v2026.05.17

🧾 v2026.05.17 — AI Audit Logs and Grok Model

✨ New Features

• 🧾 AI Audit Log — every OpenRouter evaluation now writes the exact request body and raw response to logs/ai_audit.jsonl
• 🧹 7-Day Audit Retention — audit records older than 7 days are pruned by created_at on each write
• 🤖 Grok 4.3 Default Model — the default OpenRouter model is now x-ai/grok-4.3
• 🧵 Context-Friendly Prompting — the quote prompt now explicitly asks for the minimal dialogue context when the punchline depends on setup or a reply chain

🔧 Improvements

• 🆔 The AI contract now clearly says to use internal id values for primary_id, context_ids, and scores, while message_id remains Telegram metadata only
• 🐳 Docker Compose now mounts /app/logs so file logs survive container rebuilds
• 🧼 .dockerignore keeps .env, local logs, .venv, .git, and collaborator files out of Docker build context
• 🔒 AI audit records do not include authorization headers or API keys, and are written to files instead of Docker stdout/stderr

🗃️ Database Changes

No database migration required.

✅ Validation

• .venv/bin/python -m pytest — 44 passed
• .venv/bin/alembic heads — 20260513_01 (head)
• git diff --check — clean
• WS04 docker compose config — clean for the existing quoto stack
• OpenRouter public model list confirms x-ai/grok-4.3

---

Full documentation: README.md | README.ru.md