Improve search: pymorphy3 stemming, top_k=4, RRF 0.3/0.7, restructured KB

chore: clean prompt, remove phone/operator-ban

fix: cleaner off-topic message

fix: general greeting, LLM detects off-topic → operator transfer

feat: expanded operator triggers, insoles-only focus, clear history on transfer

fix: parse URL-encoded form data from Bitrix outgoing webhooks

feat: Bitrix24 webhook вместо Telegram polling

fix: TLS 1.2 + round-robin Telegram IPs + client recreation on error

restore: exact code from b4a1c1a (search_debug + httpx polling)

restore: exact working code from 1dbfd30 (httpx, no requests, no ssl hacks)

fix: configure_http_backend + requests.Session (from forum solution)

restore: rollback to working commit 1dbfd30 (httpx polling, long poll)

fix: back to polling + requests (webhook broken on HF Spaces)

fix: switch from httpx to requests.Session for Telegram API

feat: test_tokens.py + TLS 1.2, _ssl_handshake_timeout, infinite webhook retry

fix: tg_post param name json_data vs json kwarg

fix: shared Telegram API client (reuse connection, avoid re-handshake)

fix: webhook (aiohttp) + Python 3.12.9-slim (SSL bug fixed)

fix: ssl.OP_IGNORE_UNEXPECTED_EOF for Py3.12<3.12.3 bug

fix: increase SSL handshake timeout 15→120, keep client alive

fix: short poll (timeout=0) + new connection per request, no SSL EOF

fix: recreate httpx client on SSL EOF instead of reusing stale connection

feat: dialog history (10 msg) per chat_id

Revert "fix: add bot logging, suppress transformers progress bars"

Revert "fix: typing indicator, local LLM timeout (120s), short polling"

Revert "fix: simplify prompts for Llama 3.2 3B"

fix: simplify prompts for Llama 3.2 3B

fix: typing indicator, local LLM timeout (120s), short polling

fix: add bot logging, suppress transformers progress bars

feat: search_debug(), RAG details in web logs

feat: web log viewer at /logs, last 50 interactions

rewrite: sync httpx polling, no aiogram/aiohttp

fix: set loop._ssl_handshake_timeout = 300s

fix: override _connector_init SSL context in constructor

fix: custom AiohttpSession with TLS 1.2, force IPv4

fix: remove unused TCPConnector, simplify AiohttpSession

fix: use asyncio.run + start_polling, full traceback

fix: custom SSL context TLS 1.2, TCP diagnostics

fix: force IPv4, ssl_handshake_timeout=120s

fix: retry bot on SSL timeout, custom aiohttp session with longer timeouts

hybrid search (bge-m3 + BM25 + RRF) + local Llama 3.2 3B fallback

fix: hardcode KB paths, ignore KB_PATH env var from Render

fix: switch to llama-3.3-70b, stricter prompt against hallucination

fix: strict system prompt - no price hallucination

feat: simplify operator redirect to 'Переход на оператор'

feat: split KB into general + insoles, add operator triggers

always pass questions through Groq even without KB matches

initial: ORTOS Telegram Bot v2