The easiest way to make an agent feel “real” is to separate planning from execution.
- Planning needs long-form thinking (ChatGPT is great at this).
- Execution needs fast nudges and quick replies (Telegram is great at this).
If you try to do both inside one surface, your habit agent becomes either:
- a long chat you never open, or
- a spam bot you mute.
Buffy’s design is multi-channel by default: the behavior core is channel-agnostic, and each channel is a thin adapter.
This post outlines a simple workflow you can use directly, especially if you’re orchestrating agent experiences with OpenClaw.
What you’ll learn
- A 3-message ChatGPT → Telegram loop you can run today.
- Why a shared behavior core beats two disconnected bots.
- Where to wire OpenClaw without fragmenting habit vs todo state.
Step 1: Plan in ChatGPT
In ChatGPT, define the intent clearly:
- “Weekdays, morning startup: water, planning, stretch between 7:30–8:00.”
Buffy turns that into activities (habit/task/routine) that can be executed anywhere.
If you want the ChatGPT-specific framing:
Step 2: Execute in Telegram
Telegram is where you:
- get nudges at the right moments
- reply “done” or “snooze 20”
- keep momentum without opening a dashboard
A 3-message script (copy this flow)
-
In ChatGPT (planning):
“Weekdays 7:30–8:00, morning startup: water, 10-min planning, stretch. Remind me in Telegram.”
Buffy creates the routine and wires reminders to your Telegram. -
In Telegram (first nudge):
“Morning startup window’s open. Water now, snooze 20m, or skip today?”
You reply done (or snooze 20). The same behavior core logs it. -
In Telegram (follow-up or next step):
“Planning and stretch still in the window. Do them now or snooze?”
Again: done / snooze / skip. No dashboard—just one thread.
Optional step 4 (same day): Back in ChatGPT, ask “How did my morning startup go?”—you should get an answer grounded in what you actually tapped in Telegram, not a blank chat.
That’s the full loop: plan once in ChatGPT, execute in Telegram with minimal friction.
Telegram post:
Step 3: Keep it consistent with a behavior core
The reason this workflow scales is that:
- ChatGPT and Telegram are interfaces
- the behavior core stores the truth (activities + history + reminders + memory)
That’s what prevents fragmentation across OpenClaw workflows.
Integration overview:
FAQ
Why not just use a Telegram habit bot?
Because most bots become isolated checklists. If you want habits to coordinate with tasks and routines—and to improve over time—you need a behavior engine underneath.
What’s the OpenClaw connection?
OpenClaw can orchestrate agent experiences, but you still need a stable behavior core so “habit agent” and “todo agent” don’t drift apart:
Where to go next
- Next step: Run the 3-message flow yourself—plan one routine in ChatGPT and get your first Telegram nudge in under 5 minutes: How to Get Started With Buffy Agent in 5 Minutes