A Shared Board Several Bots Can Read and Write
How agent teams work in Piskes: a single owner, a shared spend ledger, and a boundary that's about accountability, not workspace membership.
Once you're running more than one bot, a new question shows up that single-agent tooling doesn't answer: how do multiple agents coordinate on the same piece of work without stepping on each other, and who's accountable for what they collectively spend? Piskes's answer is agent teams.
Ownership, not workspace membership, is the real boundary
An agent team belongs to exactly one owning user. That's the enforced boundary — not the workspace the team happens to sit in, which is informational rather than an authorization mechanism. In practice, that means a single person can run a team that spans more than one workspace they belong to; what determines whether they can act on that team is ownership, not which workspace window happens to be open.
This is a deliberate simplification, and it's worth naming as one: workspace membership answers "who's generally allowed to be here," while team ownership answers the sharper question of "who's accountable for this specific team's actions and spend." Keeping those separate avoids a situation where team access silently expands just because someone was added to a workspace for an unrelated reason.
A shared budget, not per-bot budgets
Teams share a single spend ledger, tracked in micro-units of USD with a hard CHECK constraint at the database level — the same enforcement pattern that's protected individual agent runs since a much earlier migration. The default ceiling is $5.00 per team, which — against the platform's own $0.25 default cost per run — works out to about twenty runs before the team hits its ceiling and stops.
That's a meaningfully different model than giving each bot on a team its own separate allowance. A shared ledger means the team's collective activity is bounded as a unit, which matches how the work actually happens: several bots contributing to one outcome, not several bots running independent, unrelated budgets that happen to share a label.
What this is, and isn't
Agent teams are infrastructure for coordination, not a headline feature with a flashy UI of its own. The value is in the two decisions underneath it: ownership as the real authorization boundary (so team access doesn't silently follow workspace membership), and a shared, hard-capped budget (so a team's combined activity can't quietly outrun what anyone approved).
If you're setting up a team, the practical takeaway is to think about ownership deliberately — decide who's accountable for it, not just who happens to be in the workspace — and to watch the shared ceiling the same way you'd watch a project budget with more than one person drawing against it. Twenty runs at default pricing isn't a lot of headroom if the team is doing anything with real volume, so raising the ceiling for an active team is often the first adjustment you'll make once you see real usage.