AI agents for a more productive tomorrow
agentos.mt

agentos.mt / Documentation

Reference

How agentos.mt works

A reference for workspace owners and builders. The public API is not available yet; this page will document it when it is.

Workspaces

Everything in agentos.mt belongs to a workspace: agents, knowledge bases, conversations and usage. You can belong to several workspaces and switch between them. Each request checks your membership and role again on the server.

Roles

RoleCan
OwnerEverything, including billing and removing the workspace.
AdminManage the team and settings, publish and roll back agents.
BuilderCreate and edit drafts, manage knowledge, run tests. Cannot publish.
OperatorView agents and conversations; will handle handoffs when the inbox launches.
AnalystView agents and usage.
BillingView usage and, when paid plans launch, manage billing.
ViewerRead-only access to agents and knowledge.

Drafts, revisions and versions

Every time you save an agent, a new revision is stored. Publishing takes the current revision and the connected knowledge bases and freezes them into a numbered version with a fingerprint. Rolling back creates a new version with the same snapshot as an earlier one, so the history is never rewritten.

Publishing checklist

  • Instructions of at least a few sentences
  • An AI disclosure message and a fallback answer
  • A human escalation contact for HR, legal and finance agents
  • A successful test conversation on the current draft

Knowledge

Text entries (up to 60,000 characters) and .txt, .md or .csv files (up to 1 MB) are split into sections of about 900 characters and indexed for keyword search. For each question, up to four matching sections from the knowledge bases connected to that agent are passed to Aftaris as reference material, and the answer cites them by number.

How a test message is answered

  1. Your role and the monthly message limit are checked, and one message is reserved.
  2. Matching knowledge sections are found in the agent’s own knowledge bases.
  3. The platform rules, your instructions, the reference material and the recent conversation are sent to Aftaris.
  4. The answer, its sources, the time taken and the token counts are saved, and the reserved message is counted.

If Aftaris is unavailable, you see a clear message and the reserved message is released — nothing is invented.

Limits during early access

  • 5 agents, 2 live at a time
  • 2000 test messages a month
  • 5 knowledge bases, 100 sources, 50 MB
  • Up to 5 team members