Work

A build kit for AI agents that survive a crash mid-task, stop at their budget and touch nothing without your approval.

Crash test

Kill the container mid-task and the run resumes

Approvals

Every action set to always, first time only, or never

Spend

Capped, and broken down per agent and model

One tunnel for your whole dev team: a tester circles a bug and writes one sentence, an agent fixes it and posts the after-shot.

Findings answered

148 from 37 sessions, across 3 PRs

Tester time

3.9 h

First commit to v5.10

About 12 weeks

Booking, payments and smart locks — a venue that runs 24/7 with no staff.

Door access

A time-boxed PIN for every booking

Embeds

A booking widget for React, Vue, Svelte and Angular

In development since

2023

Like Wolt, but for home services — AI diagnosis, vetted pros and BankID-signed contracts.

Backend

89 API modules

Platforms

iOS, Android, web and admin

First line of code

5 May 2025

Prague launch

1 Aug 2026

A real-time control plane between the energy market and the machines on site.

Measured

Command response times, continuously

For a Czech financial institution’s developer teams — measured before and after.

Benchmark

15 golden tasks

Quality

78.5% before, 95.2% after