Function queryModel

src/services/api/claude.ts:1017–2892

(
  messages: Message[],
  systemPrompt: SystemPrompt,
  thinkingConfig: ThinkingConfig,
  tools: Tools,
  signal: AbortSignal,
  options: Options,
)

}

async function* queryModel(
  messages: Message[],
  systemPrompt: SystemPrompt,
  thinkingConfig: ThinkingConfig,
  tools: Tools,
  signal: AbortSignal,
  options: Options,
): AsyncGenerator<
  StreamEvent | AssistantMessage | SystemAPIErrorMessage,
  void
> {
  // Check cheap conditions first — the off-switch await blocks on GrowthBook
  // init (~10ms). For non-Opus models (haiku, sonnet) this skips the await
  // entirely. Subscribers don't hit this path at all.
  if (
    !isClaudeAISubscriber() &&
    isNonCustomOpusModel(options.model) &&
    (
      await getDynamicConfig_BLOCKS_ON_INIT<{ activated: boolean }>(
        'tengu-off-switch',
        {
          activated: false,
        },
      )
    ).activated
  ) {
    logEvent('tengu_off_switch_query', {})
    yield getAssistantMessageFromError(
      new Error(CUSTOM_OFF_SWITCH_MESSAGE),
      options.model,
    )
    return
  }

  // Derive previous request ID from the last assistant message in this query chain.
  // This is scoped per message array (main thread, subagent, teammate each have their own),
  // so concurrent agents don't clobber each other's request chain tracking.
  // Also naturally handles rollback/undo since removed messages won't be in the array.
  const previousRequestId = getPreviousRequestIdFromMessages(messages)

  const resolvedModel =
    getAPIProvider() === 'bedrock' &&
    options.model.includes('application-inference-profile')
      ? ((await getInferenceProfileBackingModel(options.model)) ??
        options.model)
      : options.model

  queryCheckpoint('query_tool_schema_build_start')
  const isAgenticQuery =
    options.querySource.startsWith('repl_main_thread') ||
    options.querySource.startsWith('agent:') ||
    options.querySource === 'sdk' ||
    options.querySource === 'hook_agent' ||
    options.querySource === 'verification_agent'
  const betas = getMergedBetas(options.model, { isAgenticQuery })

  // Always send the advisor beta header when advisor is enabled, so
  // non-agentic queries (compact, side_question, extract_memories, etc.)