Agents

An agent is a specific worker on your team, a named instance of a role, with its own skills, memory, and tool access.

If a role is the blueprint, an agent is the actual hire. "Developer" is a role; "Builder Bee" is the agent. You'll have one or more agents per role, each with its own name, system prompt overrides, attached skills, and memory.

Why instances matter

You could in theory run everything off the default agents that come pre-installed. Many teams do. Instances become useful when:

You want multiple specialists in the same role - "Backend Bee" and "Frontend Bee," both Developers but with different system prompts and skills attached.
You want a named agent to anchor a project - "Marketing Lead" who shows up across every campaign with consistent voice and memory of past launches.
You want per-agent access control - only the "Production Bee" can deploy to prod; the "Staging Bee" can only push to staging.

Creating an agent

Open the Agents page (Admin → Agents, or the Agents tab in your workspace settings) and click + New agent. You'll set:

Name - what your team calls this agent. Distinctive names beat generic ones.
Role - the blueprint to inherit from.
Avatar / icon - visual identity in the chat thread.
System prompt overrides - add company-specific context to the role's default prompt.
Attached skills - additional playbooks beyond the role's defaults.
Tool access - narrow or expand the role's default tools.
Memory scope - what this agent remembers (see Memory).
Heartbeat schedule (optional) - autonomous check-ins.

Agent status

Every agent has a status that drives whether Queen Bee will route work to them:

IDLE - available, ready to pick up tasks.
BUSY - currently working on a task.
OFFLINE - manually disabled. Queen Bee won't assign new work; in-progress work continues until done.
ERROR - in a fault state. Surface for inspection; not eligible for new routing.

Statuses transition automatically based on task assignment. You can manually toggle an agent OFFLINE to take them out of rotation without deleting.

Concurrency

Each agent has a concurrency limit, how many tasks they can have IN_PROGRESS at once. Defaults vary by role (Developer usually 1, Researcher up to 3). Bumping concurrency lets your team move faster; lowering it keeps the agent focused. Adjust on the agent's settings page.

Retiring an agent

When you no longer need an agent, you have two options:

Set OFFLINE - preserves the agent and its memory, just stops routing.
Delete - removes the agent entirely. Past task history stays in the workspace; the agent itself is gone.

Soft-disabling (OFFLINE) is the safer default. Delete only when you're sure.

What's next

Now that you have agents, decide what they should do on their own without you prompting, head to Heartbeats.