lou · bookkeeper
Watches your Stripe. Drafts the recovery email the moment a payment fails, matched to the decline and the customer.
Five AI coworkers. One job each. They turn on with one toggle, run inside guardrails you set, and leave an audit trail you can read. Drafts only by default; you approve before anything goes out.
Watches your Stripe. Drafts the recovery email the moment a payment fails, matched to the decline and the customer.
Reads every inbound email. Drafts replies in your voice. Files bug reports as Issues with reproduction steps.
Watches which customers have gone quiet and tells you why he thinks so. Drafts the check-in. Never sends it.
Runs your rhythm. The Monday brief. A monthly read on your market. A nightly pass over what the rest of the crew noticed.
Drafts the changelog you keep skipping, from the week's shipped work, in your voice. You publish it.
The crew runs in drafts by default. Here is the exact line each one won't cross without you — the same line the funded tools keep vague.
Reads every Stripe webhook and drafts a recovery email the moment a payment fails.
Sending. The recovery draft waits in your inbox until you approve or edit it.
Reads inbound support mail, classifies it, and drafts a reply from your own history and docs.
Sending. You approve or edit the reply before it reaches the customer.
Watches usage and support signals, flags accounts drifting toward churn, and drafts the outreach.
Reaching out. The at-risk flag and the draft are Hank's; the send is yours.
Gathers the week's metrics, computes the baselines, drafts the briefing, and posts it to your Today.
When a metric breaks its baseline, the briefing pauses for your review before it posts. Surprising numbers get a second look.
Drafts changelog entries and customer notes from what actually shipped.
Publishing. Nothing reaches a status page or a customer list until you say so.
Every draft and every action is written to a plain-English trail you can read: what they saw in the run, what they wrote, what they were about to do. High-stakes actions are cryptographically signed.
Kill switches are per-agent. Turn an agent off and it stops at its next step; your approval queue and audit trail stay exactly where they were.
Customer data is never used to train any AI model. It's a commitment in our privacy policy, not a setting you have to go find.
Drafts are the default across the workspace. You turn drafts-only off when you're ready, and some agents keep holding every send for your approval anyway, by design, whatever the setting.
Five is the count we ship at. Each agent absorbs more playbooks over time — Lou picks up the renewal and expansion work he doesn't do yet, Frankie gets more accurate at edge classifications, Hank learns finer churn signals. The team doesn't grow; the team gets sharper.
A sixth agent — a compliance officer for SOC 2 evidence and customer DPAs — joins if and when enterprise sales materialize. Inky will announce it in the changelog when they ship. No sooner.