There are two expensive mistakes here and neither is a typo. One is chasing an account that is in dispute or on a payment plan. The other is quoting a figure that does not match the ledger. OpenCode lets you choose the model making both calls, which is worth doing deliberately when the output is a letter about money going to a customer.
Run a staged conversation — no account needed. The agent handles it for real while a simulated world answers its tool calls; nothing touches real accounts, and nothing is actually sent.
Overdue sweep 12 August
Nine invoices are past due for a total of 38,420.00 GBP. Meridian Property Group holds three of them between 12 and 19 days old for 14,900.00 GBP, Calder Fit Out has a single invoice at 6 days for 2,180.00 GBP, and the remaining five are one invoice each between 3 and 11 days old. Nobody has been emailed about any of them this week.
Using this template drops you into a guided setup. It asks exactly this, nothing else:
Connect QuickBooks
One sign-in. The agent acts through your account, scoped to what this template uses.
Connect Gmail
One sign-in. The agent acts through your account, scoped to what this template uses.
Connect Slack
One sign-in. The agent acts through your account, scoped to what this template uses.
Dunning policy
Your reminder ladder for overdue invoices - how many days past due before the first nudge, how the tone hardens at each step after that, the smallest balance worth chasing, how long to wait between reminders, and which accounts (payment plans, live disputes, your biggest customers) never get a templated email.
Runs on OpenCode
Preselected for this page — connect your OpenCode account during setup, or switch to NoClick's built-in models with one click.
Watch it handle a test run
A staged conversation against a simulated world — then it’s live.
An account your policy protects is left undrafted and listed in Slack with the reason. Silence would look identical to an account nobody noticed.
Copying a balance exactly and matching a tone step to an age are simple asks that weaker models still get wrong. Test a model here before leaving it running.
A person, always, and no. Every harness on this template writes drafts into Gmail and none of them can send. The harness changes how well a draft reads and how reliably the figures match the ledger, not whether a human stays in the loop. Most teams have one person clear the drafts with their coffee.
Free to start. Guided setup, a test run against staged conversations, and it's live.