Introducing BotTrottr Teams, Swarms & Skills

Today we’re excited to share the next step in how BotTrottr organizes AI agents: Teams, Swarms, and a three-tier Skills system. Together they turn a pile of independent bots into a coherent, shared, and predictable operating unit.

The problem: hidden skills, unpredictable agents

Until now, an agent’s capabilities lived in a few opaque places — a system prompt, a bundle of MCP connectors, and whatever files happened to be in its storage. That works for a single bot. It breaks down the moment you have a team of agents that should share knowledge, tools, and memory.

The result was the worst kind of surprise: an agent that behaved differently in production than it did in testing, because it was silently loading a skill nobody knew about.

The building blocks

Swarms — the operational unit

A Swarm is the shared environment an agent belongs to. It owns the skills, the memory, the tools, and the connectors its agents use. Bind an agent to a Swarm and it inherits the Swarm’s capabilities automatically — no per-agent copying, no drift.

Skills — three tiers, clear precedence

Every capability is now a first-class Skill: a SKILL.md plus optional references, templates, and scripts. Skills come in three tiers, and when two share a name, the most specific wins:

1. Platform Skills — global, trusted, and signed by BotTrottr. The shared defaults everyone starts from.
2. Swarm Skills — installed by a Swarm maintainer, shared by every agent in that Swarm. These auto-apply and can’t be silently removed.
3. User Skills — your private skills, attached to an agent only when you explicitly choose them.

Progressive disclosure + version pinning

Skills load the way a good reference manual works: at startup we read only the metadata (name, description, tags). The full SKILL.md — and its referenced files — load on demand, only when a task actually matches. Each agent pins the exact version of every skill it uses, so a new release never changes behavior out from under a running Swarm.

A skill selector that shows you the truth

The Deploy Agent screen now has a Skill Selector that makes all of this visible. You see Platform Skills, the locked Swarm Skills your agent inherits, and your own User Skills — each with its name, version, and source. No more guessing what powers your agent. Upload a skill to your personal or Swarm storage, tick the ones you want, and deploy.

Teams, next

Swarms are the operational unit; Teams are the ownership boundary above them. One team owns many Swarms, with roles (Admin, Maintainer, Member) controlling who can install skills, promote changes, and attach agents. Multi-tenant isolation means one team’s experiments can never leak into another’s production.

What’s next

This is the foundation. On top of it we’re building skill promotion (dev → staging → prod with signatures), a shared Skills marketplace, and full team management. We’ll share more as each lands — subscribe to the RSS feed to follow along.

Try it today: create a Swarm, attach your agents, and see exactly what they know.