Frequently asked questions

Basics

Is my data used to train any model?

No. Conversation content, knowledge files, and tool data stay inside your organisation. We don’t share them with third parties for training. See our privacy policy for the details on what we collect and how it’s used.

What happens if an agent makes a mistake?

For irreversible actions (customer replies, destructive operations), the answer is “a human approved it first.” The approval gate sits between the agent’s draft and the action landing. For reversible mistakes (a wrong draft, a misread summary), you can re-run the task with corrected instructions or fix the artifact directly.

Can I undo a deletion?

Yes. Every delete in AgentTeams is optimistic with a 6-second undo window. Click Delete, a toast appears with an Undo button counting down; click Undo any time before the timer expires and the item comes back.

Agents

How many agents can I hire?

Depends on your plan. The Free plan lets you hire one agent (your PM) plus one specialist; the paid plans increase the cap. See pricing for the current numbers.

Can I rename an agent after I hire them?

Yes. From the agent’s detail page, click Edit on the identity card. Name, avatar, role, and team can all be changed. Renaming doesn’t break anything: tasks, knowledge, and tool connections all follow the agent.

What happens if I fire an agent?

Firing an agent removes them from your roster, ends their tool connections (their tokens are deleted), and moves their active tasks into a holding state where you can reassign them. Their chat history and audit log entries are kept for the audit trail. Firing is reversible (with the undo window) but not retroactive: if you confirm and the timer runs out, that’s final.

Reactions

Will an agent reply automatically to my customers when I connect Gmail?

No. The opt-in rule holds: reactions are blocked by default, and the safe defaults we seed (email from a teammate, internal calendar invites, etc.) are scoped to internal traffic. External customer email is in the catalogue, but ships as opt-in only. You have to explicitly create a task subscribed to gmail.inbound.external for the agent to react to outside email. See the full catalogue on the Reactions page.

Can I have an agent react to a webhook from a tool you don’t support yet?

Yes. Pick the Custom Webhook option in the New Task form. We mint a unique URL you can paste into any third-party (Zapier, Stripe, GitHub, a custom script). Every POST to that URL fires the task with the request body as context. Optional HMAC signing is supported for senders that need it.

Why didn’t my reaction fire?

Three usual causes:

• No matching task. The agent only reacts to events a task subscribes to. Check the agent’s Tasks tab to confirm the Reaction task is Active and not Paused.
• Filter mismatch. If your trigger uses sender_domain: internal_only and the event came from outside your company, it’ll be filtered out. Same for mention-required triggers when nobody @mentioned the agent.
• Token revoked. If the integration disconnected (the nightly health check will have flagged it), no events flow in. Reconnect from the agent’s Tools tab.

Tools

Can two agents share a single connection?

No. Connections are per agent, on purpose. Each agent acts under its own identity, so the audit trail in your upstream tool stays clean (a Slack message says Mira sent this, not some-shared-bot sent this). If you want two agents to cover the same surface, connect each separately and use a team to coordinate routing.

Does AgentTeams work without any tool connected?

Sort of. You can talk to your agents in the dashboard, create tasks, build a knowledge base, and the PM will coordinate things. But the agent can’t reach out to anyone or pull in real-time data until a tool is connected. Most people connect Slack and Google Workspace within the first session.

Billing

What’s in the Free plan?

One PM agent, one specialist, all integrations, all reactions. The cap is on number of agents and number of active tasks per month. See pricing for the current limits.

How do you bill for agent activity?

Per agent on the roster, not per message or per token. Hiring an agent counts; the work they do doesn’t get itemised. This means you can run an agent hard without watching a meter.

Still stuck?

Drop us a note and a human will reply. We don’t use a chatbot for support (the irony is not lost on us).