This paper uses text data from Telegram - a messaging platform with social media like features - to understand how online political activism connects to offline contentious actions in autocratic contexts. Telegram was developed by the creators of VKontakte (VK) in response to repression they faced from Russian authorities for refusing to remove opposition groups from VK. The history of Telegram is steeped in anti-regime dissent and multiple instances where the creators have denied the government access to its data. This has made it popular amongst dissenters, keen on escaping state surveillance. Yet there is limited research on Telegram. In this study, I use BERT models to classify discussion of protests and criticism of the government in 50+ public group chats on Telegram. These groups can host 200 thousand members and is a rich source of text data which remains under studied. I subsequently model what predicts these online discussions over the course of 2 years. I use features of the protest, the government’s response to it, and regional population characterizes as my main predictors. In addition, I also provide a framework for analyzing data from this platform.