VegaDūtaPlatform · Agents
Agent teams: one lead, several specialists
A team is one agent plus the other agents it may hand parts of a task to. The lead agent decides what to delegate, each member does its part, and the lead combines the answers into one reply. With two or more members the lead can send work to several of them at once instead of one after another.
Teams are set up on the agent's Team / Swarm card in the console. Delegation is deliberately bounded: it goes one level deep, the number of parallel branches is capped, each branch has a time limit, and every branch draws on a single budget for the run.
By the VegaDūta team · Last updated
In short
- A team on VegaDūta is a lead agent plus the other agents it may hand parts of a task to.
- With two or more members, a lead agent can delegate to several members in parallel and combine their answers; delegation goes one level deep.
- Each delegation call runs at most 8 parallel branches, each branch is stopped after 240 seconds, and all branches share one per-run budget.
How delegation works
Team members are ordinary agents added to the lead agent as collaborators. The lead sees them as tools it can call: ask one member, or delegate to several in parallel. Members answer the lead and cannot hand work on to further agents. Handoffs appear as their own steps in the Agent Lab run graph.
The limits, stated plainly
Parallel delegation multiplies model calls, so the limits are fixed and shown on the card before you turn a team on.
- At most 8 parallel branches per delegation call; extra members wait for a later call
- Each branch is stopped after 240 seconds
- All branches share one per-run budget, 2 US dollars by default; the run stops when it is used up
- Depth 1 only: members cannot delegate further
Presets to start from
Three presets create the member agents for you, using the lead agent's provider and model: a research crew that gathers facts, challenges them, and writes them up; a review pair where one agent drafts and one reviews; and a support triage team that classifies a request, resolves it, and escalates what is left. Existing collaborators are kept when a preset is applied.
Frequently asked questions
Can several AI agents work on one task at the same time?
Yes. When a lead agent has two or more team members it can delegate to several of them in parallel and combine the answers. Up to 8 branches run per delegation call.
Can a team member delegate to its own team?
No. Delegation is one level deep. Members answer the lead agent and cannot hand work on to further agents, which keeps cost and run time predictable.
What stops a multi-agent run from spending too much?
All branches of a run share one budget, 2 US dollars by default, and the run stops when it is used up. The figure is an estimate used as a runaway guard, not a billing amount. Each branch also has a 240 second time limit.
How is this different from a supervisor node in a workflow?
A team is decided by the lead agent at run time inside a conversation. A workflow fixes the steps in advance on a canvas, with supervisor, parallel and fan-out nodes for the same kind of split. Use a team when the agent should choose; use a workflow when you want the steps pinned down.
See it working in two minutes
The sandbox provisions a real tenant — describe an agent in one sentence and test it, no account, no card. Or browse ~90 industry workflow recipes to see what teams build.
