Research on human–agentic interaction and swarm intelligence.
Products that put it to work.
Build the business that runs itself. Kyte is a desktop app where AI agents build your company's software — CRM, ERP, HR — then keep the work moving around the clock. It runs on your machine, and none of it is sent to someone else's cloud.
+--------------+-------------------------------+-----------------------+ | o o o KYTE.SYS | +--------------+-------------------------------+-----------------------+ | > Cowork | Invoice draft for Fabrikam | Wingtip — Webshop | | Extend | | AO-WR-25219 * active | | | | | | APPS | | CUSTOMER Wingtip | | CRM | | MANAGER M. Bowen | | ERP | | START 01.04.2026 | | HR | | END 31.10.2026 | | | | BUDGET 68 000 EUR | | SESSIONS | | | | Invoice… | | --------------------- | | Q3 review | | Budget (h) 160.00 | | Book hours | +-----------------------+ | Booked 17.5 | | | | Message your agent…_ | | ##............ 11 % | | admin@acme | +-----------------------+ | | +--------------+-------------------------------+-----------------------+
Describe your company once — every AI agent works by your rules.
Goals, roles, processes and policies as markdown in git. People read them. Agents follow them. A tool checks them like code.
---
api: company-as-code.org/v0
type: process
id: invoicing
title: Monthly invoicing
owner: role/ceo
serves: [goal/wholesale-growth]
policies: [policy/four-eyes-payments]
---
Trigger: 1st of each month. Steps:
agent drafts invoices from delivery
notes; ops reviews; send; log.
Done when: all invoices sent and logged.process/invoicing
|- owner role/ceo
|- serves goal/wholesale-growth
`- policies policy/four-eyes-paymentsrole/ceo
|- owns process/invoicing
|- owns process/customer-follow-up
`- owns process/weekly-roast-planpirol is a small German AI lab researching human–agentic interaction. We build tools that let people and agents share a desk — predictable, observable and safe. We believe the future holds not one model but many — a swarm of intelligence.
Intelligence is about to become abundant and affordable. We want to shape that future: AI systems that serve people, improve lives, and shrink the cycle time of innovation from years to days. Our path there starts at the desk — with agents people can trust to do their most sensitive work.
We are a small team — every hire shapes the lab. If you want to build at the edge of agentic AI on sensitive data, we want to hear from you.
Work with us →Pirol AI is a model of models. It picks the model that solves your job best — or you reach for one of the five best known yourself. Open-weight, hosted in the EU, through one OpenAI-compatible endpoint, with no provider key of your own.
For Developers →
Grischa Jacobs · founder
Everyone is waiting for the one model that can do everything. I think the answer is many models that know what they are doing: agents that can tell what a click will do before they make it, and can be trusted with real work because of it. Europe is sitting on the data that matters most, the data that runs its factories, its clinics, its workshops. Almost none of it has been opened up, because none of it can leave the building. Make agents safe enough to work in there, and the agent takes the busywork while the human keeps the decisions. Then the things we are still slow at start taking days instead of years: a new material, a better machine, a treatment. That is the part I want to be useful to.