26 Commits

Author SHA1 Message Date
chahinebrini
55e3cdfb26 fix(transcribe): pass language=ar/tr to nova-2-general so Lyra answers in correct language
Without explicit language param, nova-2-general falls back to multilingual
auto-detect and often misdetects arabic audio as english (phonetic transcript
'salam alaikum' instead of 'السلام عليكم'). detectLang() then sees only
latin chars and answers in english.

Confirmed via Deepgram docs: nova-2-general accepts language=ar and language=tr
(only nova-3 rejects them with HTTP 400).
2026-05-31 01:37:11 +02:00
chahinebrini
3a4e1ecfba 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.
2026-05-31 01:07:10 +02:00
chahinebrini
487af4ede1 fix(coach): duplicate lastUserMsg declaration in message.post 2026-05-31 00:15:28 +02:00
chahinebrini
685782b538 fix(coach): dynamische Sprache (Text-Detection + App-Locale-Fallback)
LLM-Prompt (message.post + sos-stream):
- LANG_INSTRUCTIONS Map raus, ersetzt durch dynamische Instruktion
  'Reply in {detectedFromUser} ... fallback: {appLang}'
- Lyra matcht jetzt die Sprache der letzten User-Message (per
  detectLang Unicode-Detection); App-Locale ist nur noch Fallback
- Instruktion doppelt eingehängt (Anfang + Ende des System-Prompts)
  gegen recency bias bei langen deutschen Prompts

TTS (speak dispatcher + speak-cartesia + speak-elevenlabs):
- Kein 'de'-Default mehr für language. detectLang(text, locale) leitet
  Sprache primär aus dem Antwort-Text ab (Arabic/Cyrillic/CJK/Turkish-
  Letters), Locale als Fallback
- Cartesia + ElevenLabs: language/language_code nur senden wenn
  ableitbar, sonst Provider auto-detect statt erzwungenem 'de'
- speak-cartesia: sonic-2 → sonic-3 (Multi-Lang, war beim Dispatcher-
  Fix gestern vergessen worden)
- Google: en-US neutraler Fallback statt de-DE-Bias

Neu: server/utils/detect-lang.ts
2026-05-31 00:12:40 +02:00
chahinebrini
d31d5b3b83 fix(coach): use nova-2-general for ar/tr STT to avoid 400 errors 2026-05-30 02:14:16 +02:00
chahinebrini
f9d44a6754 feat(coach): mehrsprachiges TTS — locale durchreichen + Cartesia Sonic-3
TTS-Sprache war provider-übergreifend hart auf "de" verdrahtet, locale aus dem
Request wurde ignoriert → arabischer Text wurde deutsch-phonetisch gesprochen.

- locale aus Body auslesen → Basis-Sprachcode an alle Provider
- Pro: Cartesia sonic-2 → sonic-3 (sonic-2 kann kein Arabisch; sonic-3 = 42 Sprachen)
- Legend: ElevenLabs language_code gesetzt (turbo_v2_5 multilingual, ar dabei)
- Google-Fallback: BCP-47-Map (ar→ar-XA etc.), de-Voice nur noch für de

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-05-29 23:57:12 +02:00
chahinebrini
b0315fd177 feat(coach): Lyra-Prompt-Update (Pricing/Beta/Geräte-Limits) + fr Sprach-Instruktion
- Prompt-Rewrite via Copilot: 2-Tier-Pricing (kein Free), Beta-Phase,
  Geräte-Limits, Mail-IDLE, RebReakBinder, Pricing-Disziplin (kein Proaktiv-Pitch)
- fr zu LANG_INSTRUCTIONS (message + sos-stream) — französische App-User
  bekamen sonst deutsche Lyra-Antworten

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-05-29 23:46:25 +02:00
chahinebrini
4f788e640e fix(coach): STT auf Deepgram nova-3 — fixt Arabisch/Türkisch-Transcription
nova-2 unterstützt kein ar/tr → Deepgram 400 "No such model/language/tier
combination" → leeres Transcript ("kein Text nach Speech"). nova-3 deckt alle
gelisteten Sprachen als diskrete Codes ab (de/en/tr/ar/fr/es/pt/it), ohne
Regression. Verifiziert gg. Deepgram models-languages-overview.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-05-29 23:45:50 +02:00
chahinebrini
1ae86c03f4 feat(lyra): coach system prompt v3 — Lock-Modus Setup-Steps (Safari + AirDrop)
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-26 01:22:25 +02:00
chahinebrini
4c31be94b1 feat(lyra): coach system prompt — Self-Bind-Detection + Raynis-Branding + MDM-Korrektur
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-26 01:15:37 +02:00
chahinebrini
dc841b4275 feat(coach): Lyra kennt die iOS-Schutz-Architektur (Layer 1/2 + VIP)
Coach-Prompt (SOS + Casual) um das Zwei-Schichten-Wissen ergänzt:
Layer 1 = URL-Filter (~330k Domains), Layer 2 = VIP-Liste (Zweitschutz,
50/Land + Custom-Domains). Stale NEFilterDataProvider-Erwähnungen korrigiert.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-22 21:09:00 +02:00
chahinebrini
704958320b refactor(domains): gemeinsamer 10/20-Slot-Pool, Free-Tier entfernt
Custom-Domain-Slots sind jetzt EIN gemeinsamer Pool für web + mail
(Pro 10 / Legend 20) statt getrennter web/mail-Buckets. Free-Tier ist
entfallen — PLAN_LIMITS hat nur noch pro + legend, getPlanLimits
defaultet auf pro.

Backend:
- plan-features: customDomains ist eine Zahl (CustomDomainLimits weg)
- index.post: Slot-Check gegen Gesamt-Count, Fehler einheitlich LIMIT_REACHED
- index.get: liefert { items, count, limit }
- change-preview + coach/message an die neue Form angepasst

Frontend:
- useCustomDomains: count/limit (Zahlen) statt countsByType/limits
- AddDomainSheet: ein generischer Limit-Hinweis (error_limit_reached)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-22 18:40:28 +02:00
chahinebrini
f2b81eef54 feat(backend/plan): separate web/mail slot pools + display-name submit lock
plan-features.customDomains is now { web, mail } per plan instead of a
single number. Free 5+5, Pro 5+5, Legend 10+10 — the user explicitly
chose separate pools so users don't have to trade a website slot for a
mail-pattern slot or vice versa.

- countActiveCustomDomainsSplit(userId) groupBy type → { web, mail }
  (mail aggregates mail_domain + mail_display_name). Old single-count
  function stays as a deprecated alias for any caller still on it.
- POST /api/custom-domains: body-compat accepts both { pattern, kind }
  (current frontend) and { domain, type } (legacy / direct). kind='mail'
  is split into mail_domain vs mail_display_name server-side based on
  whether the pattern looks like a domain. Slot check is per-bucket;
  errors are WEB_LIMIT_REACHED / MAIL_LIMIT_REACHED so the UI can show
  the right limit-reached message per tab.
- GET /api/custom-domains: response shape extended to
  { items, counts: { web, mail }, limits: { web, mail } } so the
  frontend can drive the per-tab counter without client-side estimation.
- POST /api/custom-domains/:id/submit: hard-blocks mail_display_name
  with 400 DISPLAY_NAME_NOT_SUBMITTABLE. Display-name submission to the
  global blocklist is deferred to v1.1 — would require a schema split
  on BlocklistDomain that's risky pre-TestFlight. mail_domain still
  flows through the community-vote pipeline like web entries.
- auth/me.get.ts, plan/change-preview.get.ts, coach/message.post.ts
  updated for the new shape (Lyra prompts untouched, only template
  variables split web vs mail counts).

24 vitest cases in backend/tests/custom-domains/plan-limits.test.ts
cover the new shape, body compat, bucket logic, and the submit guard;
216/216 total backend tests pass.
2026-05-16 02:03:26 +02:00
chahinebrini
76f8595a4f feat(backend): lyra voice picker for legend (mvp)
- profile.ts: Whitelist (null | iFSsEDGbm0FiEd2IVH4w | Gt7OshJCH7MuzX96wFHi) + setLyraVoiceId()
- profile/me/lyra-voice.patch.ts: neuer Endpoint, Legend-Gate (403 legend_only),
  Validation gegen Whitelist (400 invalid_voice_id). DB-Wert bleibt bei Plan-Downgrade.
- coach/speak.post.ts: ElevenLabs-Voice-Prioritätskette nimmt userLyraVoiceId
  zuerst (nur wenn plan === legend), sonst voiceCfg / config / env / FALLBACK.
- auth/me.get.ts: lyraVoiceId in der Profile-Response damit Frontend hydriert.

Schema-Feld lyraVoiceId existiert bereits aus migration
20260507_profile_demographics_and_trial — keine neue Migration nötig.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-14 22:15:42 +02:00
chahinebrini
e5c9fadd1d fix(speak): explicit imports for voice-quota helpers
Nitro auto-import doesn't reliably pick up named exports from
db/voiceQuota.ts at runtime — speak endpoint threw 500 with
"ReferenceError: getRemainingVoiceQuota is not defined".

Explicit imports for getRemainingVoiceQuota, consumeVoiceQuota,
estimateAudioSeconds + getPlanLimits.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-09 18:11:42 +02:00
chahinebrini
30ed4191b6 fix(coach): markdown-strip safety-net for LLM responses
LLMs (especially Haiku) keep emitting markdown despite explicit "no markdown"
prompt rule. Mobile app has no markdown renderer — users see raw asterisks.

- New stripMarkdown() util handles **bold**, bullet-lists, headings,
  code-fences, links, blockquotes
- /api/coach/message: applies stripMarkdown(text) post-LLM as safety-net

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-09 17:58:22 +02:00
chahinebrini
b40b8465b9 feat(lyra,voice): founder-story + voice-tier-mapping + quota system
Two features in one push (both backend, deploy together):

LYRA FOUNDER-STORY (per strategist Option C — mixed/medium-detail):
- COACH_CASUAL_SYSTEM_PROMPT: GRÜNDER-STORY sub-block
  - Sharing-rules: ALWAYS on direct ask, RARELY proactive (only on
    explicit isolation expressions "niemand versteht das"), NEVER in
    SOS-mode, NEVER first-3-msgs, NEVER if user appears minor
  - Detail-level: "aus persönlicher Erfahrung mit Spielsucht in seiner
    Familie" — KEINE Namen, Verwandtschaftsgrade, Verlust-Details
  - Post-share-pivot: "...aber jetzt zu dir: was ist gerade los?"
- COACH_SYSTEM_PROMPT (SOS): SOS-MODE LOCK — hard-Verbot Gründer-Story
  zu erwähnen, auch bei direct-ask. Re-trigger-Risk zu hoch.
- DSGVO: brother bleibt komplett anonymisiert. Hans-Müller-DSB-review für
  verbal-consent-doc empfohlen.

VOICE TIER-MAPPING (per user-decision: voice für ALLE tiers):
- New plan-features.voice config: provider + model + voiceId + dailyQuotaSeconds
- Tier-mapping:
  - Free  → Google TTS Neural2-F (de-DE), 60s/day,  ~$4/1M chars
  - Pro   → Cartesia Sonic-2,            300s/day,  ~$4/1M chars + ~75ms TTFT
  - Legend → ElevenLabs Turbo v2.5,      unlimited, ~$30/1M chars
- New backend/server/db/voiceQuota.ts:
  - getRemainingVoiceQuota(userId, plan)
  - consumeVoiceQuota(userId, seconds)
  - estimateAudioSeconds(text)
- speak.post.ts komplett umgeschrieben als plan-aware dispatcher
- 14 tests passing (partial-consume, exhausted, day-rollover, edge-cases)
- Schema-migration 20260509_voice_quota:
  ADD voice_seconds_used_today, voice_quota_reset_at to profiles
  (auto-deploy via pipeline)

Pending Frontend (separate task):
- Voice-quota-UI in Settings/Profile (remaining seconds + upgrade-prompt
  bei 429 quota_exceeded)

⚠️ Schema-migration auto-deploy via b38bf17 detection.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-09 16:28:36 +02:00
chahinebrini
376f3454d6 feat(games,lyra): GameOverScreen migration + Lyra markdown-strip
GAMES (Nuxt → RN migration):
- New components/games/GameOverScreen.tsx — slide-in + fade overlay
  Props: score, bestScore, gameName, onRetry, onExit, isNewBest
- New lib/gameScores.ts — AsyncStorage helpers
  rebreak_best_snake (higher=better), _tetris (higher=better),
  _memory (lower=better, inverted isNewBest)
- UrgeGames.tsx wired: snake-collision/tetris-topout/memory-finish trigger
  GameOverScreen with retry/exit + best-score persist
- TicTacToe NICHT — round-aggregation game hat eigenen Fertig-Flow
- 7 i18n keys (gameOver.* DE+EN, 5 motivational texts statisch aus pool)

LYRA (markdown-bug fix):
- User-Report: Lyra antwortet mit ** in mobile-app, verwirrt user
- Beide system-prompts (COACH_SYSTEM_PROMPT für SOS, COACH_CASUAL_SYSTEM_PROMPT
  für Coach) bekommen "ANTWORTFORMAT - KRITISCH"-section:
  NIE Markdown (kein **bold**, _italic_, #-Headings, -Bullets) — Klartext only
- Reason: Mobile-App-bubbles rendern markdown nicht → User sieht raw `**text**`

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-09 16:17:38 +02:00
chahinebrini
33108a6774 feat(lyra): Coach-Mode persona refactor + mode-badge UI distinction
Per strategist-spec: Lyra-Coach-Mode klarer von SOS-Mode trennen.
- SOS-Mode (urge): crisis-intervention, focused, kurz
- Coach-Mode (lyra): casual, profile-building, philosophy, features

Backend (backend/server/api/coach/message.post.ts):
- COACH_CASUAL_SYSTEM_PROMPT komplett neu strukturiert (~620 tokens)
- Stärkerer Fokus: 3 explicit Aufträge (echtes Gespräch / Profile-Building /
  Rebreak sprechen)
- Profile-building-mandate: "wenn du wenig weißt, sag's ehrlich; frag nach
  Hobbies/Zielen/Menschen — eingewoben, NICHT als Checkliste"
- Cleanere Mission-Section: Bewegung, Anonymität, kein-pathologisieren,
  community-getrieben, DiGA-Listung-Ziel
- Hard-rules klarer: NIE demographics extrahieren (User-Form ist tabu),
  kein Sucht-Vokabular, kein medical-advice
- Existing PLAN_DETAILS-template-var bleibt
- Memory-system unverändert (lyra-memories table, extractAndStoreMemories
  fire-and-forget — kein schema-change nötig)

Frontend Mode-Badges:
- app/lyra.tsx (Coach-Mode): Header-pill "Coach" in brandOrange-tint neben
  Lyra-name
- app/urge.tsx (SOS-Mode): Header-pill "SOS" in error/red-tint neben
  Lyra-name (alt: "Lyra · SOS [v2]" inline-text → cleaner badge-style)

i18n:
- coach.modeBadge.coach + coach.modeBadge.sos in DE + EN

Switch-Logic: route-based (lyra.tsx vs urge.tsx → separate persona via
backend endpoint). Kein User-Toggle — User soll nicht entscheiden müssen
"bin ich grade in Krise?".

Implementation Risk: LOW — schema-neutral, prompt-only + 2 small UI badges.
Erste Beta-Testing-Phase: ~1-2 Wochen iterieren bei Feedback.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-09 16:07:48 +02:00
chahinebrini
192f67cd07 feat(coach): tier-based LLM-Routing + COACH_CASUAL_SYSTEM_PROMPT
Coach-Page ist NICHT SOS — User ist nicht in Krise, will small-talk,
Reflexion, Feature-Wuensche, Philosophie. „Lockere Lyra".

Aenderungen:
- Neuer Prompt COACH_CASUAL_SYSTEM_PROMPT (exportiert): warm/neugierig/
  manchmal humorvoll, bis 4-5 Saetze, darf eigene Empfehlungen + Mini-
  Meinungen formulieren, lädt Feedback aktiv ein. Kein Crisis-Framing.
  Sprachregeln (keine Pathologisierung, kein „Sucht") gelten unveraendert.
- Tier-LLM-Routing analog zu sos-stream:
  Free/Pro = Groq llama-3.3-70b-versatile (Fallback llama-3.1-8b)
  Legend  = OpenRouter anthropic/claude-haiku-4.5 (Fallback claude-3.5-haiku)
- max_tokens 280→500 (Coach darf laenger antworten)
- Demographics-Injection (analog sos-stream): birthYear/gender/etc als
  USER-DEMOGRAPHIE-Block in Prompt (read-only, kein Extract)
- sosMode-Branch deprecated — Frontend kann den Param noch senden, wird
  ignoriert. Folge-TODO: UI-Agent entfernt sosMode aus Coach-Call.

NICHT geändert:
- TTS-Endpoints bleiben plan-agnostisch (Frontend routet by tier)
- sos-stream.get.ts/sos-stream.post.ts unberuehrt (importieren weiter
  COACH_SYSTEM_PROMPT, kein Breaking Change)

Memory:
- project_llm_per_plan.md (tier-LLM-Default-Logic)
- feedback_anonymity_nickname.md
- feedback_demographics_user_initiated.md

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-07 21:22:51 +02:00
chahinebrini
82c6c61a5e fix(sos-stream): profile is not defined ReferenceError
let profile vor try-block hoisten, damit es im plan-routing-Fallback
(line 210, profile?.plan) sichtbar ist. Vorher: const profile innerhalb
try-block → block-scoped → ReferenceError außerhalb.

Demographics-Block-Injection added — gracefully no-op wenn neue Felder
(birthYear/gender/etc.) noch nicht im DB-Schema sind (optional chaining).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-07 21:03:27 +02:00
chahinebrini
355166c194 feat(sos): tier-based LLM-Default — Free/Pro=Groq, Legend=Haiku
Backend (sos-stream.get.ts): wenn sessionData.llmProvider === 'auto'
oder undefined, resolved zu plan-based default via profile.plan:
- legend → openrouter-haiku (Anthropic warm + sort:latency)
- pro/free → groq-llama (sachlich + schnell, ~157ms TTFB)

Frontend (llmProvider.ts): DEFAULT_PROVIDER = 'auto', neue Pill 'Auto'
in LlmProviderToggle. Explicit-Toggles (Sonnet/Haiku/Groq) sind
debug-overrides die plan-logic bypassen.
2026-05-07 04:15:31 +02:00
chahinebrini
4e5f8e6c93 fix(sos-stream): nickname-Injektion in systemPrompt
Lyra halluzinierte Namen wie 'Max' im SOS-Stream weil systemPrompt
keinen Nicknamen-Anker hatte. message.post.ts hatte das Pattern schon,
aber beim Cutover ins backend/-Layout war's nur dort, nicht in
sos-stream.{get,post}.ts.

Fix: getProfile(user.id) + Inject 'NUTZER-NAME: ...' wie in message.post.ts.
Non-blocking (catch), Memory-Pattern preserved.
2026-05-07 04:10:27 +02:00
chahinebrini
431ae6a75d fix(sos-stream): undefined 'key' var nach cutover-refactor
Im Cutover wurde 'const key = config.openrouterApiKey' rausgenommen,
aber line 311 referenzierte 'key' weiter für extractAndStoreMemories.
ReferenceError fiel in catch → 'stream failed' SSE event → App
fiel auf non-streaming fallback (4-5s wait) statt streaming.

Fix: explizit memoryExtractKey aus config.openrouterApiKey holen.
2026-05-07 03:41:11 +02:00
chahinebrini
f2e822be95 feat(sos): llmProvider toggle + sort:latency + bench scaffolding
- backend/coach: routing zu Sonnet (default) / Haiku / Groq Llama je nach
  sessionData.llmProvider. sort:latency für Anthropic-Modelle (-30..58% TTFB).
- frontend: LlmProviderToggle (Sonnet/Haiku/Groq pills), llmProvider.ts
  Storage-Helper. sosStream.ts schickt llmProvider im /sos-session-Body.
- bench: SosTtsBenchmark sammelt Marker (req->session, lyra-ttfb, lyra-done,
  tts-fired/headers/body/file, audio-loaded, first-audio); Output als console.table.
- ops: backend/scripts/llm-bench.sh + Python-Variante für realistic SOS-Prompt.
- speak-cartesia + speak-elevenlabs Endpoints (waren ungetracked, jetzt mit drin).
2026-05-06 13:58:07 +02:00
RaynisDev
b58588cf3c initial commit: rebreak-monorepo (RN app + standalone Nitro backend) 2026-05-06 07:13:43 +02:00