recap-relay/server/routes/transcribe.js

// POST /relay/transcribe — forwards an audio payload to the chosen
// backend (Gemini first, operator hardware as overflow) and returns
// the standard envelope.
//
// Request shape: multipart/form-data
//   audio:        binary audio file (required)
//   mime_type:    string (default application/octet-stream)
//   title:        string (optional, used by Gemini prompt)
//   channel:      string (optional)
//   description:  string (optional)
//   chapters:     JSON-stringified array (optional)
//   offset_seconds: number string (optional, for chunked audio)
//
// Headers:
//   X-Recap-Install-Id  (required)
//   X-Recap-Job-Id      (optional but expected — pairs with /analyze)
//   Authorization       (optional Bearer LIC1-... for licensed tiers)
//
// Response (standard envelope):
//   {
//     result: { text: "[MM:SS] ...", segments: [], duration_seconds: 0 },
//     credits_remaining, tier, credit_charged
//   }
//
// Every outcome (success / quota-refused / backend-error) writes one
// row to the audit log so the admin dashboard can compute cost,
// margin, and speed metrics.

import express from "express";
import multer from "multer";
import { resolveLicense } from "../keysat-client.js";
import { getOrCreateRow, planBackend, commitCredit } from "../credits.js";
import { lookupJob, markJobCharged, refundJob } from "../job-credits.js";
import { getConfigSnapshot, getTierQuotas } from "../config.js";
import { createGeminiBackend } from "../backends/gemini.js";
import { createHardwareBackend } from "../backends/hardware.js";
import { envelope, errorEnvelope } from "./envelope.js";
import { recordCall } from "../audit-log.js";
import { calcGeminiCost } from "../pricing.js";

const upload = multer({
  storage: multer.memoryStorage(),
  limits: { fileSize: 200 * 1024 * 1024 }, // 200 MB per request
});

export function transcribeRouter() {
  const router = express.Router();

  router.post("/transcribe", upload.single("audio"), async (req, res) => {
    const t0 = Date.now();
    const installId = req.header("X-Recap-Install-Id");
    const jobId = req.header("X-Recap-Job-Id") || null;
    const auth = req.header("Authorization");

    if (!installId) {
      const e = await errorEnvelope({
        error: "missing X-Recap-Install-Id header",
        statusHint: 400,
      });
      return res.status(400).json(e.body);
    }
    if (!req.file) {
      const e = await errorEnvelope({ error: "missing audio file", installId, statusHint: 400 });
      return res.status(400).json(e.body);
    }

    const license = await resolveLicense(auth);
    const tier = license.tier;

    const row = await getOrCreateRow(installId);
    row.tier_snapshot = tier;

    let reusedJob = false;
    let chosenBackend = null;
    const existingJob = lookupJob(installId, jobId);
    if (existingJob) {
      reusedJob = true;
      chosenBackend = existingJob.backend;
    } else {
      const cfg = await getConfigSnapshot();
      const hasHardware = !!cfg.relay_parakeet_base_url;
      const quota = await getTierQuotas();
      const preference =
        cfg.relay_transcribe_backend_preference || "gemini_first";
      const plan = planBackend(row, quota, { hasHardware, preference });
      if (!plan.allowed) {
        await recordCall({
          install_id: installId,
          tier,
          pipeline: "transcribe",
          backend: null,
          model: null,
          status: "refused",
          credit_charged: 0,
          duration_ms: Date.now() - t0,
          cost_usd: 0,
          job_id: jobId,
          error: plan.reason,
        });
        const e = await errorEnvelope({
          error: plan.reason,
          installId,
          tier,
          statusHint: 402,
        });
        return res.status(402).json(e.body);
      }
      chosenBackend = plan.backend;
    }

    const cfg = await getConfigSnapshot();
    let result;
    try {
      if (chosenBackend === "gemini") {
        const backend = createGeminiBackend({
          apiKey: cfg.relay_gemini_api_key,
          transcriptionModel: cfg.relay_gemini_transcription_model,
          analysisModel: cfg.relay_gemini_analysis_model,
        });
        result = await backend.transcribeAudio({
          audio: req.file.buffer,
          mimeType: req.body?.mime_type || req.file.mimetype || "application/octet-stream",
          title: req.body?.title || "",
          channel: req.body?.channel || "",
          description: req.body?.description || "",
          chapters: parseChaptersField(req.body?.chapters),
          offsetSeconds: Number(req.body?.offset_seconds) || 0,
        });
      } else {
        const backend = createHardwareBackend({
          parakeetBaseURL: cfg.relay_parakeet_base_url,
          gemmaBaseURL: cfg.relay_gemma_base_url,
          parakeetModel: cfg.relay_parakeet_model,
          gemmaModel: cfg.relay_gemma_model,
        });
        result = await backend.transcribeAudio({
          audio: req.file.buffer,
          mimeType: req.body?.mime_type || req.file.mimetype || "application/octet-stream",
          offsetSeconds: Number(req.body?.offset_seconds) || 0,
        });
      }
    } catch (err) {
      if (reusedJob) refundJob(installId, jobId);
      console.error(`[relay/transcribe] backend error: ${err?.message}`);
      await recordCall({
        install_id: installId,
        tier,
        pipeline: "transcribe",
        backend: chosenBackend,
        model: chosenBackend === "gemini"
          ? cfg.relay_gemini_transcription_model
          : cfg.relay_parakeet_model,
        status: "error",
        credit_charged: 0,
        duration_ms: Date.now() - t0,
        cost_usd: 0,
        job_id: jobId,
        error: (err?.message || String(err)).slice(0, 200),
      });
      const e = await errorEnvelope({
        error: err?.message || "backend_error",
        installId,
        tier,
        statusHint: err?.status || 502,
      });
      return res.status(e.statusHint).json(e.body);
    }

    let creditCharged = 0;
    if (!reusedJob) {
      await commitCredit(installId, { backend: chosenBackend, tier });
      markJobCharged(installId, jobId, { backend: chosenBackend, tier });
      creditCharged = 1;
    }

    // Success — write the audit row with cost details. Gemini's usage
    // metadata gives us token counts; calcGeminiCost translates that
    // into USD. Hardware-served calls have no token data and we
    // report cost_usd: 0 (operator's hardware is fixed-cost).
    const costDetails =
      chosenBackend === "gemini" && result.usage
        ? calcGeminiCost(result.model, result.usage)
        : {
            input_tokens: 0,
            output_tokens: 0,
            thinking_tokens: 0,
            cost_usd: 0,
          };
    await recordCall({
      install_id: installId,
      tier,
      pipeline: "transcribe",
      backend: chosenBackend,
      model: result?.model || null,
      status: "success",
      credit_charged: creditCharged,
      duration_ms: Date.now() - t0,
      job_id: jobId,
      ...costDetails,
    });

    const body = await envelope({ result, installId, tier, creditCharged });
    res.json(body);
  });

  return router;
}

function parseChaptersField(raw) {
  if (!raw) return [];
  try {
    const arr = JSON.parse(raw);
    return Array.isArray(arr) ? arr : [];
  } catch {
    return [];
  }
}