A handful of founding users use Nynch primarily through the Telegram bot. “Send a Nynch about the call with Sarah.” “What changed on the Acme account this week.” A relationship-CRM that lives in the same place as your conversations is genuinely useful.
The problem: Telegram gives a bot a fixed window to acknowledge a message before treating the conversation as dropped. Anything Nynch does that touches multiple records (a relationship brief, a synthesis across recent meetings) was reliably exceeding that window. The bot would do the work, but Telegram had already moved on. The user saw silence.
We rewrote the response pattern.
The bot now acknowledges within Telegram’s window, then processes the request in the background. When the work completes, it posts back into the same thread with the result. Every code path that touches the model or the database now has explicit error handling, so if something fails (a rate limit, a timeout downstream, a malformed query) you get a clear message back instead of nothing.
What changes for you. Silent failures stop. The bot is dramatically more reliable on the longer requests that were the most useful ones. If you stopped using the Telegram bot because it felt flaky, this is your reason to try it again.
