import { FileHelper } from '@start9labs/start-sdk'
import { Volume } from '@start9labs/start-sdk/package/lib/util/Volume'
import { z } from 'zod'

const mainVolume = new Volume('main')

// Operator-side configuration for the Recap Relay package. All fields
// are optional and ship with sensible defaults — the relay will boot
// even with an empty config, but will refuse to serve traffic until at
// least relay_gemini_api_key and relay_admin_password_hash are set.
export const configFile = FileHelper.json(
  {
    base: mainVolume,
    subpath: 'config/relay-config.json',
  },
  z.object({
    // ── Backend credentials ──
    // The relay's Gemini API key. Used for all transcribe + analyze
    // forwarding until a user exceeds their tier's Gemini cap (then
    // overflows to operator hardware below). Empty disables the
    // Gemini backend entirely — relay will then either route to
    // hardware (if configured) or 503 every request.
    relay_gemini_api_key: z.string().default(''),

    // ── Operator hardware (optional fallback) ──
    // When a Pro/Max user exceeds their monthly Gemini cap, the relay
    // routes overflow here. Leave empty to hard-cap at the Gemini limit
    // and return 402 once exceeded (no fallback).
    relay_parakeet_base_url: z.string().default(''),
    relay_gemma_base_url: z.string().default(''),
    // Model identifiers to send in the upstream request bodies. The
    // operator's Ollama or Parakeet wrapper may serve different models
    // depending on what's been pulled; making these config-driven
    // means the operator can swap models without rebuilding the relay.
    // Live-reloaded — change applies to the next request.
    relay_parakeet_model: z.string().default('parakeet-tdt-0.6b-v3'),
    relay_gemma_model: z.string().default('gemma3:27b'),

    // ── Gemini model selection ──
    // Operator can pick which Gemini SKU is used per pipeline step
    // without rebuilding the relay. Defaults match Google's typical
    // recommendations: Flash for transcription (cheap, fast,
    // multimodal-capable), Pro for analysis (higher quality on
    // structured-JSON outputs). Operators can swap to flash for
    // analysis when they want faster + cheaper at the cost of some
    // section-boundary precision.
    relay_gemini_transcription_model: z.string().default('gemini-3-flash-preview'),
    relay_gemini_analysis_model: z.string().default('gemini-3.1-pro-preview'),

    // ── Backend routing preference per pipeline ──
    // Controls whether the relay tries Gemini first (current default —
    // best quality, costs operator's Gemini API budget) or the
    // operator-hardware backend first (saves Gemini budget, may be
    // slower depending on the operator's hardware). One of:
    //   - "gemini_first"   try Gemini until per-tier cap, then hardware
    //   - "hardware_first" try hardware first, fall back to Gemini
    //   - "gemini_only"    Gemini only, fail when cap is exceeded
    //   - "hardware_only"  Hardware only, fail when not configured
    relay_transcribe_backend_preference: z
      .enum(['gemini_first', 'hardware_first', 'gemini_only', 'hardware_only'])
      .default('gemini_first'),
    relay_analyze_backend_preference: z
      .enum(['gemini_first', 'hardware_first', 'gemini_only', 'hardware_only'])
      .default('gemini_first'),

    // ── License server ──
    // URL of the Keysat license server used for the cached online
    // license-validation check. Defaults to the public endpoint;
    // operators co-located with Keysat on the same Start9 server can
    // override to the internal `http://keysat.startos:<port>` hostname
    // for a lower-latency hot path.
    relay_keysat_base_url: z.string().default('https://keysat.xyz'),

    // ── Admin dashboard auth ──
    // Username + scrypt-hashed password + session secret for the
    // /admin/* dashboard. Same shape Recap uses (see Recap's
    // server/admin-auth.js for the hash + verify code). Empty hash
    // disables /admin entirely — useful while testing the public
    // /relay/* endpoints.
    relay_admin_username: z.string().default(''),
    relay_admin_password_hash: z.string().default(''),
    relay_admin_password_salt: z.string().default(''),
    relay_admin_session_secret: z.string().default(''),

    // ── Tier quotas (operator-adjustable without redeploy) ──
    // JSON blob driving credits.js. Defaults match the v1 product
    // spec: Core lifetime-5, Pro 50/mo with 25 Gemini cap, Max
    // unlimited with 50 Gemini cap. Operators can tweak via the
    // "Adjust Tier Quotas" action without a code change or restart.
    relay_tier_quotas_json: z.string().default(
      JSON.stringify({
        // Core: 10 lifetime credits total — first 5 served via Gemini
        // (operator's cloud spend), final 5 fall through to operator
        // hardware so the user can keep going on free tier without
        // costing the operator more cloud $.
        core: {
          lifetime: 10,
          geminiCapLifetime: 5,
          monthly: null,
          geminiCapMonthly: null,
        },
        pro: { lifetime: null, monthly: 50, geminiCapMonthly: 25 },
        max: { lifetime: null, monthly: null, geminiCapMonthly: 50 },
      }),
    ),
  }),
)