Function handleSiliconFlowStream

web/src/llm-api/siliconflow.ts:192–329

({
  body,
  userId,
  stripeCustomerId,
  agentId,
  fetch,
  logger,
  insertMessageBigquery,
}: {
  body: ChatCompletionRequestBody
  userId: string
  stripeCustomerId?: string | null
  agentId: string
  fetch: typeof globalThis.fetch
  logger: Logger
  insertMessageBigquery: InsertMessageBigqueryFn
})

}

export async function handleSiliconFlowStream({
  body,
  userId,
  stripeCustomerId,
  agentId,
  fetch,
  logger,
  insertMessageBigquery,
}: {
  body: ChatCompletionRequestBody
  userId: string
  stripeCustomerId?: string | null
  agentId: string
  fetch: typeof globalThis.fetch
  logger: Logger
  insertMessageBigquery: InsertMessageBigqueryFn
}) {
  const originalModel = body.model
  const startTime = new Date()
  const { clientId, clientRequestId, costMode } = extractRequestMetadata({ body, logger })
  const auditRequest = createRequestAuditRecord(body)

  const response = await createSiliconFlowRequest({ body, originalModel, fetch })

  if (!response.ok) {
    throw await parseSiliconFlowError(response)
  }

  const reader = response.body?.getReader()
  if (!reader) {
    throw new Error('Failed to get response reader')
  }

  let heartbeatInterval: NodeJS.Timeout
  let state: StreamState = { responseText: '', reasoningText: '', ttftMs: null, billedAlready: false }
  let clientDisconnected = false

  const stream = new ReadableStream({
    async start(controller) {
      const decoder = new TextDecoder()
      let buffer = ''

      controller.enqueue(
        new TextEncoder().encode(`: connected ${new Date().toISOString()}\n`),
      )

      heartbeatInterval = setInterval(() => {
        if (!clientDisconnected) {
          try {
            controller.enqueue(
              new TextEncoder().encode(
                `: heartbeat ${new Date().toISOString()}\n\n`,
              ),
            )
          } catch {
            // client disconnected
          }
        }