AI automations where a person approves what matters.

We build AI automations and agent integrations for small businesses. This line takes paid work today.

What we build

  1. 01Intake and triage of messages and leads.
  2. 02Document question-answering with citations.
  3. 03Scheduling agents.
  4. 04Read-only connections between an AI assistant and your business database.
  5. 05Checkups of a chatbot you already run.

How every automation is wired

Each build also goes through our adversarial review loop before it ships.

See how our review loop works

  1. The model suggests.
  2. Plain code decides the route.
  3. Anything that acts on a customer waits for a person to approve it.

Seven sample builds, with their test numbers

These are sample builds on synthetic data. None of them is client work. Each one was tested on its own set.

  1. 01Lead to CRM.A person approves before any reply is sent.
  2. 02Document Q&A with citation checking.12/12 on its test set.
  3. 03Scheduling agent with staff approval.4/6 on a local model. Both misses were stopped at the approval step.
  4. 04Chatbot checkup.16 fixed questions, judged by code. A typical bot scored 7/16. Ours scored 16/16.
  5. 05Read-only database connection for an AI assistant at a clinic.Private columns are kept out of its reach. 24 attack queries behaved as expected. The local model answered 5/8.
  6. 06Website-form intake triage.Urgent messages are flagged to staff. 13/13 on its set and 7/8 on a held-out set.
  7. 07Message redaction and triage gateway.Routed right: 21/23. Planted private values leaked: 0/29.

Results on synthetic data are not a promise about your data. That is one more reason a person approves the steps that matter.

Price and how to start

Each project is quoted separately. There is no price list.

Email us what you want automated. We also take work through freelance platforms.