Handler for Admins
A tour of the dashboard concepts: what an agent profile is, how members fit in, and where governance, approvals, and audit show up in the UI.
The five things to understand
Five concepts cover everything in the dashboard. Skim this once and the rest of the UI maps cleanly.
- Agent Profiles & Instances — what an agent is in Handler.
- Superpowers — bundled capabilities you grant to a profile.
- Connected Services — your own accounts (Gmail, GitHub, Slack…) the agent can act through.
- Governance & Approvals — the rules and the human-in-the-loop queue.
- Members & Audit — real people, and the activity trail tied to them.
Agent Profiles & Instances
An agent profile is the definition of an agent: its name, the superpowers it can call, the connected services it's allowed to touch, and the rules that govern it (auto / approve / block per action, danger-zone toggle, allowed services).
A profile by itself doesn't run anything. To put it to work, you assign it to a member. That assignment creates an instance — the specific copy of that profile used by that specific person.
Where you'll see this in the dashboard:
- Agent Profiles page — create and configure profiles.
- Templates page — pre-built starting points (e.g. Cautious / Balanced / Power risk classes) you can spin profiles out of.
- Members page — assign profiles to members. Each assignment is one instance.
- Overview → your instances panel — the instances belonging to the currently-signed-in member.
Superpowers (in brief)
Superpowers are the bundled, ready-to-use capabilities (research, intel, monitor, finance, marketing, audio, generate). When you create a profile, you choose which superpowers it can call. Each call costs a flat $0.005 and is governed by your rules.
Full reference: Superpowers. Pricing: Pricing.
Connected Services (in brief)
Connected Services are your accounts — Gmail, GitHub, Slack, HubSpot, Notion, and 200+ others — that you connect via OAuth. Once connected, each service appears as its own peer-level tool alongside the superpowers in the agent's tool list. The agent acts on your behalf, governed by the same rules as superpowers.
Full reference: Connected Services.
Governance & Approvals (in brief)
Three layers of governance:
- Allowed services — which superpowers and connected services a profile can touch at all.
- Permission defaults — per-action policy: auto, require-approval, or block.
- Spend caps — org-wide monthly cap and pre-paid balance.
When an agent action requires approval, it lands in the Approvals queue and pings the configured channel (dashboard, WhatsApp, Telegram, MCP). An admin approves or rejects; the agent is notified and continues.
Full reference: Governance.
Members & Audit
A member is a real person in your organization. Members come from one of two sources:
- Identity Provider sync — connect Google Workspace (or another supported IdP) and Handler imports users and groups. Group membership maps to assignments.
- Email invite — invite teammates directly when no IdP is in use.
Every action — every superpower call, every connected-service call, every approval decision, every rule change — is logged against the member that triggered it. The Activity table on the Overview is your audit trail. Filter by agent, by member, or by outcome.
Where you'll see this in the dashboard:
- Members page — list, invite, deactivate, manage assignments.
- Identity Provider page — connect Google Workspace and configure sync.
- Groups page — group-based assignment and rule overrides.
- Overview → Activity table — full audit trail with filters.
- Usage page — spend broken down by agent and member.
The Approvals queue
When a profile rule says "require approval" for an action — or when an agent tries to spend more than its delegated tier allows — the call is held and an approval request lands in the queue.
Approvals reach admins through any of:
- Dashboard — the Approvals page, with full context (who, what, why, cost estimate).
- WhatsApp / Telegram — inline approve/reject buttons. Configure in Settings.
- MCP — orchestrator agents and master agents can approve programmatically using
handler_approvals+handler_decide.
Full reference: Governance → Approvals.