A reimagined operating system for AI agents.
Intellect OS is an open-source, AI-native operating system designed around agents instead of adding agents on top of an operating system built for humans and applications.
This project is currently in the early architecture and development stage.
Current AI agents interact with computers through screenshots, simulated mouse movements, keyboard input, and fragile application-specific integrations.
These interfaces were designed for humans.
A mouse cursor exists because a human can point at only one place at a time. A human can type into only one application at a time. An AI agent does not have these limitations—it could work across multiple interfaces simultaneously, perform tasks in the background, and process structured context directly.
Existing operating systems force agents to imitate human behavior. Intellect OS aims to provide the primitives agents actually need.
Intellect OS will expose secure and permissioned APIs for:
- Selected and visible text
- Images and other on-screen content
- Active windows and screen regions
- Mouse position and user interactions
- Rendered frames
- Files and documents
- PDF and image viewers
- Background tasks
- System actions
- Agent memory and context
- Local and cloud AI models
- Human approvals and audit logs
Instead of relying only on screenshots and UI automation, agents will be able to access structured system context through explicit APIs.
The Intellect OS SDK will allow developers to build agents that can:
- Understand what the user is viewing or working on
- React to selections, clicks, files, and system events
- Work across multiple interfaces
- Perform tasks in parallel or in the background
- Render interactive agent interfaces
- Request permissions before accessing sensitive capabilities
- Use local or cloud AI models
- Connect to external tools through protocols such as MCP
Every agent will operate through a declared set of permissions controlled by the user.
The initial architecture will include:
- A minimal, immutable Linux base
- A Rust-based Agent Supervisor
- A permission and approval system
- A model-provider abstraction layer
- System-awareness and action APIs
- Agent memory and storage
- A TypeScript, React, and Tauri desktop shell
- Isolation using Linux namespaces, cgroups, containers, and virtual machines
The AI layer will live primarily in hardened user space rather than replacing the Linux kernel.
Intellect OS is intended to be model-independent.
Planned support includes:
- OpenAI
- Anthropic
- Google Gemini
- Ollama
- Local open-source models
- Custom model providers
Users will be able to choose the model that best fits their privacy, performance, and cost requirements.
- Agent Supervisor
- Model-provider interface
- Basic agent manifests
- Permissions and approvals
- Local task execution
- Selected-text API
- Screen-region API
- Window and frame context
- File and document access
- Agent-aware PDF and image viewers
- Public SDK
- Capability adapters
- MCP integrations
- Agent development tools
- Example agents
- Isolated virtual-machine environment
- Live USB preview
- Native installation for supported hardware
Intellect OS is currently being designed and developed in public.
The first releases will focus on establishing the agent runtime, permission model, and system-awareness APIs. The project will be developed iteratively, with frequent releases as individual components become usable.
Intellect OS is at an early stage, and contributions are welcome.
You can contribute by:
- Opening an issue with an idea or use case
- Reviewing architecture proposals
- Reporting security and permission-model concerns
- Building experimental capability adapters
- Improving documentation
- Submitting pull requests
For major changes, please open an issue before beginning implementation.
- Agents should not need to pretend to be humans.
- Users must control what agents can see and do.
- Local and cloud models should both be first-class.
- High-impact actions should require explicit approval.
- Agent activity should be observable and auditable.
- The core platform and SDK should remain open.
- Existing applications should become capabilities that agents can safely use.
A license will be selected before the first public release.
Created by Nishant Hada.