You also get the real numbers for each response: decode tok/s, time-to-first-token, tokens and cost. This one came back at about 210 tok/s with the thinking expanded.

Setup is a wizard: create a Telegram bot with BotFather, pair your chat, then pick a model — DeepSeek, OpenRouter, Claude, OpenAI, or a local LM Studio server. Keys are encrypted with Windows DPAPI.

It runs a local dashboard so I can watch what it's doing live — thinking, answers, tool calls, and what each reply actually cost in tokens and dollars.

Made this for myself — Alfred: a Telegram bot that drives my Windows PC. I message it from my phone and it opens apps, browses in a real Chrome window, reads/writes files, runs commands, and replies…