claude-code-best/src/services/api/openai/index.ts

import type { BetaToolUnion } from '@anthropic-ai/sdk/resources/beta/messages/messages.mjs'
import type { SystemPrompt } from '../../../utils/systemPromptType.js'
import type {
  Message,
  StreamEvent,
  SystemAPIErrorMessage,
  AssistantMessage,
} from '../../../types/message.js'
import type { Tools } from '../../../Tool.js'
import { getOpenAIClient } from './client.js'
import { anthropicMessagesToOpenAI } from './convertMessages.js'
import {
  anthropicToolsToOpenAI,
  anthropicToolChoiceToOpenAI,
} from './convertTools.js'
import { adaptOpenAIStreamToAnthropic } from './streamAdapter.js'
import { resolveOpenAIModel } from './modelMapping.js'
import { normalizeMessagesForAPI } from '../../../utils/messages.js'
import { toolToAPISchema } from '../../../utils/api.js'
import {
  getEmptyToolPermissionContext,
  toolMatchesName,
} from '../../../Tool.js'
import { logForDebugging } from '../../../utils/debug.js'
import { addToTotalSessionCost } from '../../../cost-tracker.js'
import { calculateUSDCost } from '../../../utils/modelCost.js'
import { isEnvTruthy, isEnvDefinedFalsy } from '../../../utils/envUtils.js'
import { getModelMaxOutputTokens } from '../../../utils/context.js'
import type { Options } from '../claude.js'
import { randomUUID } from 'crypto'
import {
  createAssistantAPIErrorMessage,
  normalizeContentFromAPI,
} from '../../../utils/messages.js'
import {
  isToolSearchEnabled,
  extractDiscoveredToolNames,
} from '../../../utils/toolSearch.js'
import {
  isDeferredTool,
  TOOL_SEARCH_TOOL_NAME,
} from '../../../tools/ToolSearchTool/prompt.js'

/**
 * Detect whether DeepSeek-style thinking mode should be enabled.
 *
 * Enabled when:
 * 1. OPENAI_ENABLE_THINKING=1 is set (explicit enable), OR
 * 2. Model name contains "deepseek-reasoner" OR "DeepSeek-V3.2" (auto-detect, case-insensitive)
 *
 * Disabled when:
 * - OPENAI_ENABLE_THINKING=0/false/no/off is explicitly set (overrides model detection)
 *
 * @param model - The resolved OpenAI model name
 * @internal Exported for testing purposes only
 */
export function isOpenAIThinkingEnabled(model: string): boolean {
  // Explicit disable takes priority (overrides model auto-detect)
  if (isEnvDefinedFalsy(process.env.OPENAI_ENABLE_THINKING)) return false
  // Explicit enable
  if (isEnvTruthy(process.env.OPENAI_ENABLE_THINKING)) return true
  // Auto-detect from model name (deepseek-reasoner and DeepSeek-V3.2 support thinking mode)
  const modelLower = model.toLowerCase()
  return modelLower.includes('deepseek-reasoner') || modelLower.includes('deepseek-v3.2')
}

/**
 * Build the request body for OpenAI chat.completions.create().
 * Extracted for testability — the thinking mode params are injected here.
 *
 * DeepSeek thinking mode: inject thinking params via request body.
 * Two formats are added simultaneously to support different deployments:
 * - Official DeepSeek API: `thinking: { type: 'enabled' }`
 * - Self-hosted DeepSeek-V3.2: `enable_thinking: true` + `chat_template_kwargs: { thinking: true }`
 * OpenAI SDK passes unknown keys through to the HTTP body.
 * Each endpoint will use the format it recognizes and ignore the others.
 * @internal Exported for testing purposes only
 */
export function buildOpenAIRequestBody(params: {
  model: string
  messages: any[]
  tools: any[]
  toolChoice: any
  enableThinking: boolean
  maxTokens: number
  temperatureOverride?: number
}): Record<string, any> {
  const { model, messages, tools, toolChoice, enableThinking, maxTokens, temperatureOverride } = params
  return {
    model,
    messages,
    max_tokens: maxTokens,
    ...(tools.length > 0 && {
      tools,
      ...(toolChoice && { tool_choice: toolChoice }),
    }),
    stream: true,
    stream_options: { include_usage: true },
    // DeepSeek thinking mode: enable chain-of-thought output.
    // When active, temperature/top_p/presence_penalty/frequency_penalty are ignored by DeepSeek.
    ...(enableThinking && {
      // Official DeepSeek API format
      thinking: { type: 'enabled' },
      // Self-hosted DeepSeek-V3.2 format
      enable_thinking: true,
      chat_template_kwargs: { thinking: true },
    }),
    // Only send temperature when thinking mode is off (DeepSeek ignores it anyway,
    // but other providers may respect it)
    ...(!enableThinking && temperatureOverride !== undefined && {
      temperature: temperatureOverride,
    }),
  }
}

/**
 * Assemble the final AssistantMessage (and optional max_tokens error) from
 * accumulated stream state. Extracted to avoid duplication between the
 * `message_stop` handler and the post-loop safety fallback.
 */
function assembleFinalAssistantOutputs(params: {
  partialMessage: any
  contentBlocks: Record<number, any>
  tools: Tools
  agentId: string | undefined
  usage: { input_tokens: number; output_tokens: number; cache_creation_input_tokens: number; cache_read_input_tokens: number }
  stopReason: string | null
  maxTokens: number
}): (AssistantMessage | SystemAPIErrorMessage)[] {
  const { partialMessage, contentBlocks, tools, agentId, usage, stopReason, maxTokens } = params
  const outputs: (AssistantMessage | SystemAPIErrorMessage)[] = []

  const allBlocks = Object.keys(contentBlocks)
    .sort((a, b) => Number(a) - Number(b))
    .map(k => contentBlocks[Number(k)])
    .filter(Boolean)

  if (allBlocks.length > 0) {
    outputs.push({
      message: {
        ...partialMessage,
        content: normalizeContentFromAPI(allBlocks, tools, agentId),
        usage,
        stop_reason: stopReason,
        stop_sequence: null,
      },
      requestId: undefined,
      type: 'assistant',
      uuid: randomUUID(),
      timestamp: new Date().toISOString(),
    } as AssistantMessage)
  }

  if (stopReason === 'max_tokens') {
    outputs.push(createAssistantAPIErrorMessage({
      content: `Output truncated: response exceeded the ${maxTokens} token limit. ` +
        `Set CLAUDE_CODE_MAX_OUTPUT_TOKENS to override.`,
      apiError: 'max_output_tokens',
      error: 'max_output_tokens',
    }))
  }

  return outputs
}

/**
 * OpenAI-compatible query path. Converts Anthropic-format messages/tools to
 * OpenAI format, calls the OpenAI-compatible endpoint, and converts the
 * SSE stream back to Anthropic BetaRawMessageStreamEvent for consumption
 * by the existing query pipeline.
 */
export async function* queryModelOpenAI(
  messages: Message[],
  systemPrompt: SystemPrompt,
  tools: Tools,
  signal: AbortSignal,
  options: Options,
): AsyncGenerator<
  StreamEvent | AssistantMessage | SystemAPIErrorMessage,
  void
> {
  try {
    // 1. Resolve model name
    const openaiModel = resolveOpenAIModel(options.model)

    // 2. Normalize messages using shared preprocessing
    const messagesForAPI = normalizeMessagesForAPI(messages, tools)

    // 3. Check if tool search is enabled (similar to Anthropic path)
    const useToolSearch = await isToolSearchEnabled(
      options.model,
      tools,
      options.getToolPermissionContext ||
        (async () => getEmptyToolPermissionContext()),
      options.agents || [],
      options.querySource,
    )

    // 4. Build deferred tools set (similar to Anthropic path)
    const deferredToolNames = new Set<string>()
    if (useToolSearch) {
      for (const t of tools) {
        if (isDeferredTool(t)) deferredToolNames.add(t.name)
      }
    }

    // 5. Filter tools (similar to Anthropic path)
    let filteredTools = tools
    if (useToolSearch && deferredToolNames.size > 0) {
      const discoveredToolNames = extractDiscoveredToolNames(messages)

      filteredTools = tools.filter(tool => {
        // Always include non-deferred tools
        if (!deferredToolNames.has(tool.name)) return true
        // Always include ToolSearchTool (so it can discover more tools)
        if (toolMatchesName(tool, TOOL_SEARCH_TOOL_NAME)) return true
        // Only include deferred tools that have been discovered
        return discoveredToolNames.has(tool.name)
      })
    }

    // 6. Build tool schemas with deferLoading flag
    const toolSchemas = await Promise.all(
      filteredTools.map(tool =>
        toolToAPISchema(tool, {
          getToolPermissionContext: options.getToolPermissionContext,
          tools,
          agents: options.agents,
          allowedAgentTypes: options.allowedAgentTypes,
          model: options.model,
          deferLoading: useToolSearch && deferredToolNames.has(tool.name),
        }),
      ),
    )

    // 7. Filter out non-standard tools (server tools like advisor)
    const standardTools = toolSchemas.filter(
      (t): t is BetaToolUnion & { type: string } => {
        const anyT = t as Record<string, unknown>
        return (
          anyT.type !== 'advisor_20260301' && anyT.type !== 'computer_20250124'
        )
      },
    )

    // 8. Convert messages and tools to OpenAI format
    const enableThinking = isOpenAIThinkingEnabled(openaiModel)
    const openaiMessages = anthropicMessagesToOpenAI(messagesForAPI, systemPrompt, {
      enableThinking,
    })
    const openaiTools = anthropicToolsToOpenAI(standardTools)
    const openaiToolChoice = anthropicToolChoiceToOpenAI(options.toolChoice)

    // 9. Log tool filtering details
    if (useToolSearch) {
      const includedDeferredTools = filteredTools.filter(t =>
        deferredToolNames.has(t.name),
      ).length
      logForDebugging(
        `[OpenAI] Tool search enabled: ${includedDeferredTools}/${deferredToolNames.size} deferred tools included, total tools=${openaiTools.length}`,
      )
    } else {
      logForDebugging(
        `[OpenAI] Tool search disabled, total tools=${openaiTools.length}`,
      )
    }

    // 10. Compute max_tokens — required by most OpenAI-compatible endpoints.
    //     Without this the server uses a tiny default, and when
    //     thinking is enabled the thinking phase consumes the entire budget
    //     leaving no tokens for the final response.
    //
    //     Use upperLimit (not the slot-cap default) because the Anthropic path's
    //     slot-reservation cap (CAPPED_DEFAULT_MAX_TOKENS=8k) is paired with an
    //     auto-retry at 64k in query.ts. The OpenAI path has no such retry, so
    //     using the capped 8k default would silently truncate responses in
    //     multi-turn conversations where thinking consumes most of the budget.
    const { upperLimit } = getModelMaxOutputTokens(openaiModel)
    const maxTokens = options.maxOutputTokensOverride ?? upperLimit

    // 11. Get client
    const client = getOpenAIClient({
      maxRetries: 0,
      fetchOverride: options.fetchOverride,
      source: options.querySource,
    })

    logForDebugging(
      `[OpenAI] Calling model=${openaiModel}, messages=${openaiMessages.length}, tools=${openaiTools.length}, thinking=${enableThinking}`,
    )

    // 12. Call OpenAI API with streaming
    const requestBody = buildOpenAIRequestBody({
      model: openaiModel,
      messages: openaiMessages,
      tools: openaiTools,
      toolChoice: openaiToolChoice,
      enableThinking,
      maxTokens,
      temperatureOverride: options.temperatureOverride,
    })
    const stream = await client.chat.completions.create(
      requestBody,
      { signal },
    )

    // 12. Convert OpenAI stream to Anthropic events, then process into
    //     AssistantMessage + StreamEvent (matching the Anthropic path behavior)
    const adaptedStream = adaptOpenAIStreamToAnthropic(stream, openaiModel)

    // Accumulate content blocks and usage, same as the Anthropic path in claude.ts
    const contentBlocks: Record<number, any> = {}
    let partialMessage: any
    let stopReason: string | null = null
    let usage = {
      input_tokens: 0,
      output_tokens: 0,
      cache_creation_input_tokens: 0,
      cache_read_input_tokens: 0,
    }
    let ttftMs = 0
    const start = Date.now()

    for await (const event of adaptedStream) {
      switch (event.type) {
        case 'message_start': {
          partialMessage = (event as any).message
          ttftMs = Date.now() - start
          if ((event as any).message?.usage) {
            usage = {
              ...usage,
              ...(event as any).message.usage,
            }
          }
          break
        }
        case 'content_block_start': {
          const idx = (event as any).index
          const cb = (event as any).content_block
          if (cb.type === 'tool_use') {
            contentBlocks[idx] = { ...cb, input: '' }
          } else if (cb.type === 'text') {
            contentBlocks[idx] = { ...cb, text: '' }
          } else if (cb.type === 'thinking') {
            contentBlocks[idx] = { ...cb, thinking: '', signature: '' }
          } else {
            contentBlocks[idx] = { ...cb }
          }
          break
        }
        case 'content_block_delta': {
          const idx = (event as any).index
          const delta = (event as any).delta
          const block = contentBlocks[idx]
          if (!block) break
          if (delta.type === 'text_delta') {
            block.text = (block.text || '') + delta.text
          } else if (delta.type === 'input_json_delta') {
            block.input = (block.input || '') + delta.partial_json
          } else if (delta.type === 'thinking_delta') {
            block.thinking = (block.thinking || '') + delta.thinking
          } else if (delta.type === 'signature_delta') {
            block.signature = delta.signature
          }
          break
        }
        case 'content_block_stop': {
          // Block accumulation is complete; assembly happens at message_stop.
          break
        }
        case 'message_delta': {
          const deltaUsage = (event as any).usage
          if (deltaUsage) {
            usage = { ...usage, ...deltaUsage }
          }
          if ((event as any).delta?.stop_reason != null) {
            stopReason = (event as any).delta.stop_reason
          }
          break
        }
        case 'message_stop': {
          // Assemble ONE AssistantMessage with ALL content blocks, matching the
          // Anthropic SDK path. Real usage (input + output tokens) is available
          // here and injected so tokenCountWithEstimation() can read it.
          if (partialMessage) {
            for (const output of assembleFinalAssistantOutputs({
              partialMessage, contentBlocks, tools, agentId: options.agentId,
              usage, stopReason, maxTokens,
            })) {
              yield output
            }
            // Reset partialMessage so the post-loop safety fallback does not
            // yield a second identical AssistantMessage.
            partialMessage = null
          }
          // Track cost and token usage
          if (usage.input_tokens + usage.output_tokens > 0) {
            const costUSD = calculateUSDCost(openaiModel, usage as any)
            addToTotalSessionCost(costUSD, usage as any, options.model)
          }
          break
        }
      }

      // Also yield as StreamEvent for real-time display (matching Anthropic path)
      yield {
        type: 'stream_event',
        event,
        ...(event.type === 'message_start' ? { ttftMs } : undefined),
      } as StreamEvent
    }

    // Safety: if stream ended without message_stop, assemble and yield whatever we have
    if (partialMessage) {
      for (const output of assembleFinalAssistantOutputs({
        partialMessage, contentBlocks, tools, agentId: options.agentId,
        usage, stopReason, maxTokens,
      })) {
        yield output
      }
    }
  } catch (error) {
    const errorMessage = error instanceof Error ? error.message : String(error)
    logForDebugging(`[OpenAI] Error: ${errorMessage}`, { level: 'error' })
    yield createAssistantAPIErrorMessage({
      content: `API Error: ${errorMessage}`,
      apiError: 'api_error',
      error: error instanceof Error ? error : new Error(String(error)),
    })
  }
}