MAIOS

Multi AI Operative System · PC · privacy · continuity

Multi AI Operative System

Local agentic operating system.

MAIOS is a research and development project for safe and secure agentic systems. The project aims to develop an intelligence core that remains resilient as models and providers change, evolves and improves itself. Primary specifications: context awareness and internal and external operational capabilities.

Start the setup Explore the system

To operate: a project folder, a project-aware coding agent such as ChatGPT/Codex or Claude Code, and the human competence required to define and verify the result.

System field

The system becomes legible through its relations.

The Atlas groups the current entities without flattening their different roles and states.

  1. 01Systems and nodesMAIOS, public surfaces and situated coordination.
  2. 02KernelsMMK, RepoKernel and the Project Kernel generated for the target.
  3. 03FacultiesCompetences, meta-competences and bounded operating skills.
  4. 04Products and pathsCombinations that can be inspected, prepared or used.
  5. 05LabsMixed-maturity research cycles, tools and domain applications.

Manifesto

AI must be able to live in the user's own system.

Files, email, folders, documents, project memory and tools already define everyday work. The MAIOS project develops agentic systems in that same environment, with local operation as the reference condition.

When a task can remain on the PC, the target is cycle-closed operation without an internet connection or external provider. Frontier models and external services remain available through an explicit connection when the task requires them.

Prompts, corrections, evaluations and operating choices form knowledge about the user's work over time. MAIOS is designed to preserve and reuse that operating knowledge inside the user's system as models and providers change.

For this capability to remain legible, every work front has a responsible instance and a verifiable intent lineage: source, objective, write owner, authority, handoffs and receipts. Agents, tools, skills, competences and data remain faculties subordinate to the selected front.

Security is the condition that allows this capability to grow: user authority, explicit boundaries, verifiability and recovery of actions, privacy and continuity of work.

System form

Models, skills and tools can change. The operating state remains continuous.

Each part has a distinct role. Models propose, skills provide competence, tools act within permissions, and project state keeps the work readable over time.

  1. 01 Context

    Local sources and project memory.

  2. 02 Competence

    Selected skills and operating rules.

  3. 03 Intelligence

    Local model or externally hosted model selected explicitly.

  4. 04 Action

    Tool use within permissions and review.

  5. 05 Continuity

    Verified state retained for the next cycle.

Architecture in use

A kernel coordinates. Each instance owns one front. Faculties remain subordinate.

MMK keeps the relationship between intent, persistent instances, faculties, authority and receipts legible. MAIOS develops the operating environment in which this continuity can live inside the user's system.

  1. 01Intentsource and objective
  2. 02MMKcoordinates the chain
  3. 03Instancesone owner per front
  4. 04Facultiesagents, tools, skills and data
  5. 05Resultverified proposal or effect
  6. 06Receiptstate and resumption

MMK

The coordination kernel.

It connects source, intent, owning instance, write owner, authority and effects while keeping the work chain verifiable.

See how MMK works

Persistent instances

Responsibility for one defined work front.

Each instance preserves its own state and receives only the context required by the work it owns.

Subordinate faculties

Capability used within explicit authority.

Agents, models, tools, skills, competences and data contribute without acquiring control of the front merely because they are available.

From context to an operating system

Consulting when a decision is needed. Setup when something must be built.

Work starts from objectives, real activities, authorised sources and responsibilities. Consulting produces orientation and decisions. MAIOS Setup prepares the first project with an assistant, competences, operating memory, data rules and a RepoKernel handoff.

01

Direct consulting

Analysis, a preliminary report and human review of decisions, without starting an installation.

02

MAIOS Setup

Consultative orientation and generation of the first AI system built around how the organisation actually works.

03

Project Kernel

The project preserves state, sources, assistant, competences, rules, receipts and continuity in a portable form.

Start from your context. You can prepare the complete AI setup package or the report for a human consultation.

Open MAIOS Client Setup

Development programme

The work required to complete the system.

This programme defines the long-term direction of the project. Individual technical steps may change as evidence develops.

  1. 01 Local operating base

    Context, memory, permissions and tools inside the user's system.

  2. 02 Small local models

    Define useful roles and limits for models that run on consumer PCs.

  3. 03 MiniAGI workbench

    Test components, workflows, decisions and model roles inside MAIOS.

  4. 04 Reproducible component and task state

    Generate and validate component and task data before using them in the interface.

  5. 05 Guided decisions

    Present bounded choices when the system lacks information required to proceed.

  6. 06 Controlled PC workflows

    Extend work across files, folders, email, browser and documents with explicit authority.

  7. 07 Installable form

    Prepare a maintainable distribution for private users and organisations.

  8. 08 Evolution through use

    Use verified work to improve skills, tools, components and documentation.

Current project state

Components and current state.

The state changes only when a component changes position. It is not a daily activity log.

Status Component Current function
Public maios.it Definition, documentation, programme, access and public distribution.
In use MMK Orientation and coordination kernel; its portable operating part enters generated Project Kernels.
Private MiniAGI Local application and workbench inside MAIOS for capabilities, workflows, context and local models.
In test MAIOS Setup Guided path that prepares the context, the report and one reviewable MAIOS Project Package.
Distributed RepoKernel / Project Kernel Reviewable compilation of state, sources, skills, authority, reentry and learning into the project.
In review Apps, accounts, API and activation Registered access, persistent services, integrations and external effects still require dedicated infrastructure and proof.

Research horizon

Research questions.

01

Local intelligence

Which everyday tasks can small models perform reliably on consumer PCs?

02

Operational continuity

How can context and project state persist as tools and models change over time?

03

Controlled action

How should permissions and review work when agentic systems use files, email, browser and applications?

04

Independent operation

Which functions must remain available when a network or provider is unavailable?

05

Evolutive capability

How can verified use improve tools, skills and workflows without losing control of the system?

06

Accessible distribution

How can the system remain understandable and maintainable for people without specialist knowledge?

Support the programme

Research, test environments, technical work and resources.

MAIOS can be supported by people and organisations that may benefit from local agentic AI, or that share the objective of preserving privacy, freedom and user authority as AI becomes more autonomous.

Research support

Financial or material resources for engineering, model evaluation, security, documentation and distribution.

Pilot environments

Real workflows that can be studied with clear boundaries for data, privacy and actions.

Technical contribution

Local models, systems engineering, security, UX, validation, packaging or hardware competence.

Requirements and use

Professional or personal contexts that clarify what a private and resilient agentic system must do.

Support the research, contribute or propose a pilot context.

info@d-nd.com