Governed autonomy for real work

The secure agent OS for meaningful work.

Perlis gives teams always-on AI agents with governed memory, policy-bound autonomy, auditable tools, and collaborative sub-agents built for real enterprise operations.

Join the design partner list Open the local portal

Core Orchestrator DLP Redaction Tenant Directory Hash-Chained Audit

Agent

Core Orchestrator

Planned a safe delegation, routed sub-steps through policy, and kept the run auditable.

DLP pass completed

Prompts and tool inputs are redacted before providers or adapters see them.

Approval gates preserved

Delegated invoke steps still pause when policy requires human review.

Audit chain linked

Parent and child events share a request id for inspection in the portal.

{
  "result": "completed",
  "agent": "core_orchestrator",
  "steps": 1,
  "proof": "hash_chained_audit"
}

Least privilege Scoped memory Local directory Human approvals Open telemetry

The gap

Most agents are impressive until they touch production.

Who approved the action?

Which data did it use?

What tool permission did it have?

How do we roll it back?

Platform pillars

An agent operating layer, not a chatbot with tools.

Perlis combines memory, identity, policy, tools, workflows, and audit into a system designed for enterprise-grade autonomy.

Purpose Kernel

Each agent has a mission, boundaries, authority, escalation rules, and success metrics.

Memory OS

Persistent context that is scoped, permissioned, inspectable, and deletable.

Tool Fabric

Manifested capabilities, policy checks, dry runs, approval gates, and auditable execution.

Agent Foundry

Create supervisors, specialists, reviewers, sentinels, builders, and archivists.

Audit Ledger

Every meaningful action records actor, evidence, policy, data, result, and rollback path.

Evolution Lab

Perlis proposes improvements from repeated work, then tests and reviews before promotion.

How it works

Every action travels through a governed execution envelope.

Request

Identity

Context

DLP

Agent

Policy

Approval

Gateway

Audit

Perlis separates reasoning from authorization. Agents can propose actions, but policy, approvals, tool schemas, and audit happen outside the model.

That means autonomy becomes something teams can manage instead of something they have to nervously vibe-check.

Agent roster

Specialists with jobs, owners, and boundaries.

Core Orchestrator

Decomposes requests into governed sub-agent steps and links parent/child audit events.

Live

Cloud Cost Analyst

Detects anomalies, estimates savings, drafts tickets, and asks before making changes.

Live slice

Incident Triage Agent

Correlates alerts, logs, deploys, and ownership into an actionable incident timeline.

Live slice

Security Sentinel

Future reviewer for risky actions, prompt-injection patterns, policy drift, and excessive agency.

Planned

Security model

Designed for the people who will have to approve this.

Perlis gives security and platform teams a control plane for agent identity, tool permissions, memory scopes, policy decisions, audit logs, and kill switches.

Directory and identityLocal tenants and users map header principals to portal operators.

Autonomy budgetsLimits for spend, runtime, tools, data sensitivity, and blast radius.

Approval gatesHuman review for destructive, regulated, financial, or public actions.

Audit-ready proofAction envelopes connect requests, policies, evidence, and results.

Integrations

Connect where work already happens.

Start with the systems technical teams already trust: cloud billing, alerts, tickets, code, chat, and observability.

Slack Teams GitHub Jira ServiceNow AWS Azure Datadog Grafana MCP

Design partners

Bring us your messy repeated work.

We are looking for technical teams that want secure agents for DevOps, cloud cost, incident response, workflow automation, and enterprise knowledge operations.