> ## Documentation Index
> Fetch the complete documentation index at: https://docs.superember.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Agents, Team Agents, and Fleet

> Create fully configured nested AI teams and manage desired state safely.

Creating a team creates its Team Agent in the same operation. `teamAgent` accepts identity, profile/role brief, model, workspace source, timezone, access, skills, MCP references, app bindings, channel intents, knowledge references, and provisioning intent. Creating an agent requires a same-team `parentAgentId`; that parent can be the Team Agent or another active subagent, so agents may construct and configure their own granted descendant group.

Fleet manifests provide portable desired state with stable `ref` values. They support nested `children`, but never credentials. Run `fleet plan` first, inspect its exact changes and risk, then apply the unexpired plan. High-risk pruning requires an organization-admin approval flag. If an apply fails partway, retry the same plan with a new operation key; SuperEmber accepts it only when current state exactly matches the recorded recoverable partial state.

Operational reads are available through `/v1/activity`, `/v1/pulse`, `/v1/agents/{agentId}/conversations`, and `/v1/agents/{agentId}/approvals`. Attention acknowledgement and delegation delivery/startup recovery use idempotent mutation endpoints. Retrying an uncertain delivery or confirming an uncertain startup requires the exact observed timestamp and explicit duplicate-risk confirmation, so stale dashboard or CLI actions cannot resolve a newer incident.

```yaml theme={null}
apiVersion: superember.ai/v1alpha1
kind: Fleet
prune: false
teams:
  - ref: revenue
    name: Revenue
    slug: revenue
    teamAgent:
      ref: revenue-lead
      identity: { name: Rowan, title: Revenue lead }
      provision: true
    agents:
      - ref: prospecting
        identity: { name: Scout, title: Prospecting }
        profile: { roleBrief: Find and qualify target accounts. }
        runtime: { model: openai/gpt-5, timezone: Europe/Berlin }
        provision: true
        children:
          - ref: enrichment
            identity: { name: Atlas, title: Data enrichment }
            provision: true
```