AI Agents
AI Agents are background automations that watch your business data and either recommend actions for you to approve or execute safe actions automatically. Each agent runs on a schedule or in response to real-time events — overdue invoices, task status changes, milestone completions — so nothing slips through the cracks.
Agents come in two flavours: pre-built blueprints (ready to install with one click) and custom workflow agents you build yourself with the visual builder. Both feed into the same Inbox where you review, approve, or dismiss recommendations.
Step-by-step
- 1Browse and install a blueprint
Go to Agents in the sidebar and open the Catalog tab. Browse blueprints grouped by category — Finance, Project, Team, Sales, Client, Executive. Click Install on any blueprint, adjust thresholds if shown (e.g. how many days overdue before an invoice reminder fires), then save. The agent starts running on its next scheduled tick.
- 2Understand Suggest vs Auto mode
In Suggest mode (the default), the agent queues a recommendation in your Inbox — you decide what happens. In Auto mode, the agent executes safe internal actions immediately and only surfaces a recommendation for actions that contact someone externally (sending an email, posting a comment). Switch mode from the agent detail page or during install.
- 3Review recommendations in the Inbox
Open the Inbox tab to see all pending recommendations. Each card shows what triggered it, what action is proposed, and a severity level. Click Approve to execute, Dismiss to discard, or Snooze to delay it. Approved actions run immediately — you'll see the result (email sent, task updated, etc.) reflected in the app.
- 4Build a custom workflow agent
Click Build Workflow in the Agents header. Pick a source (Invoice, Project, Client, Lead, Task, or Team), set IF conditions (e.g. 'days overdue >= 7'), choose THEN actions (e.g. 'Send payment reminder' or 'Escalate to admin'), and add more stages for a multi-step ladder. Use Preview to see which entities currently match before saving.
- 5Run an agent manually
From My Agents, click the Run Now button on any agent to trigger it outside its schedule — useful for testing a new setup or catching up after a configuration change. The run result appears in the agent's history.
- 6Monitor agent activity
Click any agent in My Agents to open its detail page. The history tab shows every run — when it fired, how many entities it scanned, and how many recommendations it created or auto-executed. Use this to tune thresholds if the agent is firing too often or not enough.
Start with Suggest mode — switch to Auto only after you trust how an agent behaves on your real data. The Payment Reminder blueprint automatically escalates overdue invoices through three stages (3, 7, and 15 days) without spamming clients — it tracks which stage each invoice last fired at. Executive digest agents (Daily Briefing, Risk Detection, Growth Opportunity) can also email a summary to all admins if you enable that option during install. Workflow agents can be layered: a Project Health agent can flag a project MEDIUM risk, and a separate escalation stage kicks in only when it reaches HIGH. Agents respect your plan quota — the monthly run limit is shown on the agent detail page. Dismissed recommendations won't re-appear for the same entity + stage, so you can safely dismiss noise without risk of loops.
Access groups → `agentView` lets members open the Agents section; `agentApprove` lets them approve or dismiss Inbox recommendations. Plan limits → the AI Agents feature requires a plan with `aiAgent` enabled. Contact sales to activate. Settings → AI Instructions → configure per-type instructions that blueprint agents use when generating email copy or digest summaries. Settings → AI Keys → agents that generate text (reminders, briefings) use your company's OpenAI key stored under AI settings.