An offline-first, system-aware Linux OS project.
AiOS is an Arch-based operating system concept built around a local assistant that can understand the machine it runs on without sending user data to cloud services.
The goal is not to make an AI that silently controls the OS. The goal is to make a trustworthy system that can read, explain, search, plan, and ask before anything changes.
AiOS combines:
- a clean Arch Linux foundation
- local system indexing
- offline AI-assisted explanations and planning
- reversible, permission-gated system actions
- clear documentation for safety, architecture, and development phases
The system is designed around a simple rule:
The AI can help the user understand and plan. The user stays in control.
- Offline first: no cloud dependency for core AI behavior.
- Assistive, not autonomous: the AI does not silently execute system changes.
- Explainable: recommendations should be inspectable and understandable.
- Recoverable: risky operations must have clear rollback paths.
- Stable before flashy: daily-driver reliability matters more than novelty.
- Privacy-conscious: indexing starts with metadata, not file contents.
AiOS is in early alpha. The repository currently contains the project architecture, safety model, development timeline, and early implementation work for aio-indexd, the local filesystem indexing daemon.
Completed:
- project vision and principles
- architecture and safety documentation
- instruction schema design
- first
aio-indexdSQLite schema - basic filesystem scanner
- basic UNIX socket query server
- tests for index creation, lookup, and metadata refresh
In progress:
- installer and live environment planning
- hardening
aio-indexd - CLI interface design
- AI assistant integration design
Planned:
- controlled execution service
- incremental filesystem event indexing
- package ownership metadata
- desktop integration
- settings and assistant UI
AiOS/
core/
aio_indexd/ Filesystem indexing daemon prototype
aio-aid/ Offline assistant engine placeholder
aio-cli/ CLI interface placeholder
aio-execd/ Controlled execution service placeholder
docs/ Architecture, safety, schema, and timeline docs
models/ Local model notes and storage guidance
os/
installer/ Installer planning
services/ systemd service files
aio-indexd is the first core service implementation. It builds a local SQLite index of filesystem metadata so the assistant can answer questions like "where is this config file?" without scanning the whole machine every time.
Current indexed metadata:
- path
- file name
- extension
- size
- owner
- modified timestamp
Environment variables:
| Variable | Default | Purpose |
|---|---|---|
AIO_INDEXD_DB |
/var/lib/aio/index.db |
SQLite database path |
AIO_INDEXD_SOCKET |
/run/aio/indexd.sock |
UNIX socket path |
AIO_INDEXD_SCAN_ROOTS |
/etc |
os.pathsep separated scan roots |
Run the tests:
python -m unittest discover core/aio_indexd/tests| Phase | Focus | Status |
|---|---|---|
| 0 | Vision, principles, architecture | Complete |
| 1 | Installer and ISO foundation | In progress |
| 2 | Core services | In progress |
| 3 | Desktop environment | Planned |
| 4 | Offline AI read-only assistant | Planned |
| 5 | Controlled AI actions | Planned |
| 6 | Polish, docs, stability | Planned |
See docs/dev-phases.md for the detailed development plan.
AiOS intentionally separates:
- AI reasoning
- system indexing
- user approval
- execution
- logging and rollback
The assistant should never rely on unrestricted shell output or hidden automation. High-risk operations must be validated, explained, and confirmed before execution.
See docs/safety-model.md and docs/instruction-schema.md.
This project is not ready for daily use yet. It is being built in public from the foundation upward, with correctness and safety taking priority over visual polish.
The next practical milestone is making aio-indexd robust enough to serve as the local knowledge layer for the rest of the system.