Before there was a 19-agent team, there was one small automation. This is the true, unglamorous origin story of how UseTalos started - a single workflow, one channel, one job.
Before UseTalos was a company, it was one founder trying to solve one annoying problem: too many messages, not enough hands to answer them. So the very first build wasn't a product, a pitch deck, or a plan for a fleet of AI agents. It was a single automation - built in n8n, a workflow tool - that did exactly one thing well.
A message would come in on Telegram. The workflow would wake up, hand that message to GPT-4o, and GPT-4o would answer - remembering the last stretch of the conversation using a window-buffer memory node, so replies didn't feel like talking to a goldfish. That's it. No dashboard, no fleet, no org chart. Just a trigger, a model, and a memory.
That one workflow is what everything else at UseTalos was measured against later: does a new idea help one person get one answer faster, in one place they already check? If yes, it earns its keep. If it's complexity for its own sake, it gets cut.
Here's the actual shape of that first build - three steps, one loop back for memory.
A message lands in the chat and wakes the workflow up.
The model reads the message and drafts a reply.
The answer is sent back in the same chat.
This describes the shape recorded in the original workflow (Telegram trigger → GPT-4o → window-buffer memory). It's a diagram of the pattern, not a screenshot of the export.
That one-channel, one-model habit didn't stay small - it became the working pattern behind UseTalos itself. Today the team runs on a fleet of AI agents, each one still built on that same idea: give one worker one clear job and one place to do it, and let it get good at that.
Plus a dedicated security/governor agent watching the whole system, and the original personal assistant, still running as the founder's own day-to-day tool. The fleet grew agent by agent, the same way the first one was built: start with a real, boring, recurring task, wire up one channel and one model to handle it, and only add the next piece once the first one is actually being used.
Note: exact agent capabilities and internal wiring are Talos's own working system and aren't itemized here - this section describes shape and scale, not a feature list.
You don't need a "fleet" on day one. Almost nobody does. The version of this that actually works for most small businesses looks like the very first UseTalos build:
