CarsRUS: session car history – stick to discussed models on follow-up

CarsRUS: single base layout, chat main, expanded arch popup, contrast light/dark

CarsRUS: credit signature (גל שלמה בן דוד), fixed header, arch popup; examples verified

CarsRUS: Link & Co 01 normalization (and/לינק&קו), tests

CarsRUS: sticky header, single final answer (no duplicate), API link in footer

Expose Gradio chat API (/chat) and enable queue(api_open=True) for cloud tests

RAG: better normalization for Genesis GV80 + Link & Co Hebrew; add test_chat_scenarios manual script

UI: +11 chat examples (Hebrew/English); design tweaks (examples grid, label, input focus, spacing)

OpenRouter: default model google/gemini-3-flash-preview (fix 404, gemini-2.0-flash-exp:free deprecated)

Generation: use only OpenRouter when OPENROUTER_API_KEY set (no Gemini fallback)

OpenRouter: log when key missing; try more env names; doc HF secret OPENROUTER_API_KEY

OpenRouter first (faster), Gemini fallback; conversation history Q+A; optional .env

Pipeline logging + show LLM response in chat

Rate limit: longer retries (8 attempts, 3 min wait); show as 'בעיה זמנית' not 'התשובה'

Present aggregated verbal answer first in chat; strengthen RAG prompt

UI/UX: RTL layout, dark mode, stable header logo, chat/input RTL, scrollbar, focus states; agent: retrieve→generate→end, final answer separator

Fix: Hebrew Elantra N detection, final answer display, stronger prompts; UI redesign; Hebrew comparison test

RAG: comparison by supported models, partial answer for 1 known model, stronger prompts for context aggregation

Remove test_agent.py and test_rag.py

tests

agentic rag update

fixing timeout bug while waiting for gemini response

create vector offline

faster réponse

flash models for faster response

not relevant

not relevant

not relevant

rate limit fix

optimize rag flow

fix bug

bug

upload files

update

🔧 Fix: Remove problematic demo.load() call - simplify to basic buttons

🔧 Fix: Simplify button handlers with JavaScript instead of component references

🔧 Fix: Smarter rate limit handling - wait 30s on 429 errors instead of switching models

✨ Feature: Add interactive buttons for car models and comparisons in sidebar

✅ Improve: Increase minimum delay to 3s + more aggressive exponential backoff (3^n) for rate limit handling

✅ Fix: Add rate limiting with exponential backoff + response caching to prevent API quota errors

✅ Fix: Update to correct Gemini API models (gemini-2.0-flash, gemini-1.5-flash, gemini-1.5-pro) + Clean minimal chat UI

🔧 Fix: Dark chat UI + correct Gemini API model names

🎨 Simplify UI to clean professional design + 🐛 Add better error handling

🎨 Redesign UI with stunning gradients, modern cards, and premium styling

✨ Implement 10 Golden Rules for RAG

UI Fix: Force strict high-contrast (Black/White) and classic input layout

Fix API 404 (upgrade lib) and UI Readability (High Contrast)

UI Overhaul: Clean & Fun design (Claude-like)

Fix crashes: Pin numpy<2, remove invalid Gradio args, fix paths

Refactor: Move scraping logic to data_ingestion folder