RestartiX Platform

English

Română

English

Română

Appearance

AI Agents as First-Class Actors ​

Stated platform direction. This is not a feature spec — it's the commitment that shapes foundation decisions.

The commitment ​

The RestartiX platform is built around the assumption that AI agents are first-class actors, not tools that humans call. Concretely:

  • An AI agent has its own identity in the system, not a shared "AI" account.
  • An AI agent has its own role grants and permissions, scoped per organization.
  • An AI agent's actions are audited with full provenance: which model, what inputs, what confidence, what was decided.
  • An AI agent can act under human delegation ("doctor X authorized this agent to schedule follow-ups") or autonomously within scope ("monitor patient adherence overnight, escalate to the on-call specialist if X").
  • An AI agent's writes are governed by the same RLS, audit, and compliance rules as any other actor — not a separate code path.

This is a stated commitment to a 12-24 month direction. It is not a bet on a specific AI feature shape — clinical drafting, autonomous monitoring, triage, decision support, transcription — those are defined per-feature when they ship. The commitment is to the actor model, not to the feature roadmap.

What this means for the foundation ​

The actor model lives in the schema as principals as the root identity. Audit, RLS, RBAC, and every domain reference principals.id. Humans, AI agents, integration service accounts, and system jobs are all principals — humans are not privileged in the schema. See architecture/decisions.md → Why principals as the root identity for the design.

What this does NOT mean ​

  • Not "AI-first architecture." The substrate (Postgres + RLS, Go monolith, repo/service/handler split) is unchanged. AI agents are callers of the same services, governed by the same rules.
  • Not "we will ship autonomous AI tomorrow." Specific AI features land per their own feature specs, with regulatory scope (EU MDR / GDPR Art. 22) decided per feature.
  • Not a license to add AI-shaped abstractions everywhere. Foundation work supporting this direction must still meet cost asymmetry — see CLAUDE.md → Foundation work vs. no-speculation.

See also ​