Skip to content

Releases: core-euler/sna_net

InnerCore v0.5.0

15 Jun 13:51

Choose a tag to compare

InnerCore v0.4.5

15 Jun 13:42

Choose a tag to compare

InnerCore v0.4.4

15 Jun 12:24

Choose a tag to compare

InnerCore v0.4.3

15 Jun 12:07

Choose a tag to compare

InnerCore v0.4.2

23 May 11:23

Choose a tag to compare

InnerCore v0.4.2

  • Оплата через Google Play временно отключена — экран Pro показывает «оплата скоро будет доступна». Тарифы и лимиты работают как прежде.
  • Исправлена кнопка обновления приложения (теперь корректно открывает браузер).
  • Backend: исправлена отправка email-кодов (SMTP SSL + публичный DNS).

APK: app-release.apk (подписан, Android).

InnerCore v0.4.2-test (IP build)

23 May 12:30

Choose a tag to compare

Pre-release

Тестовая сборка — НЕ для публики

APK собран с API_BASE_URL=http://89.125.77.243:8080 (backend через reverse SSH-туннель, plain HTTP).

  • Подключение к бэку: анонимный и email-вход работают.
  • Google-вход: отключён в этой сборке (GOOGLE_WEB_CLIENT_ID не задан).
  • Помечен pre-release → встроенный апдейтер его НЕ предлагает обычным пользователям (latest остаётся v0.4.2).
  • Работает только пока поднят туннель 8080 на VPS.

InnerCore v0.4.1

29 Mar 18:51

Choose a tag to compare

InnerCore v0.4.1 — Auth, Billing, Fixes

JungAI v0.4.0 — Audio Chunking, User Memory, Map & Localization

26 Mar 08:50

Choose a tag to compare

What's New in PATCH 0.4

🎙️ Voice Input — Long Recordings

  • Long recordings (60+ sec) now transcribe reliably
  • Audio is automatically split into 15-second segments and processed in parallel
  • UX remains a single continuous recording — segmentation is invisible to the user
  • Re-recording appends new text instead of replacing the previous input
  • Per-segment retry with partial result fallback on failure
  • Voice input now works in the dream chat (not only in dream entry)

🧠 Long-Term User Memory (user.md)

  • LLM now maintains a persistent psyche map per user: recurring themes, dominant archetypes, emotional states, inner conflicts
  • Updated after each dream analysis and significant dialogue
  • Evolves incrementally — not rewritten from scratch; fading patterns are weakened over time
  • LLM reads user.md before every analysis and incorporates dynamics without repeating known conclusions

🗺️ Symbol & Archetype Map

  • Embeddings are L2-normalized before UMAP — points now spread across the full space rather than clustering per-dream
  • min_dist reduced from 0.08 to 0.02 — cross-dream symbol overlap is more visible
  • Deterministic jitter seeded by md5(user_id:symbol_name) — map layout is stable across refreshes
  • Max zoom increased from 3x to 10x with LOD (larger labels at high zoom)
  • Map cache bumped to v4

⏱️ Temporal Dynamics

  • Each RAG chunk now stores source_recorded_at, source_created_at, source_order
  • Recency bonus in hybrid scoring: 0.1 * exp(-days_ago / 30) — recent dreams rank higher
  • Chunks in the prompt are ordered chronologically
  • LLM sees a timestamp on every memory fragment

🌍 Localization (EN / RU)

  • All hardcoded Russian strings removed from 6 screens and the message menu widget
  • ~40 new ICU keys added to EN and RU ARB files

⚙️ Infrastructure

  • Alembic migrations: 001 baseline, 002 user_memory_docs, 003 temporal fields with backfill
  • entrypoint.sh in the backend container — alembic upgrade head runs automatically on every container start

Client version: 0.4.0+5
Minimum supported version: 0.3.2

JungAI v0.3.2

17 Mar 11:40

Choose a tag to compare

New Features

  • In-app update check — the app checks for new versions on startup and shows an update dialog with a download link.
  • Minimum version enforcement — server blocks outdated clients (< 0.3.2) with 426 Upgrade Required. Old versions (0.3.0, 0.3.1) will no longer work.

Install

Download app-release.apk and install over the previous version. Future updates will be prompted automatically.

JungAI v0.3.1

17 Mar 11:27

Choose a tag to compare

Bug Fixes

  • Chat replies now retry on LLM failures — the chat task retries up to 4 times with exponential backoff instead of failing silently on the first error.
  • Analysis no longer stuck in PENDING — when all retries are exhausted, analysis status correctly transitions to FAILED.

New Features

  • CometAPI fallback provider — when the primary LLM provider (Gonka Proxy) is down, requests automatically fall back to CometAPI (gpt-5.1). Transparent to the client.
  • Archetypes pie chart — profile screen now shows archetype distribution as a donut chart with a color legend, replacing the old progress bars.

Install

Download app-release.apk and install over the previous version.