Currently, the tg bot is using oneshot qa flow. This requires embedding the message every time a request is made. The bot should instead create sessions - such that the LLM can evaluate if new search is necessary. Prototype is already implemented in CLI chat agent. It has to be implemented in the telegram bot.
Currently, the tg bot is using oneshot qa flow. This requires embedding the message every time a request is made. The bot should instead create sessions - such that the LLM can evaluate if new search is necessary. Prototype is already implemented in CLI chat agent. It has to be implemented in the telegram bot.
Implement the logic implemented here https://github.com/svoi-fr/mirai/blob/main/backend/chattest.py In here https://github.com/svoi-fr/mirai/blob/main/backend/danswer/danswerbot/telegram/listener.py