Building your own hive

A hive is a named team of agents you can hand work to as a unit. You build it in two layers: pick the roles, then configure what each role knows and can do.

Compose the roles

Add one or more agent roles to the hive. For each, you can:

• Give it a role label within the hive (for example, "lead researcher" or "reviewer").
• Mark it required or optional.
• Set its order, which controls how roles are listed in the hive.

You need at least one role. Reuse a role across hives, or start from a pre-built hive and adjust its lineup.

Configure each role

A role is more than a title. Each one carries:

• Instructions — a system prompt that sets how the agent behaves and what "good" looks like.
• Skills — reusable step-by-step know-how (a SKILL.md plus optional scripts and reference files) you attach so the agent handles a repeatable job the same way every time.
• Memory — context packs that give durable, org- or role-scoped background the agent carries into every task, alongside the learnings it captures on its own.
• Tools and knowledge — which features and knowledge bases the agent can reach for, so a role only holds the access its job needs.

Add check-ins

Any agent can run on a heartbeat — a recurring prompt it answers on its own schedule (30 minutes by default) so it stays quiet unless something needs you. You set the interval, the prompt, and optional active hours and timezone so it only fires during your working day. See Heartbeats.

Building your own custom agents is available on the Pro plan — every plan comes with the pre-built hives ready to use.

Where it fits

• Use a hive on real work in Projects.
• Package a repeatable process for a hive with Playbooks.
• Give roles durable context in Memory.
• Automate ongoing check-ins with Automation.