Skip to content

Collaborators

A collaborator is another agent in the same tenant that this agent may hand work to. Olympusx AI offers each collaborator to the agent as a tool; the agent reads the collaborators’ routing keys and decides, turn by turn, whether a question belongs to one of them. A collaborator answers with its own model, instructions, knowledge, tools and skills. You set an agent’s collaborators in the Collaborators section of its page.

Role What they can do
Admin Read and change an agent’s collaborators with Manage.
Builder The same as an Admin.
Auditor Read the Collaborators section.
Reviewer, End user Cannot open it.
Platform admin Everything an Admin can, in any tenant.

The Collaborators section on an agent’s page shows, in order:

  1. A count of the agent’s collaborators and, for an Admin or Builder, Manage.
  2. One entry per collaborator: its name, linking to its page, and its routing key. A collaborator with no routing key reads No description — this agent will be offered and never chosen. Give it one on its own page.
  3. When there are none: This agent works alone.

The How it is connected graph at the top of the page shows the same agents under Agents it hands work to, or It works alone.

  1. Open the agent’s page and, in Collaborators, choose Manage.
  2. In Delegate to, select the agents this one may hand work to. Each checkbox shows the other agent’s routing key, or No description, so it will never be picked.
  3. Choose Save.

To remove every collaborator, clear all the checkboxes and choose Save.

The delegating agent picks a collaborator by reading routing keys only: the When should this agent be used? text on each collaborator. It never reads their instructions.

  • Give every collaborator a routing key. One without it is offered and never chosen.
  • Keep the keys distinct. When two agents claim the same questions, Olympusx AI warns while you type (Another agent claims the same territory), and the choice between them is inconsistent.
  • Add a collaborator when a question belongs to another agent, not to mirror an organization chart. Every delegation is a separate model call.

On the Preview page and in the Run section, a delegation appears as a step delegated to followed by the collaborator’s name. Tool calls made inside a collaborator are counted and billed in the run that started it; the line Tool calls made inside a collaborator, not listed above gives their number, and their detail is in the run’s trace. See Runs and Traces.

Limit Value
Collaborators per agent At most 12. Selecting more shows Too many collaborators and disables Save; the platform refuses more by any route.
Delegation depth At most two hops: this agent hands work to a collaborator, which may hand it to one of its own. A deeper chain fails the run, naming the chain.
Loops An agent cannot delegate to itself. A chain that returns to an agent already on it (A to B to A) fails the run, naming the chain.
Model The agent’s model must be able to call tools. Collaborators on a model that cannot are refused.
Cost Each hop is a separate model call. Everything a collaborator does counts against the ceilings of the agent the run started on: its Tool calls per run, Model calls per run and Tokens per run.