You don't have to rip anything out. Add Telm alongside your current bot in monitoring mode: it watches, logs what it would do, and touches nothing. Compare its journal with your bot's decisions for a few days, move your keyword lists over, then switch enforcement on and retire the old bot. Total risk during migration: zero, because only one bot is ever enforcing.
1Why groups move
The usual triggers: keyword filters started missing rephrased spam (AI reads meaning, patterns don't), a wrong ban cost a valued member and there was no appeal path, or managing hundreds of chat commands got old and a dashboard sounds better. Our honest side-by-side comparisons live here: Telm vs Rose and Telm vs Combot — including what those bots still do well.
If your current setup works for you, keep it. The migration below is for when it doesn't.
2The four-step migration
The whole point is that both bots coexist safely while you decide.
- 1. Add Telm to the group — it starts in monitoring mode and deletes nothing
- 2. Watch for 3–7 days: open Telm's journal and compare its would-be decisions with what your current bot actually did — pay attention to what Telm catches that Rose/Combot missed, and vice versa
- 3. Move your custom bits: keyword blocklists become custom rules, trusted domains go to the whitelist, welcome text into welcome messages
- 4. Switch Telm to enforcement, then remove or mute the old bot — running two enforcing bots makes them race on deletions
3What carries over (and what doesn't)
Carries over easily: keyword and domain lists (paste them into rules), welcome message text, your ban ladder logic (warn → mute → ban is configurable), captcha-on-join behavior.
Doesn't transfer: historical analytics from Combot (export their reports first if you need them), Rose federations (Telm has its own cross-group ban network — bans you make in Telm propagate across your groups), and existing warn counters — members start with a clean slate, which most communities consider a feature.
4The first week after switching
Keep the journal open. If you see a wrong decision, fix it with one click — the appeal system also catches mistakes automatically, and every correction teaches the system your group's normal.
Give the personal ML model your chat export once you're on a plan that includes it: that's the single biggest accuracy jump for communities with their own slang.