Pricing
Priced for a small book of invoices.
Two plans. Start on Free with demo data, and move to Pro when you want the easy reminders sent for you.
Free
$0
For a handful of overdue invoices and a careful first look.
- Up to 10 active invoices
- Manual approval of every message
- CSV import
- Demo data to try it on
Pro
Test mode$29/month
For when the approval queue is the slowest part of your week.
- Unlimited invoices
- Auto-send step-one reminders at 0.80 confidence or higher
- Stripe invoice sync
- Weekly digest email
Test mode. Checkout runs on Stripe’s sandbox; no card is charged.
Side by side
| Feature | Free | Pro |
|---|---|---|
| Active invoices | 10 | Unlimited |
| Approval | Every message | Every message, or auto-send step one |
| Import | CSV, manual, demo data | CSV, manual, demo data, Stripe sync |
| Reply reading | Included | Included |
| Audit trail | Included | Included |
| Weekly digest email | Not included | Included |
| Price | $0 | $29/month, test mode |
Questions
Does it send without asking?
Not on the Free plan. Every message waits in your approval queue until you approve it, edit it, or reject it.
On Pro you can let step-one reminders go out on their own when the draft’s confidence is 0.80 or higher. Firmer steps still wait for you unless you change the autonomy setting yourself. A draft that fails a guardrail check drops to zero confidence and always waits.
What does it never do?
- Threaten, or make legal claims. Those phrases are blocked before a draft is saved.
- Contact a customer marked do-not-contact, or one who asked to unsubscribe.
- Send outside your send window: weekdays, 08:00 to 18:00 in your timezone.
- Touch the same invoice more than once in five days.
- Keep chasing once an invoice is paid, or while a reply is waiting to be handled.
Where do replies come from in the demo?
From the Demo Inbox. Paste a reply yourself, or press “Simulate reply” and a model writes a plausible customer response, clearly labelled as simulated. Outgoing email lands in an in-app outbox by default, so nothing reaches a real customer while you try it.
Which models does it use?
Groq-hosted gpt-oss writes the drafts and classifies replies. Gemini Flash-Lite is the fallback when Groq is unavailable. Every call is logged with the model, prompt version, token counts and latency.
Is my data used for training?
Dunnit does not train models. The public demo uses synthetic data only, and the free Gemini tier is never given real customer data. Please don’t enter real customer details into the demo.