTools)
|
||||
this.useSystemPromptForTools = false
|
||||
}
|
||||
|
||||
return { tools }
|
||||
}
|
||||
}
|
||||
|
||||
@ -1,6 +1,7 @@
|
||||
import {
|
||||
Content,
|
||||
File,
|
||||
FunctionCall,
|
||||
GenerateContentConfig,
|
||||
GenerateContentResponse,
|
||||
GoogleGenAI,
|
||||
@ -11,8 +12,9 @@ import {
|
||||
PartUnion,
|
||||
SafetySetting,
|
||||
ThinkingConfig,
|
||||
ToolListUnion
|
||||
Tool
|
||||
} from '@google/genai'
|
||||
import { nanoid } from '@reduxjs/toolkit'
|
||||
import {
|
||||
findTokenLimit,
|
||||
isGeminiReasoningModel,
|
||||
@ -35,17 +37,25 @@ import {
|
||||
EFFORT_RATIO,
|
||||
FileType,
|
||||
FileTypes,
|
||||
MCPCallToolResponse,
|
||||
MCPTool,
|
||||
MCPToolResponse,
|
||||
Model,
|
||||
Provider,
|
||||
Suggestion,
|
||||
ToolCallResponse,
|
||||
Usage,
|
||||
WebSearchSource
|
||||
} from '@renderer/types'
|
||||
import { BlockCompleteChunk, Chunk, ChunkType, LLMWebSearchCompleteChunk } from '@renderer/types/chunk'
|
||||
import type { Message, Response } from '@renderer/types/newMessage'
|
||||
import { removeSpecialCharactersForTopicName } from '@renderer/utils'
|
||||
import { mcpToolCallResponseToGeminiMessage, parseAndCallTools } from '@renderer/utils/mcp-tools'
|
||||
import {
|
||||
geminiFunctionCallToMcpTool,
|
||||
mcpToolCallResponseToGeminiMessage,
|
||||
mcpToolsToGeminiTools,
|
||||
parseAndCallTools
|
||||
} from '@renderer/utils/mcp-tools'
|
||||
import { findFileBlocks, findImageBlocks, getMainTextContent } from '@renderer/utils/messageUtils/find'
|
||||
import { buildSystemPrompt } from '@renderer/utils/prompt'
|
||||
import { MB } from '@shared/config/constant'
|
||||
@ -263,7 +273,7 @@ export default class GeminiProvider extends BaseProvider {
|
||||
}: CompletionsParams): Promise<void> {
|
||||
const defaultModel = getDefaultModel()
|
||||
const model = assistant.model || defaultModel
|
||||
const { contextCount, maxTokens, streamOutput } = getAssistantSettings(assistant)
|
||||
const { contextCount, maxTokens, streamOutput, enableToolUse } = getAssistantSettings(assistant)
|
||||
|
||||
const userMessages = filterUserRoleStartMessages(
|
||||
filterEmptyMessages(filterContextMessages(takeRight(messages, contextCount + 2)))
|
||||
@ -280,12 +290,16 @@ export default class GeminiProvider extends BaseProvider {
|
||||
|
||||
let systemInstruction = assistant.prompt
|
||||
|
||||
if (mcpTools && mcpTools.length > 0) {
|
||||
const { tools } = this.setupToolsConfig<Tool>({
|
||||
mcpTools,
|
||||
model,
|
||||
enableToolUse
|
||||
})
|
||||
|
||||
if (this.useSystemPromptForTools) {
|
||||
systemInstruction = buildSystemPrompt(assistant.prompt || '', mcpTools)
|
||||
}
|
||||
|
||||
// const tools = mcpToolsToGeminiTools(mcpTools)
|
||||
const tools: ToolListUnion = []
|
||||
const toolResponses: MCPToolResponse[] = []
|
||||
|
||||
if (assistant.enableWebSearch && isWebSearchModel(model)) {
|
||||
@ -351,6 +365,224 @@ export default class GeminiProvider extends BaseProvider {
|
||||
|
||||
const { cleanup, abortController } = this.createAbortController(userLastMessage?.id, true)
|
||||
|
||||
const processToolResults = async (toolResults: Awaited<ReturnType<typeof parseAndCallTools>>, idx: number) => {
|
||||
if (toolResults.length === 0) return
|
||||
const newChat = this.sdk.chats.create({
|
||||
model: model.id,
|
||||
config: generateContentConfig,
|
||||
history: history as Content[]
|
||||
})
|
||||
|
||||
const newStream = await newChat.sendMessageStream({
|
||||
message: flatten(toolResults.map((ts) => (ts as Content).parts)) as PartUnion,
|
||||
config: {
|
||||
...generateContentConfig,
|
||||
abortSignal: abortController.signal
|
||||
}
|
||||
})
|
||||
await processStream(newStream, idx + 1)
|
||||
}
|
||||
|
||||
const processToolCalls = async (toolCalls: FunctionCall[]) => {
|
||||
const mcpToolResponses: ToolCallResponse[] = toolCalls
|
||||
.map((toolCall) => {
|
||||
const mcpTool = geminiFunctionCallToMcpTool(mcpTools, toolCall)
|
||||
if (!mcpTool) return undefined
|
||||
|
||||
const parsedArgs = (() => {
|
||||
try {
|
||||
return typeof toolCall.args === 'string' ? JSON.parse(toolCall.args) : toolCall.args
|
||||
} catch {
|
||||
return toolCall.args
|
||||
}
|
||||
})()
|
||||
|
||||
return {
|
||||
id: toolCall.id || nanoid(),
|
||||
toolCallId: toolCall.id,
|
||||
tool: mcpTool,
|
||||
arguments: parsedArgs,
|
||||
status: 'pending'
|
||||
} as ToolCallResponse
|
||||
})
|
||||
.filter((t): t is ToolCallResponse => typeof t !== 'undefined')
|
||||
|
||||
return await parseAndCallTools(
|
||||
mcpToolResponses,
|
||||
toolResponses,
|
||||
onChunk,
|
||||
this.mcpToolCallResponseToMessage,
|
||||
model,
|
||||
mcpTools
|
||||
)
|
||||
}
|
||||
|
||||
const processToolUses = async (content: string) => {
|
||||
return await parseAndCallTools(
|
||||
content,
|
||||
toolResponses,
|
||||
onChunk,
|
||||
this.mcpToolCallResponseToMessage,
|
||||
model,
|
||||
mcpTools
|
||||
)
|
||||
}
|
||||
|
||||
const processStream = async (
|
||||
stream: AsyncGenerator<GenerateContentResponse> | GenerateContentResponse,
|
||||
idx: number
|
||||
) => {
|
||||
history.push(messageContents)
|
||||
|
||||
let functionCalls: FunctionCall[] = []
|
||||
|
||||
if (stream instanceof GenerateContentResponse) {
|
||||
let content = ''
|
||||
const time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
|
||||
const toolResults: Awaited<ReturnType<typeof parseAndCallTools>> = []
|
||||
if (stream.text?.length) {
|
||||
toolResults.push(...(await processToolUses(stream.text)))
|
||||
}
|
||||
stream.candidates?.forEach((candidate) => {
|
||||
if (candidate.content) {
|
||||
history.push(candidate.content)
|
||||
|
||||
candidate.content.parts?.forEach((part) => {
|
||||
if (part.functionCall) {
|
||||
functionCalls.push(part.functionCall)
|
||||
}
|
||||
if (part.text) {
|
||||
content += part.text
|
||||
onChunk({ type: ChunkType.TEXT_DELTA, text: part.text })
|
||||
}
|
||||
})
|
||||
}
|
||||
})
|
||||
if (content.length) {
|
||||
onChunk({ type: ChunkType.TEXT_COMPLETE, text: content })
|
||||
}
|
||||
if (functionCalls.length) {
|
||||
toolResults.push(...(await processToolCalls(functionCalls)))
|
||||
}
|
||||
if (stream.text?.length) {
|
||||
toolResults.push(...(await processToolUses(stream.text)))
|
||||
}
|
||||
if (toolResults.length) {
|
||||
await processToolResults(toolResults, idx)
|
||||
}
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
text: stream.text,
|
||||
usage: {
|
||||
prompt_tokens: stream.usageMetadata?.promptTokenCount || 0,
|
||||
thoughts_tokens: stream.usageMetadata?.thoughtsTokenCount || 0,
|
||||
completion_tokens: stream.usageMetadata?.candidatesTokenCount || 0,
|
||||
total_tokens: stream.usageMetadata?.totalTokenCount || 0
|
||||
},
|
||||
metrics: {
|
||||
completion_tokens: stream.usageMetadata?.candidatesTokenCount,
|
||||
time_completion_millsec,
|
||||
time_first_token_millsec: 0
|
||||
},
|
||||
webSearch: {
|
||||
results: stream.candidates?.[0]?.groundingMetadata,
|
||||
source: 'gemini'
|
||||
}
|
||||
} as Response
|
||||
} as BlockCompleteChunk)
|
||||
} else {
|
||||
let content = ''
|
||||
let final_time_completion_millsec = 0
|
||||
let lastUsage: Usage | undefined = undefined
|
||||
for await (const chunk of stream) {
|
||||
if (window.keyv.get(EVENT_NAMES.CHAT_COMPLETION_PAUSED)) break
|
||||
|
||||
// --- Calculate Metrics ---
|
||||
if (time_first_token_millsec == 0 && chunk.text !== undefined) {
|
||||
// Update based on text arrival
|
||||
time_first_token_millsec = new Date().getTime() - start_time_millsec
|
||||
}
|
||||
|
||||
// 1. Text Content
|
||||
if (chunk.text !== undefined) {
|
||||
content += chunk.text
|
||||
onChunk({ type: ChunkType.TEXT_DELTA, text: chunk.text })
|
||||
}
|
||||
|
||||
// 2. Usage Data
|
||||
if (chunk.usageMetadata) {
|
||||
lastUsage = {
|
||||
prompt_tokens: chunk.usageMetadata.promptTokenCount || 0,
|
||||
completion_tokens: chunk.usageMetadata.candidatesTokenCount || 0,
|
||||
total_tokens: chunk.usageMetadata.totalTokenCount || 0
|
||||
}
|
||||
final_time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
}
|
||||
|
||||
// 4. Image Generation
|
||||
const generateImage = this.processGeminiImageResponse(chunk, onChunk)
|
||||
if (generateImage?.images?.length) {
|
||||
onChunk({ type: ChunkType.IMAGE_COMPLETE, image: generateImage })
|
||||
}
|
||||
|
||||
if (chunk.candidates?.[0]?.finishReason) {
|
||||
if (chunk.text) {
|
||||
onChunk({ type: ChunkType.TEXT_COMPLETE, text: content })
|
||||
}
|
||||
if (chunk.candidates?.[0]?.groundingMetadata) {
|
||||
// 3. Grounding/Search Metadata
|
||||
const groundingMetadata = chunk.candidates?.[0]?.groundingMetadata
|
||||
onChunk({
|
||||
type: ChunkType.LLM_WEB_SEARCH_COMPLETE,
|
||||
llm_web_search: {
|
||||
results: groundingMetadata,
|
||||
source: WebSearchSource.GEMINI
|
||||
}
|
||||
} as LLMWebSearchCompleteChunk)
|
||||
}
|
||||
if (chunk.functionCalls) {
|
||||
chunk.candidates?.forEach((candidate) => {
|
||||
if (candidate.content) {
|
||||
history.push(candidate.content)
|
||||
}
|
||||
})
|
||||
functionCalls = functionCalls.concat(chunk.functionCalls)
|
||||
}
|
||||
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
metrics: {
|
||||
completion_tokens: lastUsage?.completion_tokens,
|
||||
time_completion_millsec: final_time_completion_millsec,
|
||||
time_first_token_millsec
|
||||
},
|
||||
usage: lastUsage
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
// --- End Incremental onChunk calls ---
|
||||
|
||||
// Call processToolUses AFTER potentially processing text content in this chunk
|
||||
// This assumes tools might be specified within the text stream
|
||||
// Note: parseAndCallTools inside should handle its own onChunk for tool responses
|
||||
let toolResults: Awaited<ReturnType<typeof parseAndCallTools>> = []
|
||||
if (functionCalls.length) {
|
||||
toolResults = await processToolCalls(functionCalls)
|
||||
}
|
||||
if (content.length) {
|
||||
toolResults = toolResults.concat(await processToolUses(content))
|
||||
}
|
||||
if (toolResults.length) {
|
||||
await processToolResults(toolResults, idx)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (!streamOutput) {
|
||||
const response = await chat.sendMessage({
|
||||
message: messageContents as PartUnion,
|
||||
@ -359,32 +591,10 @@ export default class GeminiProvider extends BaseProvider {
|
||||
abortSignal: abortController.signal
|
||||
}
|
||||
})
|
||||
const time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
text: response.text,
|
||||
usage: {
|
||||
prompt_tokens: response.usageMetadata?.promptTokenCount || 0,
|
||||
thoughts_tokens: response.usageMetadata?.thoughtsTokenCount || 0,
|
||||
completion_tokens: response.usageMetadata?.candidatesTokenCount || 0,
|
||||
total_tokens: response.usageMetadata?.totalTokenCount || 0
|
||||
},
|
||||
metrics: {
|
||||
completion_tokens: response.usageMetadata?.candidatesTokenCount,
|
||||
time_completion_millsec,
|
||||
time_first_token_millsec: 0
|
||||
},
|
||||
webSearch: {
|
||||
results: response.candidates?.[0]?.groundingMetadata,
|
||||
source: 'gemini'
|
||||
}
|
||||
} as Response
|
||||
} as BlockCompleteChunk)
|
||||
return
|
||||
onChunk({ type: ChunkType.LLM_RESPONSE_CREATED })
|
||||
return await processStream(response, 0).then(cleanup)
|
||||
}
|
||||
|
||||
// 等待接口返回流
|
||||
onChunk({ type: ChunkType.LLM_RESPONSE_CREATED })
|
||||
const userMessagesStream = await chat.sendMessageStream({
|
||||
message: messageContents as PartUnion,
|
||||
@ -394,105 +604,6 @@ export default class GeminiProvider extends BaseProvider {
|
||||
}
|
||||
})
|
||||
|
||||
const processToolUses = async (content: string, idx: number) => {
|
||||
const toolResults = await parseAndCallTools(
|
||||
content,
|
||||
toolResponses,
|
||||
onChunk,
|
||||
idx,
|
||||
mcpToolCallResponseToGeminiMessage,
|
||||
mcpTools,
|
||||
isVisionModel(model)
|
||||
)
|
||||
if (toolResults && toolResults.length > 0) {
|
||||
history.push(messageContents)
|
||||
const newChat = this.sdk.chats.create({
|
||||
model: model.id,
|
||||
config: generateContentConfig,
|
||||
history: history as Content[]
|
||||
})
|
||||
const newStream = await newChat.sendMessageStream({
|
||||
message: flatten(toolResults.map((ts) => (ts as Content).parts)) as PartUnion,
|
||||
config: {
|
||||
...generateContentConfig,
|
||||
abortSignal: abortController.signal
|
||||
}
|
||||
})
|
||||
await processStream(newStream, idx + 1)
|
||||
}
|
||||
}
|
||||
|
||||
const processStream = async (stream: AsyncGenerator<GenerateContentResponse>, idx: number) => {
|
||||
let content = ''
|
||||
let final_time_completion_millsec = 0
|
||||
let lastUsage: Usage | undefined = undefined
|
||||
for await (const chunk of stream) {
|
||||
if (window.keyv.get(EVENT_NAMES.CHAT_COMPLETION_PAUSED)) break
|
||||
|
||||
// --- Calculate Metrics ---
|
||||
if (time_first_token_millsec == 0 && chunk.text !== undefined) {
|
||||
// Update based on text arrival
|
||||
time_first_token_millsec = new Date().getTime() - start_time_millsec
|
||||
}
|
||||
|
||||
// 1. Text Content
|
||||
if (chunk.text !== undefined) {
|
||||
content += chunk.text
|
||||
onChunk({ type: ChunkType.TEXT_DELTA, text: chunk.text })
|
||||
}
|
||||
|
||||
// 2. Usage Data
|
||||
if (chunk.usageMetadata) {
|
||||
lastUsage = {
|
||||
prompt_tokens: chunk.usageMetadata.promptTokenCount || 0,
|
||||
completion_tokens: chunk.usageMetadata.candidatesTokenCount || 0,
|
||||
total_tokens: chunk.usageMetadata.totalTokenCount || 0
|
||||
}
|
||||
final_time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
}
|
||||
|
||||
// 4. Image Generation
|
||||
const generateImage = this.processGeminiImageResponse(chunk, onChunk)
|
||||
if (generateImage?.images?.length) {
|
||||
onChunk({ type: ChunkType.IMAGE_COMPLETE, image: generateImage })
|
||||
}
|
||||
|
||||
if (chunk.candidates?.[0]?.finishReason) {
|
||||
if (chunk.text) {
|
||||
onChunk({ type: ChunkType.TEXT_COMPLETE, text: content })
|
||||
}
|
||||
if (chunk.candidates?.[0]?.groundingMetadata) {
|
||||
// 3. Grounding/Search Metadata
|
||||
const groundingMetadata = chunk.candidates?.[0]?.groundingMetadata
|
||||
onChunk({
|
||||
type: ChunkType.LLM_WEB_SEARCH_COMPLETE,
|
||||
llm_web_search: {
|
||||
results: groundingMetadata,
|
||||
source: WebSearchSource.GEMINI
|
||||
}
|
||||
} as LLMWebSearchCompleteChunk)
|
||||
}
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
metrics: {
|
||||
completion_tokens: lastUsage?.completion_tokens,
|
||||
time_completion_millsec: final_time_completion_millsec,
|
||||
time_first_token_millsec
|
||||
},
|
||||
usage: lastUsage
|
||||
}
|
||||
})
|
||||
}
|
||||
// --- End Incremental onChunk calls ---
|
||||
|
||||
// Call processToolUses AFTER potentially processing text content in this chunk
|
||||
// This assumes tools might be specified within the text stream
|
||||
// Note: parseAndCallTools inside should handle its own onChunk for tool responses
|
||||
await processToolUses(content, idx)
|
||||
}
|
||||
}
|
||||
|
||||
await processStream(userMessagesStream, 0).finally(cleanup)
|
||||
|
||||
const final_time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
@ -841,4 +952,32 @@ export default class GeminiProvider extends BaseProvider {
|
||||
public generateImageByChat(): Promise<void> {
|
||||
throw new Error('Method not implemented.')
|
||||
}
|
||||
|
||||
public convertMcpTools<T>(mcpTools: MCPTool[]): T[] {
|
||||
return mcpToolsToGeminiTools(mcpTools) as T[]
|
||||
}
|
||||
|
||||
public mcpToolCallResponseToMessage = (mcpToolResponse: MCPToolResponse, resp: MCPCallToolResponse, model: Model) => {
|
||||
if ('toolUseId' in mcpToolResponse && mcpToolResponse.toolUseId) {
|
||||
return mcpToolCallResponseToGeminiMessage(mcpToolResponse, resp, isVisionModel(model))
|
||||
} else if ('toolCallId' in mcpToolResponse) {
|
||||
const toolCallOut = {
|
||||
role: 'user',
|
||||
parts: [
|
||||
{
|
||||
functionResponse: {
|
||||
id: mcpToolResponse.toolCallId,
|
||||
name: mcpToolResponse.tool.id,
|
||||
response: {
|
||||
output: !resp.isError ? resp.content : undefined,
|
||||
error: resp.isError ? resp.content : undefined
|
||||
}
|
||||
}
|
||||
}
|
||||
]
|
||||
} satisfies Content
|
||||
return toolCallOut
|
||||
}
|
||||
return
|
||||
}
|
||||
}
|
||||
|
||||
@ -31,10 +31,13 @@ import {
|
||||
Assistant,
|
||||
EFFORT_RATIO,
|
||||
FileTypes,
|
||||
MCPCallToolResponse,
|
||||
MCPTool,
|
||||
MCPToolResponse,
|
||||
Model,
|
||||
Provider,
|
||||
Suggestion,
|
||||
ToolCallResponse,
|
||||
Usage,
|
||||
WebSearchSource
|
||||
} from '@renderer/types'
|
||||
@ -48,7 +51,12 @@ import {
|
||||
convertLinksToOpenRouter,
|
||||
convertLinksToZhipu
|
||||
} from '@renderer/utils/linkConverter'
|
||||
import { mcpToolCallResponseToOpenAICompatibleMessage, parseAndCallTools } from '@renderer/utils/mcp-tools'
|
||||
import {
|
||||
mcpToolCallResponseToOpenAICompatibleMessage,
|
||||
mcpToolsToOpenAIChatTools,
|
||||
openAIToolsToMcpTool,
|
||||
parseAndCallTools
|
||||
} from '@renderer/utils/mcp-tools'
|
||||
import { findFileBlocks, findImageBlocks, getMainTextContent } from '@renderer/utils/messageUtils/find'
|
||||
import { buildSystemPrompt } from '@renderer/utils/prompt'
|
||||
import { asyncGeneratorToReadableStream, readableStreamAsyncIterable } from '@renderer/utils/stream'
|
||||
@ -57,18 +65,22 @@ import OpenAI, { AzureOpenAI } from 'openai'
|
||||
import {
|
||||
ChatCompletionContentPart,
|
||||
ChatCompletionCreateParamsNonStreaming,
|
||||
ChatCompletionMessageParam
|
||||
ChatCompletionMessageParam,
|
||||
ChatCompletionMessageToolCall,
|
||||
ChatCompletionTool,
|
||||
ChatCompletionToolMessageParam
|
||||
} from 'openai/resources'
|
||||
|
||||
import { CompletionsParams } from '.'
|
||||
import OpenAIProvider from './OpenAIProvider'
|
||||
import { BaseOpenAiProvider } from './OpenAIProvider'
|
||||
|
||||
// 1. 定义联合类型
|
||||
export type OpenAIStreamChunk =
|
||||
| { type: 'reasoning' | 'text-delta'; textDelta: string }
|
||||
| { type: 'tool-calls'; delta: any }
|
||||
| { type: 'finish'; finishReason: any; usage: any; delta: any; chunk: any }
|
||||
|
||||
export default class OpenAICompatibleProvider extends OpenAIProvider {
|
||||
export default class OpenAICompatibleProvider extends BaseOpenAiProvider {
|
||||
constructor(provider: Provider) {
|
||||
super(provider)
|
||||
|
||||
@ -313,6 +325,24 @@ export default class OpenAICompatibleProvider extends OpenAIProvider {
|
||||
return {}
|
||||
}
|
||||
|
||||
public convertMcpTools<T>(mcpTools: MCPTool[]): T[] {
|
||||
return mcpToolsToOpenAIChatTools(mcpTools) as T[]
|
||||
}
|
||||
|
||||
public mcpToolCallResponseToMessage = (mcpToolResponse: MCPToolResponse, resp: MCPCallToolResponse, model: Model) => {
|
||||
if ('toolUseId' in mcpToolResponse && mcpToolResponse.toolUseId) {
|
||||
return mcpToolCallResponseToOpenAICompatibleMessage(mcpToolResponse, resp, isVisionModel(model))
|
||||
} else if ('toolCallId' in mcpToolResponse && mcpToolResponse.toolCallId) {
|
||||
const toolCallOut: ChatCompletionToolMessageParam = {
|
||||
role: 'tool',
|
||||
tool_call_id: mcpToolResponse.toolCallId,
|
||||
content: JSON.stringify(resp.content)
|
||||
}
|
||||
return toolCallOut
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
/**
|
||||
* Generate completions for the assistant
|
||||
* @param messages - The messages
|
||||
@ -330,7 +360,7 @@ export default class OpenAICompatibleProvider extends OpenAIProvider {
|
||||
const defaultModel = getDefaultModel()
|
||||
const model = assistant.model || defaultModel
|
||||
|
||||
const { contextCount, maxTokens, streamOutput } = getAssistantSettings(assistant)
|
||||
const { contextCount, maxTokens, streamOutput, enableToolUse } = getAssistantSettings(assistant)
|
||||
const isEnabledWebSearch = assistant.enableWebSearch || !!assistant.webSearchProviderId
|
||||
messages = addImageFileToContents(messages)
|
||||
const enableReasoning =
|
||||
@ -344,7 +374,9 @@ export default class OpenAICompatibleProvider extends OpenAIProvider {
|
||||
content: `Formatting re-enabled${systemMessage ? '\n' + systemMessage.content : ''}`
|
||||
}
|
||||
}
|
||||
if (mcpTools && mcpTools.length > 0) {
|
||||
const { tools } = this.setupToolsConfig<ChatCompletionTool>({ mcpTools, model, enableToolUse })
|
||||
|
||||
if (this.useSystemPromptForTools) {
|
||||
systemMessage.content = buildSystemPrompt(systemMessage.content || '', mcpTools)
|
||||
}
|
||||
|
||||
@ -379,53 +411,86 @@ export default class OpenAICompatibleProvider extends OpenAIProvider {
|
||||
|
||||
const toolResponses: MCPToolResponse[] = []
|
||||
|
||||
const processToolUses = async (content: string, idx: number) => {
|
||||
const toolResults = await parseAndCallTools(
|
||||
const processToolResults = async (toolResults: Awaited<ReturnType<typeof parseAndCallTools>>, idx: number) => {
|
||||
if (toolResults.length === 0) return
|
||||
|
||||
toolResults.forEach((ts) => reqMessages.push(ts as ChatCompletionMessageParam))
|
||||
|
||||
console.debug('[tool] reqMessages before processing', model.id, reqMessages)
|
||||
reqMessages = processReqMessages(model, reqMessages)
|
||||
console.debug('[tool] reqMessages', model.id, reqMessages)
|
||||
|
||||
onChunk({ type: ChunkType.LLM_RESPONSE_CREATED })
|
||||
const newStream = await this.sdk.chat.completions
|
||||
// @ts-ignore key is not typed
|
||||
.create(
|
||||
{
|
||||
model: model.id,
|
||||
messages: reqMessages,
|
||||
temperature: this.getTemperature(assistant, model),
|
||||
top_p: this.getTopP(assistant, model),
|
||||
max_tokens: maxTokens,
|
||||
keep_alive: this.keepAliveTime,
|
||||
stream: isSupportStreamOutput(),
|
||||
tools: !isEmpty(tools) ? tools : undefined,
|
||||
...getOpenAIWebSearchParams(assistant, model),
|
||||
...this.getReasoningEffort(assistant, model),
|
||||
...this.getProviderSpecificParameters(assistant, model),
|
||||
...this.getCustomParameters(assistant)
|
||||
},
|
||||
{
|
||||
signal
|
||||
}
|
||||
)
|
||||
await processStream(newStream, idx + 1)
|
||||
}
|
||||
|
||||
const processToolCalls = async (mcpTools, toolCalls: ChatCompletionMessageToolCall[]) => {
|
||||
const mcpToolResponses = toolCalls
|
||||
.map((toolCall) => {
|
||||
const mcpTool = openAIToolsToMcpTool(mcpTools, toolCall as ChatCompletionMessageToolCall)
|
||||
if (!mcpTool) return undefined
|
||||
|
||||
const parsedArgs = (() => {
|
||||
try {
|
||||
return JSON.parse(toolCall.function.arguments)
|
||||
} catch {
|
||||
return toolCall.function.arguments
|
||||
}
|
||||
})()
|
||||
|
||||
return {
|
||||
id: toolCall.id,
|
||||
toolCallId: toolCall.id,
|
||||
tool: mcpTool,
|
||||
arguments: parsedArgs,
|
||||
status: 'pending'
|
||||
} as ToolCallResponse
|
||||
})
|
||||
.filter((t): t is ToolCallResponse => typeof t !== 'undefined')
|
||||
return await parseAndCallTools(
|
||||
mcpToolResponses,
|
||||
toolResponses,
|
||||
onChunk,
|
||||
this.mcpToolCallResponseToMessage,
|
||||
model,
|
||||
mcpTools
|
||||
)
|
||||
}
|
||||
|
||||
const processToolUses = async (content: string) => {
|
||||
return await parseAndCallTools(
|
||||
content,
|
||||
toolResponses,
|
||||
onChunk,
|
||||
idx,
|
||||
mcpToolCallResponseToOpenAICompatibleMessage,
|
||||
mcpTools,
|
||||
isVisionModel(model)
|
||||
this.mcpToolCallResponseToMessage,
|
||||
model,
|
||||
mcpTools
|
||||
)
|
||||
|
||||
if (toolResults.length > 0) {
|
||||
reqMessages.push({
|
||||
role: 'assistant',
|
||||
content: content
|
||||
} as ChatCompletionMessageParam)
|
||||
toolResults.forEach((ts) => reqMessages.push(ts as ChatCompletionMessageParam))
|
||||
|
||||
reqMessages = processReqMessages(model, reqMessages)
|
||||
const newStream = await this.sdk.chat.completions
|
||||
// @ts-ignore key is not typed
|
||||
.create(
|
||||
{
|
||||
model: model.id,
|
||||
messages: reqMessages,
|
||||
temperature: this.getTemperature(assistant, model),
|
||||
top_p: this.getTopP(assistant, model),
|
||||
max_tokens: maxTokens,
|
||||
keep_alive: this.keepAliveTime,
|
||||
stream: isSupportStreamOutput(),
|
||||
// tools: tools,
|
||||
service_tier: this.getServiceTier(model),
|
||||
...getOpenAIWebSearchParams(assistant, model),
|
||||
...this.getReasoningEffort(assistant, model),
|
||||
...this.getProviderSpecificParameters(assistant, model),
|
||||
...this.getCustomParameters(assistant)
|
||||
},
|
||||
{
|
||||
signal,
|
||||
timeout: this.getTimeout(model)
|
||||
}
|
||||
)
|
||||
await processStream(newStream, idx + 1)
|
||||
}
|
||||
}
|
||||
|
||||
const processStream = async (stream: any, idx: number) => {
|
||||
const toolCalls: ChatCompletionMessageToolCall[] = []
|
||||
// Handle non-streaming case (already returns early, no change needed here)
|
||||
if (!isSupportStreamOutput()) {
|
||||
const time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
@ -439,10 +504,59 @@ export default class OpenAICompatibleProvider extends OpenAIProvider {
|
||||
// Create a synthetic usage object if stream.usage is undefined
|
||||
const finalUsage = stream.usage
|
||||
// Separate onChunk calls for text and usage/metrics
|
||||
if (stream.choices[0].message?.content) {
|
||||
onChunk({ type: ChunkType.TEXT_COMPLETE, text: stream.choices[0].message.content })
|
||||
let content = ''
|
||||
stream.choices.forEach((choice) => {
|
||||
// reasoning
|
||||
if (choice.message.reasoning) {
|
||||
onChunk({ type: ChunkType.THINKING_DELTA, text: choice.message.reasoning })
|
||||
onChunk({
|
||||
type: ChunkType.THINKING_COMPLETE,
|
||||
text: choice.message.reasoning,
|
||||
thinking_millsec: time_completion_millsec
|
||||
})
|
||||
}
|
||||
// text
|
||||
if (choice.message.content) {
|
||||
content += choice.message.content
|
||||
onChunk({ type: ChunkType.TEXT_DELTA, text: choice.message.content })
|
||||
}
|
||||
// tool call
|
||||
if (choice.message.tool_calls && choice.message.tool_calls.length) {
|
||||
choice.message.tool_calls.forEach((t) => toolCalls.push(t))
|
||||
}
|
||||
|
||||
reqMessages.push({
|
||||
role: choice.message.role,
|
||||
content: choice.message.content,
|
||||
tool_calls: toolCalls.length
|
||||
? toolCalls.map((toolCall) => ({
|
||||
id: toolCall.id,
|
||||
function: {
|
||||
...toolCall.function,
|
||||
arguments:
|
||||
typeof toolCall.function.arguments === 'string'
|
||||
? toolCall.function.arguments
|
||||
: JSON.stringify(toolCall.function.arguments)
|
||||
},
|
||||
type: 'function'
|
||||
}))
|
||||
: undefined
|
||||
})
|
||||
})
|
||||
|
||||
if (content.length) {
|
||||
onChunk({ type: ChunkType.TEXT_COMPLETE, text: content })
|
||||
}
|
||||
|
||||
const toolResults: Awaited<ReturnType<typeof parseAndCallTools>> = []
|
||||
if (toolCalls.length) {
|
||||
toolResults.push(...(await processToolCalls(mcpTools, toolCalls)))
|
||||
}
|
||||
if (stream.choices[0].message?.content) {
|
||||
toolResults.push(...(await processToolUses(stream.choices[0].message?.content)))
|
||||
}
|
||||
await processToolResults(toolResults, idx)
|
||||
|
||||
// Always send usage and metrics data
|
||||
onChunk({ type: ChunkType.BLOCK_COMPLETE, response: { usage: finalUsage, metrics: finalMetrics } })
|
||||
return
|
||||
@ -486,6 +600,9 @@ export default class OpenAICompatibleProvider extends OpenAIProvider {
|
||||
if (delta?.content) {
|
||||
yield { type: 'text-delta', textDelta: delta.content }
|
||||
}
|
||||
if (delta?.tool_calls) {
|
||||
yield { type: 'tool-calls', delta: delta }
|
||||
}
|
||||
|
||||
const finishReason = chunk.choices[0]?.finish_reason
|
||||
if (!isEmpty(finishReason)) {
|
||||
@ -563,6 +680,25 @@ export default class OpenAICompatibleProvider extends OpenAIProvider {
|
||||
onChunk({ type: ChunkType.TEXT_DELTA, text: textDelta })
|
||||
break
|
||||
}
|
||||
case 'tool-calls': {
|
||||
chunk.delta.tool_calls.forEach((toolCall) => {
|
||||
const { id, index, type, function: fun } = toolCall
|
||||
if (id && type === 'function' && fun) {
|
||||
const { name, arguments: args } = fun
|
||||
toolCalls.push({
|
||||
id,
|
||||
function: {
|
||||
name: name || '',
|
||||
arguments: args || ''
|
||||
},
|
||||
type: 'function'
|
||||
})
|
||||
} else if (fun?.arguments) {
|
||||
toolCalls[index].function.arguments += fun.arguments
|
||||
}
|
||||
})
|
||||
break
|
||||
}
|
||||
case 'finish': {
|
||||
const finishReason = chunk.finishReason
|
||||
const usage = chunk.usage
|
||||
@ -624,7 +760,33 @@ export default class OpenAICompatibleProvider extends OpenAIProvider {
|
||||
} as LLMWebSearchCompleteChunk)
|
||||
}
|
||||
}
|
||||
await processToolUses(content, idx)
|
||||
reqMessages.push({
|
||||
role: 'assistant',
|
||||
content: content,
|
||||
tool_calls: toolCalls.length
|
||||
? toolCalls.map((toolCall) => ({
|
||||
id: toolCall.id,
|
||||
function: {
|
||||
...toolCall.function,
|
||||
arguments:
|
||||
typeof toolCall.function.arguments === 'string'
|
||||
? toolCall.function.arguments
|
||||
: JSON.stringify(toolCall.function.arguments)
|
||||
},
|
||||
type: 'function'
|
||||
}))
|
||||
: undefined
|
||||
})
|
||||
let toolResults: Awaited<ReturnType<typeof parseAndCallTools>> = []
|
||||
if (toolCalls.length) {
|
||||
toolResults = await processToolCalls(mcpTools, toolCalls)
|
||||
}
|
||||
if (content.length) {
|
||||
toolResults = toolResults.concat(await processToolUses(content))
|
||||
}
|
||||
if (toolResults.length) {
|
||||
await processToolResults(toolResults, idx)
|
||||
}
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
@ -657,7 +819,7 @@ export default class OpenAICompatibleProvider extends OpenAIProvider {
|
||||
max_tokens: maxTokens,
|
||||
keep_alive: this.keepAliveTime,
|
||||
stream: isSupportStreamOutput(),
|
||||
// tools: tools,
|
||||
tools: !isEmpty(tools) ? tools : undefined,
|
||||
service_tier: this.getServiceTier(model),
|
||||
...getOpenAIWebSearchParams(assistant, model),
|
||||
...this.getReasoningEffort(assistant, model),
|
||||
|
||||
@ -21,10 +21,13 @@ import {
|
||||
Assistant,
|
||||
FileTypes,
|
||||
GenerateImageParams,
|
||||
MCPCallToolResponse,
|
||||
MCPTool,
|
||||
MCPToolResponse,
|
||||
Model,
|
||||
Provider,
|
||||
Suggestion,
|
||||
ToolCallResponse,
|
||||
Usage,
|
||||
WebSearchSource
|
||||
} from '@renderer/types'
|
||||
@ -33,7 +36,12 @@ import { Message } from '@renderer/types/newMessage'
|
||||
import { removeSpecialCharactersForTopicName } from '@renderer/utils'
|
||||
import { addImageFileToContents } from '@renderer/utils/formats'
|
||||
import { convertLinks } from '@renderer/utils/linkConverter'
|
||||
import { mcpToolCallResponseToOpenAIMessage, parseAndCallTools } from '@renderer/utils/mcp-tools'
|
||||
import {
|
||||
mcpToolCallResponseToOpenAIMessage,
|
||||
mcpToolsToOpenAIResponseTools,
|
||||
openAIToolsToMcpTool,
|
||||
parseAndCallTools
|
||||
} from '@renderer/utils/mcp-tools'
|
||||
import { findFileBlocks, findImageBlocks, getMainTextContent } from '@renderer/utils/messageUtils/find'
|
||||
import { buildSystemPrompt } from '@renderer/utils/prompt'
|
||||
import { isEmpty, takeRight } from 'lodash'
|
||||
@ -45,7 +53,7 @@ import { FileLike, toFile } from 'openai/uploads'
|
||||
import { CompletionsParams } from '.'
|
||||
import BaseProvider from './BaseProvider'
|
||||
|
||||
export default class OpenAIProvider extends BaseProvider {
|
||||
export abstract class BaseOpenAiProvider extends BaseProvider {
|
||||
protected sdk: OpenAI
|
||||
|
||||
constructor(provider: Provider) {
|
||||
@ -61,6 +69,14 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
})
|
||||
}
|
||||
|
||||
abstract convertMcpTools<T>(mcpTools: MCPTool[]): T[]
|
||||
|
||||
abstract mcpToolCallResponseToMessage: (
|
||||
mcpToolResponse: MCPToolResponse,
|
||||
resp: MCPCallToolResponse,
|
||||
model: Model
|
||||
) => OpenAI.Responses.ResponseInputItem | ChatCompletionMessageParam | undefined
|
||||
|
||||
/**
|
||||
* Extract the file content from the message
|
||||
* @param message - The message
|
||||
@ -91,16 +107,23 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
return ''
|
||||
}
|
||||
|
||||
private async getReponseMessageParam(message: Message, model: Model): Promise<OpenAI.Responses.EasyInputMessage> {
|
||||
private async getReponseMessageParam(message: Message, model: Model): Promise<OpenAI.Responses.ResponseInputItem> {
|
||||
const isVision = isVisionModel(model)
|
||||
const content = await this.getMessageContent(message)
|
||||
const fileBlocks = findFileBlocks(message)
|
||||
const imageBlocks = findImageBlocks(message)
|
||||
|
||||
if (fileBlocks.length === 0 && imageBlocks.length === 0) {
|
||||
return {
|
||||
role: message.role === 'system' ? 'user' : message.role,
|
||||
content: content ? [{ type: 'input_text', text: content }] : []
|
||||
if (message.role === 'assistant') {
|
||||
return {
|
||||
role: 'assistant',
|
||||
content: content
|
||||
}
|
||||
} else {
|
||||
return {
|
||||
role: message.role === 'system' ? 'user' : message.role,
|
||||
content: content ? [{ type: 'input_text', text: content }] : []
|
||||
} as OpenAI.Responses.EasyInputMessage
|
||||
}
|
||||
}
|
||||
|
||||
@ -285,10 +308,8 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
}
|
||||
const defaultModel = getDefaultModel()
|
||||
const model = assistant.model || defaultModel
|
||||
const { contextCount, maxTokens, streamOutput } = getAssistantSettings(assistant)
|
||||
|
||||
const { contextCount, maxTokens, streamOutput, enableToolUse } = getAssistantSettings(assistant)
|
||||
const isEnabledWebSearch = assistant.enableWebSearch || !!assistant.webSearchProviderId
|
||||
onChunk({ type: ChunkType.LLM_RESPONSE_CREATED })
|
||||
// 退回到 OpenAI 兼容模式
|
||||
if (isOpenAIWebSearch(model)) {
|
||||
const systemMessage = { role: 'system', content: assistant.prompt || '' }
|
||||
@ -387,7 +408,7 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
})
|
||||
return
|
||||
}
|
||||
const tools: OpenAI.Responses.Tool[] = []
|
||||
let tools: OpenAI.Responses.Tool[] = []
|
||||
if (isEnabledWebSearch) {
|
||||
tools.push({
|
||||
type: 'web_search_preview'
|
||||
@ -408,7 +429,15 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
systemMessage.role = 'developer'
|
||||
}
|
||||
|
||||
if (mcpTools && mcpTools.length > 0) {
|
||||
const { tools: extraTools } = this.setupToolsConfig<OpenAI.Responses.Tool>({
|
||||
mcpTools,
|
||||
model,
|
||||
enableToolUse
|
||||
})
|
||||
|
||||
tools = tools.concat(extraTools)
|
||||
|
||||
if (this.useSystemPromptForTools) {
|
||||
systemMessageInput.text = buildSystemPrompt(systemMessageInput.text || '', mcpTools)
|
||||
}
|
||||
systemMessageContent.push(systemMessageInput)
|
||||
@ -418,7 +447,7 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
)
|
||||
|
||||
onFilterMessages(_messages)
|
||||
const userMessage: OpenAI.Responses.EasyInputMessage[] = []
|
||||
const userMessage: OpenAI.Responses.ResponseInputItem[] = []
|
||||
for (const message of _messages) {
|
||||
userMessage.push(await this.getReponseMessageParam(message, model))
|
||||
}
|
||||
@ -431,7 +460,7 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
const { signal } = abortController
|
||||
|
||||
// 当 systemMessage 内容为空时不发送 systemMessage
|
||||
let reqMessages: OpenAI.Responses.EasyInputMessage[]
|
||||
let reqMessages: OpenAI.Responses.ResponseInput
|
||||
if (!systemMessage.content) {
|
||||
reqMessages = [...userMessage]
|
||||
} else {
|
||||
@ -440,48 +469,84 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
|
||||
const toolResponses: MCPToolResponse[] = []
|
||||
|
||||
const processToolUses = async (content: string, idx: number) => {
|
||||
const toolResults = await parseAndCallTools(
|
||||
const processToolResults = async (toolResults: Awaited<ReturnType<typeof parseAndCallTools>>, idx: number) => {
|
||||
if (toolResults.length === 0) return
|
||||
|
||||
toolResults.forEach((ts) => reqMessages.push(ts as OpenAI.Responses.EasyInputMessage))
|
||||
|
||||
onChunk({ type: ChunkType.LLM_RESPONSE_CREATED })
|
||||
const stream = await this.sdk.responses.create(
|
||||
{
|
||||
model: model.id,
|
||||
input: reqMessages,
|
||||
temperature: this.getTemperature(assistant, model),
|
||||
top_p: this.getTopP(assistant, model),
|
||||
max_output_tokens: maxTokens,
|
||||
stream: streamOutput,
|
||||
tools: !isEmpty(tools) ? tools : undefined,
|
||||
service_tier: this.getServiceTier(model),
|
||||
...this.getResponseReasoningEffort(assistant, model),
|
||||
...this.getCustomParameters(assistant)
|
||||
},
|
||||
{
|
||||
signal,
|
||||
timeout: this.getTimeout(model)
|
||||
}
|
||||
)
|
||||
await processStream(stream, idx + 1)
|
||||
}
|
||||
|
||||
const processToolCalls = async (mcpTools, toolCalls: OpenAI.Responses.ResponseFunctionToolCall[]) => {
|
||||
const mcpToolResponses = toolCalls
|
||||
.map((toolCall) => {
|
||||
const mcpTool = openAIToolsToMcpTool(mcpTools, toolCall as OpenAI.Responses.ResponseFunctionToolCall)
|
||||
if (!mcpTool) return undefined
|
||||
|
||||
const parsedArgs = (() => {
|
||||
try {
|
||||
return JSON.parse(toolCall.arguments)
|
||||
} catch {
|
||||
return toolCall.arguments
|
||||
}
|
||||
})()
|
||||
|
||||
return {
|
||||
id: toolCall.call_id,
|
||||
toolCallId: toolCall.call_id,
|
||||
tool: mcpTool,
|
||||
arguments: parsedArgs,
|
||||
status: 'pending'
|
||||
} as ToolCallResponse
|
||||
})
|
||||
.filter((t): t is ToolCallResponse => typeof t !== 'undefined')
|
||||
|
||||
return await parseAndCallTools<OpenAI.Responses.ResponseInputItem | ChatCompletionMessageParam>(
|
||||
mcpToolResponses,
|
||||
toolResponses,
|
||||
onChunk,
|
||||
this.mcpToolCallResponseToMessage,
|
||||
model,
|
||||
mcpTools
|
||||
)
|
||||
}
|
||||
|
||||
const processToolUses = async (content: string) => {
|
||||
return await parseAndCallTools(
|
||||
content,
|
||||
toolResponses,
|
||||
onChunk,
|
||||
idx,
|
||||
mcpToolCallResponseToOpenAIMessage,
|
||||
mcpTools,
|
||||
isVisionModel(model)
|
||||
this.mcpToolCallResponseToMessage,
|
||||
model,
|
||||
mcpTools
|
||||
)
|
||||
|
||||
if (toolResults.length > 0) {
|
||||
reqMessages.push({
|
||||
role: 'assistant',
|
||||
content: content
|
||||
})
|
||||
toolResults.forEach((ts) => reqMessages.push(ts as OpenAI.Responses.EasyInputMessage))
|
||||
const newStream = await this.sdk.responses.create(
|
||||
{
|
||||
model: model.id,
|
||||
input: reqMessages,
|
||||
temperature: this.getTemperature(assistant, model),
|
||||
top_p: this.getTopP(assistant, model),
|
||||
max_output_tokens: maxTokens,
|
||||
stream: true,
|
||||
service_tier: this.getServiceTier(model),
|
||||
...this.getResponseReasoningEffort(assistant, model),
|
||||
...this.getCustomParameters(assistant)
|
||||
},
|
||||
{
|
||||
signal,
|
||||
timeout: this.getTimeout(model)
|
||||
}
|
||||
)
|
||||
await processStream(newStream, idx + 1)
|
||||
}
|
||||
}
|
||||
|
||||
const processStream = async (
|
||||
stream: Stream<OpenAI.Responses.ResponseStreamEvent> | OpenAI.Responses.Response,
|
||||
idx: number
|
||||
) => {
|
||||
const toolCalls: OpenAI.Responses.ResponseFunctionToolCall[] = []
|
||||
|
||||
if (!streamOutput) {
|
||||
const nonStream = stream as OpenAI.Responses.Response
|
||||
const time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
@ -499,11 +564,15 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
prompt_tokens: nonStream.usage?.input_tokens || 0,
|
||||
total_tokens
|
||||
}
|
||||
let content = ''
|
||||
|
||||
for (const output of nonStream.output) {
|
||||
switch (output.type) {
|
||||
case 'message':
|
||||
if (output.content[0].type === 'output_text') {
|
||||
onChunk({ type: ChunkType.TEXT_DELTA, text: output.content[0].text })
|
||||
onChunk({ type: ChunkType.TEXT_COMPLETE, text: output.content[0].text })
|
||||
content += output.content[0].text
|
||||
if (output.content[0].annotations && output.content[0].annotations.length > 0) {
|
||||
onChunk({
|
||||
type: ChunkType.LLM_WEB_SEARCH_COMPLETE,
|
||||
@ -522,8 +591,32 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
thinking_millsec: new Date().getTime() - start_time_millsec
|
||||
})
|
||||
break
|
||||
case 'function_call':
|
||||
toolCalls.push(output)
|
||||
}
|
||||
}
|
||||
|
||||
if (content) {
|
||||
reqMessages.push({
|
||||
role: 'assistant',
|
||||
content: content
|
||||
})
|
||||
}
|
||||
if (toolCalls.length) {
|
||||
toolCalls.forEach((toolCall) => {
|
||||
reqMessages.push(toolCall)
|
||||
})
|
||||
}
|
||||
|
||||
const toolResults: Awaited<ReturnType<typeof parseAndCallTools>> = []
|
||||
if (toolCalls.length) {
|
||||
toolResults.push(...(await processToolCalls(mcpTools, toolCalls)))
|
||||
}
|
||||
if (content.length) {
|
||||
toolResults.push(...(await processToolUses(content)))
|
||||
}
|
||||
await processToolResults(toolResults, idx)
|
||||
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
@ -534,6 +627,9 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
return
|
||||
}
|
||||
let content = ''
|
||||
|
||||
const outputItems: OpenAI.Responses.ResponseOutputItem[] = []
|
||||
|
||||
let lastUsage: Usage | undefined = undefined
|
||||
let final_time_completion_millsec_delta = 0
|
||||
for await (const chunk of stream as Stream<OpenAI.Responses.ResponseStreamEvent>) {
|
||||
@ -544,6 +640,12 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
case 'response.created':
|
||||
time_first_token_millsec = new Date().getTime()
|
||||
break
|
||||
case 'response.output_item.added':
|
||||
if (chunk.item.type === 'function_call') {
|
||||
outputItems.push(chunk.item)
|
||||
}
|
||||
break
|
||||
|
||||
case 'response.reasoning_summary_text.delta':
|
||||
onChunk({
|
||||
type: ChunkType.THINKING_DELTA,
|
||||
@ -571,6 +673,21 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
text: chunk.text
|
||||
})
|
||||
break
|
||||
case 'response.function_call_arguments.done': {
|
||||
const outputItem: OpenAI.Responses.ResponseOutputItem | undefined = outputItems.find(
|
||||
(item) => item.id === chunk.item_id
|
||||
)
|
||||
if (outputItem) {
|
||||
if (outputItem.type === 'function_call') {
|
||||
toolCalls.push({
|
||||
...outputItem,
|
||||
arguments: chunk.arguments
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
break
|
||||
}
|
||||
case 'response.content_part.done':
|
||||
if (chunk.part.type === 'output_text' && chunk.part.annotations && chunk.part.annotations.length > 0) {
|
||||
onChunk({
|
||||
@ -607,9 +724,31 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
})
|
||||
break
|
||||
}
|
||||
|
||||
// --- End of Incremental onChunk calls ---
|
||||
} // End of for await loop
|
||||
if (content) {
|
||||
reqMessages.push({
|
||||
role: 'assistant',
|
||||
content: content
|
||||
})
|
||||
}
|
||||
if (toolCalls.length) {
|
||||
toolCalls.forEach((toolCall) => {
|
||||
reqMessages.push(toolCall)
|
||||
})
|
||||
}
|
||||
|
||||
await processToolUses(content, idx)
|
||||
// Call processToolUses AFTER the loop finishes processing the main stream content
|
||||
// Note: parseAndCallTools inside processToolUses should handle its own onChunk for tool responses
|
||||
const toolResults: Awaited<ReturnType<typeof parseAndCallTools>> = []
|
||||
if (toolCalls.length) {
|
||||
toolResults.push(...(await processToolCalls(mcpTools, toolCalls)))
|
||||
}
|
||||
if (content) {
|
||||
toolResults.push(...(await processToolUses(content)))
|
||||
}
|
||||
await processToolResults(toolResults, idx)
|
||||
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
@ -624,6 +763,7 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
})
|
||||
}
|
||||
|
||||
onChunk({ type: ChunkType.LLM_RESPONSE_CREATED })
|
||||
const stream = await this.sdk.responses.create(
|
||||
{
|
||||
model: model.id,
|
||||
@ -1072,3 +1212,31 @@ export default class OpenAIProvider extends BaseProvider {
|
||||
return data.data[0].embedding.length
|
||||
}
|
||||
}
|
||||
|
||||
export default class OpenAIProvider extends BaseOpenAiProvider {
|
||||
constructor(provider: Provider) {
|
||||
super(provider)
|
||||
}
|
||||
|
||||
public convertMcpTools<T>(mcpTools: MCPTool[]) {
|
||||
return mcpToolsToOpenAIResponseTools(mcpTools) as T[]
|
||||
}
|
||||
|
||||
public mcpToolCallResponseToMessage = (
|
||||
mcpToolResponse: MCPToolResponse,
|
||||
resp: MCPCallToolResponse,
|
||||
model: Model
|
||||
): OpenAI.Responses.ResponseInputItem | undefined => {
|
||||
if ('toolUseId' in mcpToolResponse && mcpToolResponse.toolUseId) {
|
||||
return mcpToolCallResponseToOpenAIMessage(mcpToolResponse, resp, isVisionModel(model))
|
||||
} else if ('toolCallId' in mcpToolResponse && mcpToolResponse.toolCallId) {
|
||||
const toolCallOut: OpenAI.Responses.ResponseInputItem = {
|
||||
type: 'function_call_output',
|
||||
call_id: mcpToolResponse.toolCallId,
|
||||
output: JSON.stringify(resp.content)
|
||||
}
|
||||
return toolCallOut
|
||||
}
|
||||
return
|
||||
}
|
||||
}
|
||||
|
||||
@ -107,6 +107,7 @@ export const getAssistantSettings = (assistant: Assistant): AssistantSettings =>
|
||||
enableMaxTokens: assistant?.settings?.enableMaxTokens ?? false,
|
||||
maxTokens: getAssistantMaxTokens(),
|
||||
streamOutput: assistant?.settings?.streamOutput ?? true,
|
||||
enableToolUse: assistant?.settings?.enableToolUse ?? false,
|
||||
hideMessages: assistant?.settings?.hideMessages ?? false,
|
||||
defaultModel: assistant?.defaultModel ?? undefined,
|
||||
customParameters: assistant?.settings?.customParameters ?? []
|
||||
|
||||
@ -423,7 +423,17 @@ const fetchAndProcessAssistantResponseImpl = async (
|
||||
}
|
||||
},
|
||||
onToolCallInProgress: (toolResponse: MCPToolResponse) => {
|
||||
if (toolResponse.status === 'invoking') {
|
||||
if (lastBlockType === MessageBlockType.UNKNOWN && lastBlockId) {
|
||||
lastBlockType = MessageBlockType.TOOL
|
||||
const changes = {
|
||||
type: MessageBlockType.TOOL,
|
||||
status: MessageBlockStatus.PROCESSING,
|
||||
metadata: { rawMcpToolResponse: toolResponse }
|
||||
}
|
||||
dispatch(updateOneBlock({ id: lastBlockId, changes }))
|
||||
saveUpdatedBlockToDB(lastBlockId, assistantMsgId, topicId, getState)
|
||||
toolCallIdToBlockIdMap.set(toolResponse.id, lastBlockId)
|
||||
} else if (toolResponse.status === 'invoking') {
|
||||
const toolBlock = createToolBlock(assistantMsgId, toolResponse.id, {
|
||||
toolName: toolResponse.tool.name,
|
||||
status: MessageBlockStatus.PROCESSING,
|
||||
|
||||
@ -55,6 +55,7 @@ export type AssistantSettings = {
|
||||
maxTokens: number | undefined
|
||||
enableMaxTokens: boolean
|
||||
streamOutput: boolean
|
||||
enableToolUse: boolean
|
||||
hideMessages: boolean
|
||||
defaultModel?: Model
|
||||
customParameters?: AssistantSettingCustomParameters[]
|
||||
@ -570,13 +571,25 @@ export interface MCPConfig {
|
||||
servers: MCPServer[]
|
||||
}
|
||||
|
||||
export interface MCPToolResponse {
|
||||
id: string // tool call id, it should be unique
|
||||
tool: MCPTool // tool info
|
||||
interface BaseToolResponse {
|
||||
id: string // unique id
|
||||
tool: MCPTool
|
||||
arguments: Record<string, unknown> | undefined
|
||||
status: string // 'invoking' | 'done'
|
||||
response?: any
|
||||
}
|
||||
|
||||
export interface ToolUseResponse extends BaseToolResponse {
|
||||
toolUseId: string
|
||||
}
|
||||
|
||||
export interface ToolCallResponse extends BaseToolResponse {
|
||||
// gemini tool call id might be undefined
|
||||
toolCallId?: string
|
||||
}
|
||||
|
||||
export type MCPToolResponse = ToolUseResponse | ToolCallResponse
|
||||
|
||||
export interface MCPToolResultContent {
|
||||
type: 'text' | 'image' | 'audio' | 'resource'
|
||||
text?: string
|
||||
@ -586,6 +599,7 @@ export interface MCPToolResultContent {
|
||||
uri?: string
|
||||
text?: string
|
||||
mimeType?: string
|
||||
blob?: string
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@ -1,18 +1,31 @@
|
||||
import { ContentBlockParam, ToolUnion, ToolUseBlock } from '@anthropic-ai/sdk/resources'
|
||||
import { MessageParam } from '@anthropic-ai/sdk/resources'
|
||||
import { Content, FunctionCall, Part } from '@google/genai'
|
||||
import {
|
||||
ContentBlockParam,
|
||||
MessageParam,
|
||||
ToolResultBlockParam,
|
||||
ToolUnion,
|
||||
ToolUseBlock
|
||||
} from '@anthropic-ai/sdk/resources'
|
||||
import { Content, FunctionCall, Part, Tool, Type as GeminiSchemaType } from '@google/genai'
|
||||
import { isVisionModel } from '@renderer/config/models'
|
||||
import store from '@renderer/store'
|
||||
import { addMCPServer } from '@renderer/store/mcp'
|
||||
import { MCPCallToolResponse, MCPServer, MCPTool, MCPToolResponse } from '@renderer/types'
|
||||
import { MCPCallToolResponse, MCPServer, MCPTool, MCPToolResponse, Model, ToolUseResponse } from '@renderer/types'
|
||||
import type { MCPToolCompleteChunk, MCPToolInProgressChunk } from '@renderer/types/chunk'
|
||||
import { ChunkType } from '@renderer/types/chunk'
|
||||
import { isArray, isObject, pull, transform } from 'lodash'
|
||||
import { nanoid } from 'nanoid'
|
||||
import OpenAI from 'openai'
|
||||
import { ChatCompletionContentPart, ChatCompletionMessageParam, ChatCompletionMessageToolCall } from 'openai/resources'
|
||||
import {
|
||||
ChatCompletionContentPart,
|
||||
ChatCompletionMessageParam,
|
||||
ChatCompletionMessageToolCall,
|
||||
ChatCompletionTool
|
||||
} from 'openai/resources'
|
||||
|
||||
import { CompletionsParams } from '../providers/AiProvider'
|
||||
|
||||
const MCP_AUTO_INSTALL_SERVER_NAME = '@cherry/mcp-auto-install'
|
||||
const EXTRA_SCHEMA_KEYS = ['schema', 'headers']
|
||||
|
||||
// const ensureValidSchema = (obj: Record<string, any>) => {
|
||||
// // Filter out unsupported keys for Gemini
|
||||
@ -153,77 +166,116 @@ const MCP_AUTO_INSTALL_SERVER_NAME = '@cherry/mcp-auto-install'
|
||||
// return processedProperties
|
||||
// }
|
||||
|
||||
// export function mcpToolsToOpenAITools(mcpTools: MCPTool[]): Array<ChatCompletionTool> {
|
||||
// return mcpTools.map((tool) => ({
|
||||
// type: 'function',
|
||||
// name: tool.name,
|
||||
// function: {
|
||||
// name: tool.id,
|
||||
// description: tool.description,
|
||||
// parameters: {
|
||||
// type: 'object',
|
||||
// properties: filterPropertieAttributes(tool)
|
||||
// }
|
||||
// }
|
||||
// }))
|
||||
// }
|
||||
|
||||
export function openAIToolsToMcpTool(
|
||||
mcpTools: MCPTool[] | undefined,
|
||||
llmTool: ChatCompletionMessageToolCall
|
||||
): MCPTool | undefined {
|
||||
if (!mcpTools) {
|
||||
return undefined
|
||||
export function filterProperties(
|
||||
properties: Record<string, any> | string | number | boolean | Array<Record<string, any> | string | number | boolean>,
|
||||
supportedKeys: string[]
|
||||
) {
|
||||
// If it is an array, recursively process each element
|
||||
if (isArray(properties)) {
|
||||
return properties.map((item) => filterProperties(item, supportedKeys))
|
||||
}
|
||||
|
||||
const tool = mcpTools.find(
|
||||
(mcptool) => mcptool.id === llmTool.function.name || mcptool.name === llmTool.function.name
|
||||
)
|
||||
// If it is an object, recursively process each property
|
||||
if (isObject(properties)) {
|
||||
return transform(
|
||||
properties,
|
||||
(result, value, key) => {
|
||||
if (key === 'properties') {
|
||||
result[key] = transform(value, (acc, v, k) => {
|
||||
acc[k] = filterProperties(v, supportedKeys)
|
||||
})
|
||||
|
||||
if (!tool) {
|
||||
console.warn('No MCP Tool found for tool call:', llmTool)
|
||||
return undefined
|
||||
result['additionalProperties'] = false
|
||||
result['required'] = pull(Object.keys(value), ...EXTRA_SCHEMA_KEYS)
|
||||
} else if (key === 'oneOf') {
|
||||
// openai only supports anyOf
|
||||
result['anyOf'] = filterProperties(value, supportedKeys)
|
||||
} else if (supportedKeys.includes(key)) {
|
||||
result[key] = filterProperties(value, supportedKeys)
|
||||
if (key === 'type' && value === 'object') {
|
||||
result['additionalProperties'] = false
|
||||
}
|
||||
}
|
||||
},
|
||||
{}
|
||||
)
|
||||
}
|
||||
|
||||
console.log(
|
||||
`[MCP] OpenAI Tool to MCP Tool: ${tool.serverName} ${tool.name}`,
|
||||
tool,
|
||||
'args',
|
||||
llmTool.function.arguments
|
||||
)
|
||||
// use this to parse the arguments and avoid parsing errors
|
||||
let args: any = {}
|
||||
try {
|
||||
args = JSON.parse(llmTool.function.arguments)
|
||||
} catch (e) {
|
||||
console.error('Error parsing arguments', e)
|
||||
}
|
||||
|
||||
return {
|
||||
id: tool.id,
|
||||
serverId: tool.serverId,
|
||||
serverName: tool.serverName,
|
||||
name: tool.name,
|
||||
description: tool.description,
|
||||
inputSchema: args
|
||||
}
|
||||
// Return other types directly (e.g., string, number, etc.)
|
||||
return properties
|
||||
}
|
||||
|
||||
export async function callMCPTool(tool: MCPTool): Promise<MCPCallToolResponse> {
|
||||
console.log(`[MCP] Calling Tool: ${tool.serverName} ${tool.name}`, tool)
|
||||
export function mcpToolsToOpenAIResponseTools(mcpTools: MCPTool[]): OpenAI.Responses.Tool[] {
|
||||
const schemaKeys = ['type', 'description', 'items', 'enum', 'additionalProperties', 'anyof']
|
||||
return mcpTools.map(
|
||||
(tool) =>
|
||||
({
|
||||
type: 'function',
|
||||
name: tool.id,
|
||||
parameters: {
|
||||
type: 'object',
|
||||
properties: filterProperties(tool.inputSchema, schemaKeys).properties,
|
||||
required: pull(Object.keys(tool.inputSchema.properties), ...EXTRA_SCHEMA_KEYS),
|
||||
additionalProperties: false
|
||||
},
|
||||
strict: true
|
||||
}) satisfies OpenAI.Responses.Tool
|
||||
)
|
||||
}
|
||||
|
||||
export function mcpToolsToOpenAIChatTools(mcpTools: MCPTool[]): Array<ChatCompletionTool> {
|
||||
return mcpTools.map(
|
||||
(tool) =>
|
||||
({
|
||||
type: 'function',
|
||||
function: {
|
||||
name: tool.id,
|
||||
description: tool.description,
|
||||
parameters: {
|
||||
type: 'object',
|
||||
properties: tool.inputSchema.properties,
|
||||
required: tool.inputSchema.required
|
||||
}
|
||||
}
|
||||
}) as ChatCompletionTool
|
||||
)
|
||||
}
|
||||
|
||||
export function openAIToolsToMcpTool(
|
||||
mcpTools: MCPTool[],
|
||||
toolCall: OpenAI.Responses.ResponseFunctionToolCall | ChatCompletionMessageToolCall
|
||||
): MCPTool | undefined {
|
||||
const tool = mcpTools.find((mcpTool) => {
|
||||
if ('name' in toolCall) {
|
||||
return mcpTool.id === toolCall.name || mcpTool.name === toolCall.name
|
||||
} else {
|
||||
return mcpTool.id === toolCall.function.name || mcpTool.name === toolCall.function.name
|
||||
}
|
||||
})
|
||||
|
||||
if (!tool) {
|
||||
console.warn('No MCP Tool found for tool call:', toolCall)
|
||||
return undefined
|
||||
}
|
||||
|
||||
return tool
|
||||
}
|
||||
|
||||
export async function callMCPTool(toolResponse: MCPToolResponse): Promise<MCPCallToolResponse> {
|
||||
console.log(`[MCP] Calling Tool: ${toolResponse.tool.serverName} ${toolResponse.tool.name}`, toolResponse.tool)
|
||||
try {
|
||||
const server = getMcpServerByTool(tool)
|
||||
const server = getMcpServerByTool(toolResponse.tool)
|
||||
|
||||
if (!server) {
|
||||
throw new Error(`Server not found: ${tool.serverName}`)
|
||||
throw new Error(`Server not found: ${toolResponse.tool.serverName}`)
|
||||
}
|
||||
|
||||
const resp = await window.api.mcp.callTool({
|
||||
server,
|
||||
name: tool.name,
|
||||
args: tool.inputSchema
|
||||
name: toolResponse.tool.name,
|
||||
args: toolResponse.arguments
|
||||
})
|
||||
if (tool.serverName === MCP_AUTO_INSTALL_SERVER_NAME) {
|
||||
if (toolResponse.tool.serverName === MCP_AUTO_INSTALL_SERVER_NAME) {
|
||||
if (resp.data) {
|
||||
const mcpServer: MCPServer = {
|
||||
id: `f${nanoid()}`,
|
||||
@ -241,16 +293,16 @@ export async function callMCPTool(tool: MCPTool): Promise<MCPCallToolResponse> {
|
||||
}
|
||||
}
|
||||
|
||||
console.log(`[MCP] Tool called: ${tool.serverName} ${tool.name}`, resp)
|
||||
console.log(`[MCP] Tool called: ${toolResponse.tool.serverName} ${toolResponse.tool.name}`, resp)
|
||||
return resp
|
||||
} catch (e) {
|
||||
console.error(`[MCP] Error calling Tool: ${tool.serverName} ${tool.name}`, e)
|
||||
console.error(`[MCP] Error calling Tool: ${toolResponse.tool.serverName} ${toolResponse.tool.name}`, e)
|
||||
return Promise.resolve({
|
||||
isError: true,
|
||||
content: [
|
||||
{
|
||||
type: 'text',
|
||||
text: `Error calling tool ${tool.name}: ${e instanceof Error ? e.stack || e.message || 'No error details available' : JSON.stringify(e)}`
|
||||
text: `Error calling tool ${toolResponse.tool.name}: ${e instanceof Error ? e.stack || e.message || 'No error details available' : JSON.stringify(e)}`
|
||||
}
|
||||
]
|
||||
})
|
||||
@ -262,7 +314,7 @@ export function mcpToolsToAnthropicTools(mcpTools: MCPTool[]): Array<ToolUnion>
|
||||
const t: ToolUnion = {
|
||||
name: tool.id,
|
||||
description: tool.description,
|
||||
// @ts-ignore no check
|
||||
// @ts-ignore ignore type as it it unknow
|
||||
input_schema: tool.inputSchema
|
||||
}
|
||||
return t
|
||||
@ -275,53 +327,68 @@ export function anthropicToolUseToMcpTool(mcpTools: MCPTool[] | undefined, toolU
|
||||
if (!tool) {
|
||||
return undefined
|
||||
}
|
||||
// @ts-ignore ignore type as it it unknow
|
||||
tool.inputSchema = toolUse.input
|
||||
return tool
|
||||
}
|
||||
|
||||
// export function mcpToolsToGeminiTools(mcpTools: MCPTool[] | undefined): geminiTool[] {
|
||||
// if (!mcpTools || mcpTools.length === 0) {
|
||||
// // No tools available
|
||||
// return []
|
||||
// }
|
||||
// const functions: FunctionDeclaration[] = []
|
||||
|
||||
// for (const tool of mcpTools) {
|
||||
// const properties = filterPropertieAttributes(tool, true)
|
||||
// const functionDeclaration: FunctionDeclaration = {
|
||||
// name: tool.id,
|
||||
// description: tool.description,
|
||||
// parameters: {
|
||||
// type: SchemaType.OBJECT,
|
||||
// properties:
|
||||
// Object.keys(properties).length > 0
|
||||
// ? Object.fromEntries(
|
||||
// Object.entries(properties).map(([key, value]) => [key, ensureValidSchema(value as Record<string, any>)])
|
||||
// )
|
||||
// : { _empty: { type: SchemaType.STRING } as SimpleStringSchema }
|
||||
// } as FunctionDeclarationSchema
|
||||
// }
|
||||
// functions.push(functionDeclaration)
|
||||
// }
|
||||
// const tool: geminiTool = {
|
||||
// functionDeclarations: functions
|
||||
// }
|
||||
// return [tool]
|
||||
// }
|
||||
/**
|
||||
* @param mcpTools
|
||||
* @returns
|
||||
*/
|
||||
export function mcpToolsToGeminiTools(mcpTools: MCPTool[]): Tool[] {
|
||||
/**
|
||||
* @typedef {import('@google/genai').Schema} Schema
|
||||
*/
|
||||
const schemaKeys = [
|
||||
'example',
|
||||
'pattern',
|
||||
'default',
|
||||
'maxLength',
|
||||
'minLength',
|
||||
'minProperties',
|
||||
'maxProperties',
|
||||
'anyOf',
|
||||
'description',
|
||||
'enum',
|
||||
'format',
|
||||
'items',
|
||||
'maxItems',
|
||||
'maximum',
|
||||
'minItems',
|
||||
'minimum',
|
||||
'nullable',
|
||||
'properties',
|
||||
'propertyOrdering',
|
||||
'required',
|
||||
'title',
|
||||
'type'
|
||||
]
|
||||
return [
|
||||
{
|
||||
functionDeclarations: mcpTools?.map((tool) => {
|
||||
return {
|
||||
name: tool.id,
|
||||
description: tool.description,
|
||||
parameters: {
|
||||
type: GeminiSchemaType.OBJECT,
|
||||
properties: filterProperties(tool.inputSchema, schemaKeys).properties,
|
||||
required: tool.inputSchema.required
|
||||
}
|
||||
}
|
||||
})
|
||||
}
|
||||
]
|
||||
}
|
||||
|
||||
export function geminiFunctionCallToMcpTool(
|
||||
mcpTools: MCPTool[] | undefined,
|
||||
fcall: FunctionCall | undefined
|
||||
toolCall: FunctionCall | undefined
|
||||
): MCPTool | undefined {
|
||||
if (!fcall) return undefined
|
||||
if (!toolCall) return undefined
|
||||
if (!mcpTools) return undefined
|
||||
const tool = mcpTools.find((tool) => tool.id === fcall.name)
|
||||
const tool = mcpTools.find((tool) => tool.id === toolCall.name)
|
||||
if (!tool) {
|
||||
return undefined
|
||||
}
|
||||
// @ts-ignore schema is not a valid property
|
||||
tool.inputSchema = fcall.args
|
||||
return tool
|
||||
}
|
||||
|
||||
@ -368,13 +435,13 @@ export function getMcpServerByTool(tool: MCPTool) {
|
||||
return servers.find((s) => s.id === tool.serverId)
|
||||
}
|
||||
|
||||
export function parseToolUse(content: string, mcpTools: MCPTool[]): MCPToolResponse[] {
|
||||
export function parseToolUse(content: string, mcpTools: MCPTool[]): ToolUseResponse[] {
|
||||
if (!content || !mcpTools || mcpTools.length === 0) {
|
||||
return []
|
||||
}
|
||||
const toolUsePattern =
|
||||
/<tool_use>([\s\S]*?)<name>([\s\S]*?)<\/name>([\s\S]*?)<arguments>([\s\S]*?)<\/arguments>([\s\S]*?)<\/tool_use>/g
|
||||
const tools: MCPToolResponse[] = []
|
||||
const tools: ToolUseResponse[] = []
|
||||
let match
|
||||
let idx = 0
|
||||
// Find all tool use blocks
|
||||
@ -401,10 +468,9 @@ export function parseToolUse(content: string, mcpTools: MCPTool[]): MCPToolRespo
|
||||
// Add to tools array
|
||||
tools.push({
|
||||
id: `${toolName}-${idx++}`, // Unique ID for each tool use
|
||||
tool: {
|
||||
...mcpTool,
|
||||
inputSchema: parsedArgs
|
||||
},
|
||||
toolUseId: mcpTool.id,
|
||||
tool: mcpTool,
|
||||
arguments: parsedArgs,
|
||||
status: 'pending'
|
||||
})
|
||||
|
||||
@ -414,36 +480,69 @@ export function parseToolUse(content: string, mcpTools: MCPTool[]): MCPToolRespo
|
||||
return tools
|
||||
}
|
||||
|
||||
export async function parseAndCallTools(
|
||||
content: string,
|
||||
toolResponses: MCPToolResponse[],
|
||||
export async function parseAndCallTools<R>(
|
||||
tools: MCPToolResponse[],
|
||||
allToolResponses: MCPToolResponse[],
|
||||
onChunk: CompletionsParams['onChunk'],
|
||||
idx: number,
|
||||
convertToMessage: (
|
||||
toolCallId: string,
|
||||
resp: MCPCallToolResponse,
|
||||
isVisionModel: boolean
|
||||
) => ChatCompletionMessageParam | MessageParam | Content | OpenAI.Responses.EasyInputMessage,
|
||||
mcpTools?: MCPTool[],
|
||||
isVisionModel: boolean = false
|
||||
): Promise<(ChatCompletionMessageParam | MessageParam | Content | OpenAI.Responses.EasyInputMessage)[]> {
|
||||
const toolResults: (ChatCompletionMessageParam | MessageParam | Content | OpenAI.Responses.EasyInputMessage)[] = []
|
||||
// process tool use
|
||||
const tools = parseToolUse(content, mcpTools || [])
|
||||
if (!tools || tools.length === 0) {
|
||||
convertToMessage: (mcpToolResponse: MCPToolResponse, resp: MCPCallToolResponse, model: Model) => R | undefined,
|
||||
model: Model,
|
||||
mcpTools?: MCPTool[]
|
||||
): Promise<
|
||||
(ChatCompletionMessageParam | MessageParam | Content | OpenAI.Responses.ResponseInputItem | ToolResultBlockParam)[]
|
||||
>
|
||||
|
||||
export async function parseAndCallTools<R>(
|
||||
content: string,
|
||||
allToolResponses: MCPToolResponse[],
|
||||
onChunk: CompletionsParams['onChunk'],
|
||||
convertToMessage: (mcpToolResponse: MCPToolResponse, resp: MCPCallToolResponse, model: Model) => R | undefined,
|
||||
model: Model,
|
||||
mcpTools?: MCPTool[]
|
||||
): Promise<
|
||||
(ChatCompletionMessageParam | MessageParam | Content | OpenAI.Responses.ResponseInputItem | ToolResultBlockParam)[]
|
||||
>
|
||||
|
||||
export async function parseAndCallTools<R>(
|
||||
content: string | MCPToolResponse[],
|
||||
allToolResponses: MCPToolResponse[],
|
||||
onChunk: CompletionsParams['onChunk'],
|
||||
convertToMessage: (mcpToolResponse: MCPToolResponse, resp: MCPCallToolResponse, model: Model) => R | undefined,
|
||||
model: Model,
|
||||
mcpTools?: MCPTool[]
|
||||
): Promise<R[]> {
|
||||
const toolResults: R[] = []
|
||||
let curToolResponses: MCPToolResponse[] = []
|
||||
if (Array.isArray(content)) {
|
||||
curToolResponses = content
|
||||
} else {
|
||||
// process tool use
|
||||
curToolResponses = parseToolUse(content, mcpTools || [])
|
||||
}
|
||||
if (!curToolResponses || curToolResponses.length === 0) {
|
||||
return toolResults
|
||||
}
|
||||
for (let i = 0; i < tools.length; i++) {
|
||||
const tool = tools[i]
|
||||
upsertMCPToolResponse(toolResponses, { id: `${tool.id}-${idx}-${i}`, tool: tool.tool, status: 'invoking' }, onChunk)
|
||||
for (let i = 0; i < curToolResponses.length; i++) {
|
||||
const toolResponse = curToolResponses[i]
|
||||
upsertMCPToolResponse(
|
||||
allToolResponses,
|
||||
{
|
||||
...toolResponse,
|
||||
status: 'invoking'
|
||||
},
|
||||
onChunk
|
||||
)
|
||||
}
|
||||
|
||||
const toolPromises = tools.map(async (tool, i) => {
|
||||
const toolPromises = curToolResponses.map(async (toolResponse) => {
|
||||
const images: string[] = []
|
||||
const toolCallResponse = await callMCPTool(tool.tool)
|
||||
const toolCallResponse = await callMCPTool(toolResponse)
|
||||
upsertMCPToolResponse(
|
||||
toolResponses,
|
||||
{ id: `${tool.id}-${idx}-${i}`, tool: tool.tool, status: 'done', response: toolCallResponse },
|
||||
allToolResponses,
|
||||
{
|
||||
...toolResponse,
|
||||
status: 'done',
|
||||
response: toolCallResponse
|
||||
},
|
||||
onChunk
|
||||
)
|
||||
|
||||
@ -466,15 +565,15 @@ export async function parseAndCallTools(
|
||||
})
|
||||
}
|
||||
|
||||
return convertToMessage(tool.tool.id, toolCallResponse, isVisionModel)
|
||||
return convertToMessage(toolResponse, toolCallResponse, model)
|
||||
})
|
||||
|
||||
toolResults.push(...(await Promise.all(toolPromises)))
|
||||
toolResults.push(...(await Promise.all(toolPromises)).filter((t) => typeof t !== 'undefined'))
|
||||
return toolResults
|
||||
}
|
||||
|
||||
export function mcpToolCallResponseToOpenAICompatibleMessage(
|
||||
toolCallId: string,
|
||||
mcpToolResponse: MCPToolResponse,
|
||||
resp: MCPCallToolResponse,
|
||||
isVisionModel: boolean = false
|
||||
): ChatCompletionMessageParam {
|
||||
@ -488,7 +587,7 @@ export function mcpToolCallResponseToOpenAICompatibleMessage(
|
||||
const content: ChatCompletionContentPart[] = [
|
||||
{
|
||||
type: 'text',
|
||||
text: `Here is the result of tool call ${toolCallId}:`
|
||||
text: `Here is the result of mcp tool use \`${mcpToolResponse.tool.name}\`:`
|
||||
}
|
||||
]
|
||||
|
||||
@ -541,7 +640,7 @@ export function mcpToolCallResponseToOpenAICompatibleMessage(
|
||||
}
|
||||
|
||||
export function mcpToolCallResponseToOpenAIMessage(
|
||||
toolCallId: string,
|
||||
mcpToolResponse: MCPToolResponse,
|
||||
resp: MCPCallToolResponse,
|
||||
isVisionModel: boolean = false
|
||||
): OpenAI.Responses.EasyInputMessage {
|
||||
@ -555,7 +654,7 @@ export function mcpToolCallResponseToOpenAIMessage(
|
||||
const content: OpenAI.Responses.ResponseInputContent[] = [
|
||||
{
|
||||
type: 'input_text',
|
||||
text: `Here is the result of tool call ${toolCallId}:`
|
||||
text: `Here is the result of mcp tool use \`${mcpToolResponse.tool.name}\`:`
|
||||
}
|
||||
]
|
||||
|
||||
@ -597,9 +696,9 @@ export function mcpToolCallResponseToOpenAIMessage(
|
||||
}
|
||||
|
||||
export function mcpToolCallResponseToAnthropicMessage(
|
||||
toolCallId: string,
|
||||
mcpToolResponse: MCPToolResponse,
|
||||
resp: MCPCallToolResponse,
|
||||
isVisionModel: boolean = false
|
||||
model: Model
|
||||
): MessageParam {
|
||||
const message = {
|
||||
role: 'user'
|
||||
@ -610,10 +709,10 @@ export function mcpToolCallResponseToAnthropicMessage(
|
||||
const content: ContentBlockParam[] = [
|
||||
{
|
||||
type: 'text',
|
||||
text: `Here is the result of tool call ${toolCallId}:`
|
||||
text: `Here is the result of mcp tool use \`${mcpToolResponse.tool.name}\`:`
|
||||
}
|
||||
]
|
||||
if (isVisionModel) {
|
||||
if (isVisionModel(model)) {
|
||||
for (const item of resp.content) {
|
||||
switch (item.type) {
|
||||
case 'text':
|
||||
@ -665,7 +764,7 @@ export function mcpToolCallResponseToAnthropicMessage(
|
||||
}
|
||||
|
||||
export function mcpToolCallResponseToGeminiMessage(
|
||||
toolCallId: string,
|
||||
mcpToolResponse: MCPToolResponse,
|
||||
resp: MCPCallToolResponse,
|
||||
isVisionModel: boolean = false
|
||||
): Content {
|
||||
@ -682,7 +781,7 @@ export function mcpToolCallResponseToGeminiMessage(
|
||||
} else {
|
||||
const parts: Part[] = [
|
||||
{
|
||||
text: `Here is the result of tool call ${toolCallId}:`
|
||||
text: `Here is the result of mcp tool use \`${mcpToolResponse.tool.name}\`:`
|
||||
}
|
||||
]
|
||||
if (isVisionModel) {
|
||||
|
||||
@ -147,7 +147,7 @@ ${availableTools}
|
||||
</tools>`
|
||||
}
|
||||
|
||||
export const buildSystemPrompt = (userSystemPrompt: string, tools: MCPTool[]): string => {
|
||||
export const buildSystemPrompt = (userSystemPrompt: string, tools?: MCPTool[]): string => {
|
||||
if (tools && tools.length > 0) {
|
||||
return SYSTEM_PROMPT.replace('{{ USER_SYSTEM_PROMPT }}', userSystemPrompt)
|
||||
.replace('{{ TOOL_USE_EXAMPLES }}', ToolUseExamples)
|
||||
|
||||
Loading…
Reference in New Issue
Block a user