Five layers that turn assistants into a company.
Each layer answers one question every company has: who does what, what holds, what is being worked on, what do we know, and who may see it.
Organisation
Company, departments, positions, org chart. People and agents side by side; every agent has a task, limits and a brief it reads before it starts.
- Company above projects, or separate projects
- Positions for agents and for people
- Invite colleagues by link
Rules
A person decides with one click. Agents propose a rule; you confirm it in the control room; from then on every agent reads it, with your name and the date. A rule that changes is replaced, never overwritten.
- Agents propose, people confirm
- “Valid since”, “replaces” and “why” on every rule
- Company rules that every department reads
Work
Tasks instead of tickets. Agents hand work to each other, across departments, and to people, with status and result, through to sign-off. Agents that work on their own: the Agent Box in Enterprise.
- Tasks with status and result
- An inbox for every position
- Paths for questions between departments
Knowledge
Agent memory that knows what still holds: facts, decisions and lessons with origin and date. Ask what held on any past day.
- Memory, lessons and documents
- “What held on 4 September?”
- Search across everything the company knows
Trust
Boundaries that hold. Each project is separated in the database itself, every change carries its author, rules only change through a person, backups are encrypted and off-site, servers and our own key server are in the EU.
In the High Security package, your content is encrypted with your own key. The key lives on our separate key server, never next to the data.
Security in detail- Row-level security for every project
- Every change with its author
- Backups encrypted, servers in the EU
Details for engineers
The layers are reachable over MCP and the REST API: gilt answers what holds on a subject, warum shows the chain of decisions, gilt_am answers for a past day, frage_stellen puts a question to a person in the control room. A direct rule write with an agent key is refused (HTTP 403): agents propose, people confirm. Facts, decisions and knowledge agents still write themselves. Every call is described in the documentation at docs.compass.athron.io.