LoRa's long range but low bandwidth already enables telemetry messages from as far as orbiting satellites with $20 hardware (https://tinygs.com/). For motivated, low time preference users, such as Nostriches, I think there must be a way to hyper-efficiently encode/decode longer text messages using local language models, trading work at sender and receiver for better transmission efficiency, mitigating bandwidth constraints and enabling far wider mesh coverage. Whether or not it can be made to work, I'm finding it interesting to explore https://njump.me/