Trigger rebuild for demospace

Update README

Update space

Update backend with Google OAuth and refresh documentation

Update space

Added security checks, updated README.md to reflect that

Finalized changes, minimal clean-up

Fixed chat permanency

UI FIX

UI FIX

fixed prompt build again

fixed prompt build

changed websearch api from duckduckgo to tavily AGAIN

changed websearch api from duckduckgo to tavily

fixed requirements

downgraded model to fit ram constrains

fixed web search

changed parse user mode, making it easier for the code to detect casual talk

resource tunning

tunned prompt

turned off f16_kv

Reduced max tokens from 1024 to 512

Reduced n_ctx from 8000 to 4000

Fixed web search and file upload

Changed README to reflect code changes

Finetuning model and parameters again, ram increase

Finetuning model and parameters

Finetuning model and parameters

fix: returning to full production app.py

fix: testing stream

fix: added stay alive to stream

fix: readloop

fixed generate function

fixed f-string error

fixed minimal stream route

changed to a smaller qwen model

fix: align prompt budget with context window and add keep-alive to stream

fix: resolve NameError for MODEL and optimize RAM context

fix: resolve NameError for MODEL and optimize RAM context

fix: remove ollama dependency and use self-contained llama-cpp

fix: update gunicorn path to find app in backend folder

Initial clean production-ready commit