Sign-in, sign-up and passkeys are live. What's new

paice
The platform

Six areas that interlock.

A canvas for flows, agents with a mandate, connected systems whose actions are classified, and an audit trail that records every step.

Draw processes on a canvas.

Triggers, tool steps, model steps, knowledge lookups and branches, drawn with the mouse. A process stands on its own: several agents can run the same one, and each brings its own accountability.

Autonomy in five levels.

Every agent has a level from 1 to 5 and a named accountable person. The level belongs to the agent and therefore holds everywhere — including in processes somebody else built.

Connect systems over MCP, every action classified.

Over MCP. On connect, the platform reads which actions a system offers and classifies each one: reads only, changes data, irreversible. Values in the call can raise the classification — an amount above a threshold turns “changes data” into “irreversible”.

Make your own documents searchable.

Documents are split and embedded locally. Agents look things up before they judge. If the embedding model is unavailable, the search falls back to words and says so in the result.

Decisions stay with the people accountable for them.

Whatever stops lands in the inbox of the responsible person. The agent works, you decide — including runs you started yourself. Where dual control is required, it can be switched on; a second person then approves. The server enforces both.

Every run fully traceable.

Which step ran, which branch was taken and why, where it stopped, who decided and on what grounds. In plain language, readable for audit.

Autonomy

Authority in five levels.

The level belongs to the agent, not to the flow. It therefore holds everywhere — including in processes somebody else built.

  1. Level 1AssistiertSchlägt vor, führt nichts selbst aus.
  2. Level 2BegleitetFührt aus, jeder Schritt wird bestätigt.
  3. Level 3ÜberwachtArbeitet selbstständig, Schreibzugriffe brauchen Freigabe.
  4. Level 4Weitgehend autonomHandelt im Budget allein, meldet Unwiderrufliches.
  5. Level 5AutonomHandelt allein, Bericht im Nachhinein.

The core

Approvals follow from consequence and autonomy level.

From the consequence of an action and the autonomy level of the agent follows whether a step runs or stops. To dissolve a stop you raise the agent's level — and that is exactly what the audit trail then records.

This is how the engine decides. The rule itself, running.

Every action of a connected system is classified

liest nur
runs from level 1
ändert Daten
runs from level 4
nicht rückholbar
runs from level 5

Values in the call can raise the classification — an amount above a threshold turns “changes data” into “irreversible”.

Triggers

How a flow starts.

Auf Zuruf

Ein Mensch startet den Ablauf, wenn er ihn braucht.

Zeitplan

Wiederkehrend — täglich, stündlich oder nach eigenem Muster.

Webhook

Ein fremdes System ruft eine Adresse auf und startet damit den Ablauf.

Neue Nachricht

Sobald etwas im überwachten Postfach eintrifft.

Neue Datei

Sobald eine Datei in einem beobachteten Ordner landet.

Neuer Beleg

Sobald im angebundenen System ein Vorgang entsteht.

See for yourself

A configured installation you can operate right away. It runs on your machine.

Book a demo