When I first set up OpenClaw, I thought I'd use it for a few things. Email triage, maybe some quick research.
That was naive.
Within weeks it was handling customer success, writing content, booking meetings, monitoring my inbox, sending follow-ups, doing research, drafting proposals. The moment it actually works, you keep giving it more. That's just how it goes.
And then the single chat thread started breaking down.
Marketing ideas from Tuesday buried under calendar confirmations from Wednesday. A half-finished blog post sandwiched between a customer reply I hadn't approved yet. Ask it something about the content calendar and you'd get a solid answer with just enough noise from an unrelated conversation earlier to be slightly off. Not wrong — just contaminated.
At some point it became genuinely impossible to run multiple workstreams through one conversation. The context was a mess, the agent kept switching modes, and I was constantly re-establishing what we were doing and why.
I needed structure.
What I built instead
I kept the main 1-on-1 chat for day-to-day things — quick questions, personal requests, whatever comes up in real time.
Then I created a private Telegram group alongside it with Topics enabled. That's where the professional workstreams live. Think of it like departments in a company: the main chat is your direct line to the assistant, the group is the office.
Customer Success. Content. Marketing. Research. Operations. Ideation. Each one its own room, with its own conversation history, its own context, and no bleed from the others.
This blog post? Written in the Marketing topic. The customer email I approved this morning? Customer Success topic. The feature research from last week? Still sitting in Research, exactly where I left it, ready to pick up.
That's the thing that's hard to explain until you try it. When you walk into the Content topic, the agent is in content mode. It knows what we've been building, what the tone is, what's already written. It's not distracted by the scheduling conversation from three topics over.
The model thing nobody talks about
OpenClaw lets you set a different AI model per topic. In practice: run a cheaper, faster model in topics where you need quick outputs — status checks, simple lookups, notifications — and a more capable model where quality actually matters, like content writing or strategic decisions.
My Ops topic runs on a lighter model. My Content topic gets the better one. The cost difference is real — more capable models cost significantly more per token — and most of the work doesn't need the expensive option. Matching model to task cuts the waste without touching quality where it counts.
The associative thinker problem
I have 12 topics now. I'm still catching myself trying to fire messages into three different threads while something else is mid-process. That's not the tool's fault — that's just how I think, and I suspect anyone who ends up building a serious AI workflow has a similar tendency.
But the structure helps anyway. Even when I'm jumping around, each thread stays coherent. I can disappear from the Content topic for three hours, come back, and pick up exactly where it left off. No rebuilding context, no re-explaining what project this is.
For that alone it's been a game changer. Everything else is a bonus.
How to set it up
Keep your existing 1-on-1 chat — that stays as your personal day-to-day assistant.
Then:
- Create a new Telegram Group
- Go to Group Settings and enable Topics
- Add your OpenClaw bot to the group and promote it to Admin
- Create a topic for each workstream you want to separate
One bot, two interfaces. The direct line for personal requests, the group for structured professional work.
MoltBot Ninja is the managed way to run OpenClaw — per-topic model configuration included, no server setup required.