How Trebuh Runs 4 OpenClaw Agents as a Team from One Telegram Chat
Built by Trebuh · · Source: GitHub
Result: Trebuh described the setup as "a real small team available 24/7." The source gives no cost or revenue numbers.
What they built
According to the awesome-openclaw-usecases repo, Trebuh is a solo founder who set up 4 OpenClaw agents on a VPS and controls all of them through a single Telegram chat. Trebuh described the pattern on X, and the repo turned it into a write-up with an example team. The problems it lists are familiar to anyone working alone: one agent can't do everything well, generic prompts give generic output, the founder burns out, and marketing research never reaches dev priorities.
How it works
The write-up doesn't share Trebuh's exact configuration. It gives an example team of four agents, each with its own role, personality and model:
- Milo, strategy lead. Runs on Claude Opus. Handles planning, priorities and coordination, posts a standup at 8:00 AM and a recap at 6:00 PM, and drafts weekly priorities on Mondays.
- Josh, business and growth. Runs on Claude Sonnet. Covers pricing, competitor analysis and KPIs, pulls metrics at 9:00 AM, checks competitor pricing weekly, and compiles a metrics report on Fridays.
- Marketing agent. Runs on Gemini. Surfaces content ideas at 10:00 AM, does SEO research, and monitors Reddit and X every two hours.
- Dev agent. Runs on Claude Opus or Codex. Writes and reviews code, fixes bugs, checks CI/CD health and reviews open PRs.
Three pieces hold it together, according to the source:
- Shared memory. A team folder holds goals, a decisions log and project status that every agent can read. Each agent also has a private folder for its own context.
- Telegram routing. Everything happens in one group chat. Agents answer when tagged (@milo, @josh, @marketing, @dev or @all), and untagged messages go to Milo.
- Scheduled tasks. A schedule file defines the daily, ongoing and weekly jobs, so the agents do work without being asked.
The listed requirements are a Telegram skill, the session spawn and send tools for coordination, a shared notes system, API keys for each model provider, and an always-on machine or VPS.
The write-up adds that other OpenClaw users have described similar setups, including @jdrhyne with more than 15 agents across 3 machines and @danpeguine with two OpenClaw instances working together in one WhatsApp group.
The result
The source gives no cost, revenue or time numbers. It reports that Trebuh described the setup as "a real small team available 24/7." The authors say the scheduled, proactive work delivers the most value, and that distinct personalities make the team easier to work with.
How to build something similar
- Start with two agents, such as a lead and one specialist, and add more only when you hit a real bottleneck. The write-up gives the same advice.
- Give each agent a clear role, a name and a model that fits the difficulty of its work. Save the expensive model for planning and code.
- Set up one shared folder for goals, decisions and status, plus a private folder per agent, before you add any schedules.
- Put everyone in one group chat with tag based routing and a default agent for untagged messages, then add a few scheduled jobs and adjust them after a week of use.
The original post
Want to build something like this?
Join the Shipping Skool community free. Paid members get live bootcamp calls every week.
Join Shipping Skool — Free