-
Notifications
You must be signed in to change notification settings - Fork 0
Home
This wiki is the canonical detailed design and implementation guide for XAI OS. The main repository README.md is the short entry point; the wiki holds the architecture, platform plans, benchmarking strategy, roadmap, and Codex build tasks.
XAI OS is a design-stage server-only operating system for CPU-only AI app agents. It targets applications that embed local AI agents capable of understanding their own source code, accepting human instructions, generating patches, rebuilding, testing, syncing with Git, and improving the running application.
The main repository currently boots an AArch64 QEMU-on-macOS prototype with UEFI handoff, serial logging, controlled exception diagnostics, generic timer discovery, GIC discovery, PSCI SMP bring-up for four CPUs, split VirtIO-MMIO transport/block/network drivers, a VirtIO-backed versioned read-only filesystem, parsed config manifest, a real EL0 /init ELF selected from that manifest, an explicit syscall table, process/capability metadata, user pointer validation, bad syscall tests, userspace service-manager policy, build/test sandbox metadata with rollback state, syscall-based service control, a VMM-backed kernel heap, VMM map/unmap checks, guarded user stack checks, a generic PMM/VMM-backed arena manager, AI Cell lifecycle/resource enforcement, explicit hot-core lease metadata, zero migration/context-switch telemetry for leased AI cores, shared read-only model arenas, private KV/cache and source-index arenas, PMM/VMM checks, and JSON-like boot telemetry.
These are design targets until measured against tuned Linux/BSD baselines.
| Area | Target |
|---|---|
| TCP/UDP latency | Up to 10-45% lower latency |
| Effective CPU-AI memory bandwidth | 3-18% higher |
| Sustained usable CPU-core performance | 2-12% higher |
| Scheduler jitter/migration | Near-zero on hot AI paths |
XAI OS cannot exceed physical silicon limits. The expected gains come from reducing scheduler migration, context switching, page faults after warmup, generic network overhead, memory duplication, bad NUMA placement, and unrelated interrupts.
- Project Purpose
- Architecture
- AI Cells
- Project Tracker
- Implementation Plan
- Codex Work Packages
- Testing and Benchmarking
- Performance Targets
- Benchmark Result Format
- Open Decisions
- Risk Register
- Roadmap
- Production Readiness Roadmap
- Glossary
- FAQ
This page defines the GitHub Wiki navigation sidebar.
- Architecture
- AI Cells
- CPU AI Runtime
- App Agents
- Memory System
- Networking
- Scheduler and Core Isolation
- Filesystem and Storage
- Driver Model
- Security Model
- Build System
- Build System
- Project Tracker
- Implementation Plan
- QEMU Full OS Core Workdown
- QEMU 100 Completion Plan
- Example Apps
- Codex Work Packages
- Testing and Benchmarking