A new dictionary for the ways you manage agents

Every new layer of the infrastructure stack eventually needs its own vocabulary. Containers gave us "pods" and "sidecars." APIs gave us "endpoints" and "rate limits."

Agents are no exception. And we've been maintaining a dictionary to keep it all straight.

Read on for a quick summary of what's changed or head straight to our glossary to check it out for yourself.

The agent landscape is evolving by the day—and so is the vocabulary that explains it.

Here are just a few of the language gaps we've been running into at Guild: 

• When we talked about an agent that's technically running but not yet visible to a team, "deployed" doesn't quite fit. 
• When we talked about a temporary build generated during a test run, "version" is misleading — it's not really a version, it never shows up in history, and it disappears the moment it's not needed. 
• When we talked about the difference between what an agent wants to run on and what an org allows it to run on, we need two separate words, not one ambiguous one.

So we've been doing what every technical field eventually does: naming things properly.

We've added a batch of new terms to the Guild Glossary, which covers the vocabulary that's grown up around how agents get built, shared, run, and governed on Guild, as well as broader terms used in the industry. A few examples:

• Ephemeral Build — the disposable, cached agent version created when you run guild agent test or guild agent chat, which doesn’t enter version history.
• Qualified Name — our naming convention for agents and resources (owner~resource-name), used everywhere in the CLI.
• LLM Preferences vs. Model Policies — the distinction between what an agent asks for and what an organization actually enforces.

There are twenty new entries in total, spanning core platform concepts (Agent Hub, Smith, Skills, Triggers), the mechanics of how agents actually get built and versioned (Saved / Validated / Published states, Guild Native Agents, Semver Ranges), and the open standards we plug into (Goose, Jinja Templates, OpenAPI, Zod).

Some of these terms describe things that are genuinely new — patterns that didn't need names until agents started running in production at scale. Others are our attempt to bring precision to words the industry already uses loosely, so that when we say "session" or "skill," we mean something specific and testable.

If you're building on Guild, you'll run into these terms in the CLI, the docs, and conversations with our team.

Check out the full Glossary here.