Custom agents
Don’t see the agent you need? Build it.
Describe the signal, the reasoning and the actions it may take. No code, no services engagement, no waiting on our roadmap.
Four steps, start to running.
- 01
Describe it
Name the agent, pick the section it belongs in, and write the system prompt in your own words. If an existing agent already does something very similar, Clynto tells you rather than letting you build a duplicate.
- 02
Scope what it may do
Set the minutes it saves per action, a default tier, and per-segment overrides. The tier is the whole safety model - it decides whether the agent notifies, waits for approval, or acts.
- 03
Test it on real history
Run it in test mode against historical accounts, so you can see what it *would* have done on accounts you already know the outcome for. Nothing it proposes in test mode leaves the building.
- 04
Promote it when you trust it
Drop it into a workflow lane, or run it from an account. You can create one directly on the workflow canvas without navigating away, so nothing in progress is lost.
What you’re actually defining
The signal
What it watches: usage, tickets, CRM fields, email and calendar activity, contract dates.
The reasoning
What the signal means for this account, in this segment, at this stage of the journey.
The allowed actions
Notify, create a task, raise a signal, flag the account, draft an email, or update a field.
The audit trail
Every decision is logged with the reasoning behind it - the same as every built-in agent.
New agent
- Name
- Sandbox Decay Detector
- Section
- Health & Risk
- Watches
- Mixpanel sandbox vs. production events
- May do
- Raise signal · Create task
- Strategic
- Notify only
- Core
- Suggest & approve
- Longtail
- Fully autonomous
A custom agent gets exactly the same leash.
It runs through the same tier system as every built-in agent, so nothing you build can quietly acquire more authority than you granted it.
Notify only
You get an in-app notification. Nothing is done.
Suggest & approve
The action waits in the Approvals queue for a human.
Fully autonomous
It executes immediately, and is audit-logged.
Three teams have already built
Illustrative, not customer data - but each is the kind of gap a built-in library never quite covers.
Security Review Watcher
Watches for accounts whose annual security review is inside 60 days and drafts the questionnaire request to the right contact.
Sandbox Decay Detector
Flags accounts whose sandbox usage has fallen while production is flat - an early sign the team has stopped building.
Invoice Chase Agent
Picks up a failed payment, checks whether the account is otherwise healthy, and routes to finance or to the CSM accordingly.
It slots into the library you already have
Custom agents appear alongside the built-in ones - in the agent dropdown on the workflow canvas, in the library, and in the audit log. Pick which of the nine sections it belongs to.
- Health & Risk
- Renewal
- Growth
- Relationship Risk
- Onboarding
- Email & Communication
- Reporting
- Automation
- Generic Ops
Build the agent your team keeps asking for.
Start free for a week and ship one before the trial is out.