For agent builders

Fleet identity, per-agent budgets, group defaults, and attribution — the setup for teams building and running autonomous agents on Relixr.

An agent isn't a person, so the questions you need answered for it are different: not "who can access billing" but "which agent spent this, and can I stop it right now."

Start with agent identity, not a shared key

In Relixr, an agent is a first-class identity with its own bound key — not a role your app's shared credential plays. Use the Agents sandbox to create one: pick a template (Support, Code review, Research), get a default daily cap and model allowlist, and copy the bound key once from the first-run panel.

Groups for shared defaults

Agents doing the same job usually should share the same spend limit and model allowlist. Set the budget and allowlist on the group once, then attach agents to it — each agent still keeps its own key and its own usage attribution, but you're not re-configuring the same policy per agent as the fleet grows.

Attribution is per key, not per inference

Every response and every usage row is tagged with the agent's key id, so "what did this agent cost this week" is a filter in Usage analytics, not a manual reconciliation. This is also what makes a runaway agent containable: freeze it to stop spend instantly, or archive it to revoke the bound key outright — neither action touches any other agent in the fleet.

Scaling the fleet

The pattern for agent #50 is the same three steps as agent #2 — see Agent fleet on Relixr for the full reference architecture, and Setting spend controls for how per-key budgets interact with workspace-level caps.