MVP FOUNDATION IMPLEMENTED / VALIDATION ACTIVE

AURA / PERMISSIONED AI OPERATING SYSTEM

Not another assistant.
An operating system for authorized intelligence.

Aura AIOS turns a bounded objective into a reviewable mission graph, executes approved work through specialized agents and returns evidence, cost and audit history without surrendering human control.

SYSTEM SHAPEMODULAR CORE + ISOLATED WORKERS
CONTROLPLAN + ACTION APPROVALS
MEMORYTENANT-SCOPED RETRIEVAL
STATUSLOCAL MVP / NOT RELEASED
THE AIOS THESIS

From prompts
to bounded missions.

The first release proves safe orchestration—not maximal autonomy. A professional or small team defines an objective, constraints, approved sources, budget and deadline; Aura plans the work, pauses at consequential decisions and preserves the evidence needed to review the result.

  1. Human-reviewed task graphs instead of invisible agent loops.
  2. Durable execution with pause, resume, cancel and recovery controls.
  3. Default-deny tools with short-lived, least-privilege grants.
  4. Provider-neutral models and replaceable infrastructure interfaces.
  5. Evidence, provenance and unsupported-claim marking by design.
MISSION ENGINEMEMORY ENGINEMODEL ROUTERPOLICY GATE
AURA KERNEL

Intent becomes
an accountable graph.

01DRAFT
02PLANNED
03APPROVAL
04QUEUED
05RUNNING
06VERIFYING
07RESULT BUNDLE
08AUDIT RECORD
HUMAN AUTHORITY — High-impact actions remain subject to explicit scopes, tool permissions, approval gates and auditable results. Persistence does not mean unrestricted autonomy.

One kernel.
Many replaceable systems.

UX

Mission console

Mission authoring, plan review, approval inbox, live timeline and result views.

ID

Identity & policy

OIDC identity, tenant context, role checks, approval proofs and action policy.

WF

Durable orchestration

Mission state, deterministic task graphs, timers, retries, checkpoints and recovery.

AI

Agent & tool boundary

Provider-neutral model routing, capability registry and schema-validated sandboxed tools.

APPLICATION BASELINETYPESCRIPT

A modular core, API and separately identified workers keep the MVP coherent and inspectable.

STATE & RETRIEVALSQL + VECTOR

Transactional mission state and permission-filtered semantic retrieval share one authoritative data boundary.

WORKFLOWDURABLE

Failure-aware orchestration preserves mission history and avoids duplicate committed effects.

OBSERVABILITYTRACEABLE

Correlation-aware logs, metrics, events, approvals, budgets and action receipts.

Remember what matters.
Forget by policy.

Aura’s memory specification separates memory categories, owners, classifications and retention rules so useful context does not become an ungoverned data lake.

01 / WORKING

Working memory

Short-lived task context isolated from durable storage.

02 / EPISODIC

Episodic memory

Versioned mission and interaction history with provenance.

03 / SEMANTIC

Semantic memory

Facts and concepts retrieved through permission-aware ranking.

04 / PROCEDURAL

Procedural memory

Approved workflows, tool patterns and repeatable operating knowledge.

05 / USER

User memory

Explicitly authorized preferences and personal context.

06 / ORGANISATION

Organisation memory

Tenant-isolated knowledge with classification and access policy.

PERMISSION-AWARE RETRIEVALRequest → Policy → Search → Rank → Filter → Context

Every memory operation carries a unique ID, owner, classification, version, integrity check and audit reference.

A working foundation.
Still inside the lab.

The current repository contains mission APIs, planning and state transitions, worker orchestration, policy and tool boundaries, tenant-scoped memory, observability and an operator-facing web console. Local QA suites exercise the highest-risk paths; independent security and production validation remain open gates.

IMPLEMENTED

Mission control plane

Create, plan, approve, pause, resume and cancel flows with idempotency and audit context.

LOCAL QA

Security-critical tests

Tenant isolation, identity tokens, sandboxed tools, retrieval, event streaming, orchestration recovery and interface validation.

OPEN GATE

Release assurance

Red-team review, backup restore, accessibility, load, operations and independent security assessment before public release.

EVIDENCE BOUNDARY — Repository implementation and local tests demonstrate engineering progress. They do not establish production availability, certification, enterprise service levels or unrestricted autonomous capability.

Control first.
Autonomy last.

GATE / 01

Kernel foundation

Identity, mission model, task graph, event bus and immutable audit.

GATE / 02

Permissioned agents

Agent registry, model router, sandboxed tools and approval workflows.

GATE / 03

Governed memory

Policy-aware retrieval, retention, versioning and tenant isolation.

GATE / 04

Persistent missions

Long-running execution with verification, recovery and human oversight.

FOUNDATION

Core path implemented

Mission, policy, orchestration, memory, telemetry and console components now exist as an integrated development baseline.

BUILD NEXT

Reference missions

Complete sourced research, user-file analysis and structured document generation with reproducible evidence bundles.

OUTSIDE MVP

Open-ended autonomy

Unrestricted browser control, self-directed finance and unreviewed third-party write actions remain explicitly excluded.

AIOS ENGINEERING & ASSURANCE PARTNERS

Challenge the governed kernel.

We are interested in durable systems, agent safety, identity, red-team, memory, observability and human-computer interaction expertise.

PRODUCT PORTFOLIO

See where Aura fits in the wider launch sequence alongside PoP L1, orbital compute and SPIDER.

View all products →