AI in the loop · Distri
The agent behind the Copilot
The Copilot runs on Distri. It reads the picture through tools, proposes into the console, and the console works without it.
Distri is the agent platform the Copilot runs on. It supplies the model, the conversation and the tool-calling loop; DomeCommand supplies the tools and the surfaces a proposal lands on.
The console never depends on it
If Distri is unreachable, misconfigured or switched off, the chat panel goes quiet and everything else works. The air picture, fusion, the planner, posture and the gates are all in the server and none of them ask an agent anything.
This is a hard requirement rather than graceful degradation for its own sake. An air picture that stops updating because a language model timed out would be worse than no assistant.
How the browser reaches it
The console never holds an API key. It asks the DomeCommand server for a short-lived token, which the server mints from its own credential and refreshes on a cadence.
So the key stays server-side, a browser session carries something that expires, and revoking access is done once in the server config rather than per user.
Two kinds of tool
Read tools answer questions about the current state. The situation, the tracks, recent assessments, the candidate plans and their scoring, a finished run's timeline. The agent calls these itself as it works.
Proposal tools produce something for you to decide on. There are two, and they behave identically in the one way that matters:
| Tool | Produces |
|---|---|
| Compose a scenario | A scenario draft, rendered in the scenario form |
| Draft a rule | A rule draft, rendered in the rule editor with its verdict |
Neither saves. The agent hands the console a proposal, the console renders it on the page that owns it, and you commit it or discard it. The chat is where you talk; the page is where you decide.
A drafted rule is validated before it is offered, so a proposal that would be rejected shows you why in the editor rather than failing on save.
What it cannot do
The Copilot cannot approve a plan, change a posture, issue a grant, or task an asset. It has no tool that commits an action, so the limit is structural rather than a rule it has been asked to follow.
Authority runs through the gate ladder, and the gate ladder does not have an entry for an agent. See Decision support.
Configuring it
A DomeCommand deployment needs a Distri workspace, an agent in it, and a credential. Set those on the server and the chat panel comes up.
Without a configured model provider the assistant runs deterministic, which is useful for tests and demonstrations and is not useful for briefing. If the panel answers but the language is stiff and repetitive, that is the state it is in.
Where to go next
- The C2 Copilot for what you can actually ask it.
- Policy & rules for the drafting flow end to end.