Fig.2

Agent systems for people and teams

A private AI operating system for one operator, and Centaur, a self-hosted Slack-native agent platform for organizations.

Personal Agent Systems

Dashboards

pull contextact through toolsfile learnings

Centaur - Company/Multiplayer Agent Systems

API

Sandbox

Tools

channel scopepermissioned executionobservable output

Operating model

Same agent pattern, different boundary

Personal stack

Personal Agent Systems

Dashboards

pull contextact through toolsfile learnings

Context

Obsidian vault, agent memory, public surface, and private data feeds.

Operators

Runner as chief-of-staff, Codex for coding, and Cotypist at the input layer.

Surfaces

Dashboards, APIs, scheduled automations, filed learnings, and daily workflows.

Team stack

Centaur - Company/Multiplayer Agent Systems

API

Sandbox

Tools

channel scopepermissioned executionobservable output

Slack-native

Requests begin where the team already works, with channel-scoped context.

Self-hosted core

API, Postgres, sandbox, proxy, and observability run inside the company boundary.

Useful loops

Linear triage, code-collab follow-through, digesting, and scoped readouts.