feat(coach): switch Lyra to Gemini 2.5 Flash Lite (Groq+OpenRouter quotas dead)

- Primary: gemini-2.5-flash-lite (~789ms TTFR, ~10x cheaper than Haiku, no reasoning overhead)
- Fallback 1: gemini-2.5-flash (smarter when Lite overloaded)
- Fallback 2: gpt-4o-mini (anchor on different provider)
- message.post.ts: candidates chain replaced
- sos-stream.get.ts: gemini-flash-lite default + auto-fallback to gpt-4o-mini if key missing
- nitro.config.ts: geminiApiKey runtimeConfig
- start-staging.sh: GEMINI_API_KEY export + NITRO_GEMINI_API_KEY

OpenRouter credits = 0, Groq TPD exhausted - users get 503 currently.

backend/nitro.config.ts

@@ -40,6 +40,7 @@ export default defineNitroConfig({
     openaiApiKey: process.env.OPENAI_API_KEY ?? process.env.NUXT_OPENAI_API_KEY ?? "",
     groqApiKey: process.env.GROQ_API_KEY ?? process.env.NUXT_GROQ_API_KEY ?? "",
     googleAiApiKey: process.env.GOOGLE_AI_API_KEY ?? "",
+    geminiApiKey: process.env.GEMINI_API_KEY ?? "",
 
     // ─── TTS-Provider ────────────────────────────────────────────────────
     googleApiKey: process.env.GOOGLE_API_KEY ?? process.env.NUXT_GOOGLE_API_KEY ?? "",

backend/server/api/coach/message.post.ts

@@ -361,6 +361,14 @@ const PROVIDER_CONFIG = {
     url: "https://openrouter.ai/api/v1/chat/completions",
     keyName: "openrouterApiKey" as const,
   },
+  gemini: {
+    url: "https://generativelanguage.googleapis.com/v1beta/openai/chat/completions",
+    keyName: "geminiApiKey" as const,
+  },
+  openai: {
+    url: "https://api.openai.com/v1/chat/completions",
+    keyName: "openaiApiKey" as const,
+  },
 } as const;
 
 const FEEDBACK_DETECTION_PROMPT = `Du analysierst eine Nutzer-Nachricht aus einer Gambling-Recovery-App.
@@ -595,28 +603,21 @@ export default defineEventHandler(async (event) => {
   }
 
   // ─── Tier-basiertes LLM-Routing (analog sos-stream.get.ts) ─────────────────
-  // Free / Pro  → Groq Llama 3.3 70B (schnell, sachlich)
+  // Default-Chain: Gemini Flash Lite (schnell+billig+warm) → Gemini 2.5 Flash
-  // Legend      → OpenRouter Haiku 4.5 (warm, premium)
+  // (smarter Fallback) → OpenAI gpt-4o-mini (Last-Resort, anderer Provider).
-  // Kein sosMode-Override mehr — Coach-Page hat eigenes Routing.
+  // OpenRouter + Groq sind aktuell ohne Quota/Credits — entfernt aus Chain.
   const planRaw = (profile?.plan ?? "free").toLowerCase();
   const plan = planRaw === "premium" ? "legend" : planRaw === "standard" ? "pro" : planRaw;
-  const llmProvider = plan === "legend" ? "openrouter-haiku" : "groq-llama";
+  const llmProvider = "gemini-flash-lite";
 
-  type Candidate = { provider: "groq" | "openrouter"; model: string };
+  type Candidate = { provider: "groq" | "openrouter" | "gemini" | "openai"; model: string };
-  const candidates: Candidate[] =
+  const candidates: Candidate[] = [
-    llmProvider === "openrouter-haiku"
+    { provider: "gemini", model: "gemini-2.5-flash-lite" },
-      ? [
+    { provider: "gemini", model: "gemini-2.5-flash" },
-          { provider: "openrouter", model: "anthropic/claude-haiku-4.5" },
+    { provider: "openai", model: "gpt-4o-mini" },
-          { provider: "openrouter", model: "anthropic/claude-3.5-haiku" },
+  ];
-          { provider: "groq", model: "llama-3.3-70b-versatile" },
-        ]
-      : [
-          { provider: "groq", model: "llama-3.3-70b-versatile" },
-          { provider: "groq", model: "llama-3.1-8b-instant" },
-          { provider: "openrouter", model: "meta-llama/llama-3.3-70b-instruct" },
-        ];
 
-  async function tryModel(providerName: "groq" | "openrouter", model: string) {
+  async function tryModel(providerName: "groq" | "openrouter" | "gemini" | "openai", model: string) {
     const p = PROVIDER_CONFIG[providerName];
     const key = config[p.keyName];
     if (!key) return null;

backend/server/api/coach/sos-stream.get.ts

@@ -219,17 +219,20 @@ export default defineEventHandler(async (event) => {
   if (userToggle && userToggle !== "auto") {
     llmProvider = userToggle;
   } else {
-    const planRaw = (profile?.plan ?? "free").toLowerCase();
+    // Default chain: Gemini Flash Lite (schnell+billig+warm) für alle Pläne.
-    // legacy "premium"/"standard" → legend/pro
+    // OpenRouter (Haiku/Sonnet) + Groq haben aktuell keine Quota/Credits.
-    const plan = planRaw === "premium" ? "legend" : planRaw === "standard" ? "pro" : planRaw;
+    llmProvider = "gemini-flash-lite";
-    llmProvider = plan === "legend" ? "openrouter-haiku" : "groq-llama";
   }
   let upstreamUrl: string;
   let upstreamKey: string | undefined;
   let upstreamModel: string;
   const upstreamHeaders: Record<string, string> = { "Content-Type": "application/json" };
   let upstreamProviderField: { sort: string } | undefined;
-  if (llmProvider === "groq-llama") {
+  if (llmProvider === "gemini-flash-lite" || llmProvider === "gemini-flash") {
+    upstreamUrl = "https://generativelanguage.googleapis.com/v1beta/openai/chat/completions";
+    upstreamKey = config.geminiApiKey as string | undefined;
+    upstreamModel = llmProvider === "gemini-flash" ? "gemini-2.5-flash" : "gemini-2.5-flash-lite";
+  } else if (llmProvider === "groq-llama") {
     upstreamUrl = "https://api.groq.com/openai/v1/chat/completions";
     upstreamKey = config.groqApiKey as string | undefined;
     upstreamModel = "llama-3.3-70b-versatile";
@@ -243,6 +246,15 @@ export default defineEventHandler(async (event) => {
     upstreamHeaders["X-Title"] = "ReBreak SOS";
     upstreamProviderField = { sort: "latency" };
   }
+  // Fallback: wenn gewählter Provider keinen Key hat → OpenAI gpt-4o-mini als Anker
+  if (!upstreamKey && config.openaiApiKey) {
+    console.warn(`[coach/sos-stream] ${llmProvider} key missing → fallback openai/gpt-4o-mini`);
+    llmProvider = "openai-mini";
+    upstreamUrl = "https://api.openai.com/v1/chat/completions";
+    upstreamKey = config.openaiApiKey as string;
+    upstreamModel = "gpt-4o-mini";
+    upstreamProviderField = undefined;
+  }
   if (!upstreamKey) {
     throw createError({ statusCode: 503, message: `API key for ${llmProvider} fehlt` });
   }

backend/start-staging.sh

@@ -44,6 +44,7 @@ exec infisical run \
     export OPENROUTER_API_KEY="${OPENROUTER_API_KEY:-${NUXT_OPENROUTER_API_KEY:-}}"
     export GROQ_API_KEY="${GROQ_API_KEY:-${NUXT_GROQ_API_KEY:-}}"
     export GOOGLE_API_KEY="${GOOGLE_API_KEY:-${NUXT_GOOGLE_API_KEY:-}}"
+    export GEMINI_API_KEY="${GEMINI_API_KEY:-}"
     export DEEPGRAM_API_KEY="${DEEPGRAM_API_KEY:-${NUXT_DEEPGRAM_API_KEY:-}}"
     export DATABASE_URL="${DATABASE_URL:-${NUXT_DATABASE_URL:-}}"
     export LYRA_BOT_USER_ID="${LYRA_BOT_USER_ID:-${NUXT_LYRA_BOT_USER_ID:-}}"
@@ -59,6 +60,7 @@ exec infisical run \
     [[ -n "${GROQ_API_KEY:-}" ]] && export NITRO_GROQ_API_KEY="$GROQ_API_KEY"
     [[ -n "${GOOGLE_AI_API_KEY:-}" ]] && export NITRO_GOOGLE_AI_API_KEY="$GOOGLE_AI_API_KEY"
     [[ -n "${GOOGLE_API_KEY:-}" ]] && export NITRO_GOOGLE_API_KEY="$GOOGLE_API_KEY"
+    [[ -n "${GEMINI_API_KEY:-}" ]] && export NITRO_GEMINI_API_KEY="$GEMINI_API_KEY"
     [[ -n "${DEEPGRAM_API_KEY:-}" ]] && export NITRO_DEEPGRAM_API_KEY="$DEEPGRAM_API_KEY"
     [[ -n "${CARTESIA_API_KEY:-}" ]] && export NITRO_CARTESIA_API_KEY="$CARTESIA_API_KEY"
     [[ -n "${ELEVENLABS_API_KEY:-}" ]] && export NITRO_ELEVENLABS_API_KEY="$ELEVENLABS_API_KEY"