AIKernel.Control is the physical execution layer for AIKernel semantic graphs. Use this documentation when you need to run, inspect, or package Control engines for .NET hosts.
These docs describe Control as the AIOS SDK governance, security, and physical execution layer. Control maps semantic graphs from Core onto explicit policies, deterministic schedulers, diagnostics, and execution engines.
AIKernel.Monolith is the official AIOS distribution now in development. It is planned as the standard reference distribution that integrates the control plane with the Semantic OS layers after the 0.1.x line stabilizes.
Shared repository boundaries, v0.1.2 development versioning, dependency order, PyPI Trusted Publishing, and Python wrapper scope are defined by Package Release Alignment v0.1.2. The historical v0.1.1.1 validation rules remain available in AIKernel Repository Alignment v0.1.1.1. When a change crosses repositories, start with the Cross-Repository Developer Guide v0.1.1.1.
Control owns orchestration, policy application, Core gate invocation, runtime control, and execution coordination. It must not reimplement Decision Gate or Trajectory Gate truth tables, provider semantic evaluation, or browser runtime execution.
- User Guide
- Architecture
- Bonsai mapping
- Bonsai-1.7B built-in provider
- Execution engine
- Q1_0 CPU execution kernel
- Control pipelines
- CTG Control integration
- Concept Elevation Notes / 概念昇格ノート
- Python governance wrapper
- Licensing
- Read the User Guide when you want the install commands and the smallest deterministic emulator/CPU validation path.
- Read Architecture when you need to understand why semantic graphs are owned by Core while physical execution engines are owned here.
- Read Bonsai mapping before wiring Bonsai-style graphs to emulator, CPU, or GPU execution.
- Read CTG Control integration when you need to opt in to Core CTG gate evaluation during the Apply Policy stage.
- Read Python governance wrapper when consuming Control from Python through
aikernel-governance. Python remains a managed assembly bridge and must not reimplement CTG Gate logic.
Start with CPU and Emulator validation before binding GPU execution:
dotnet build AIKernel.Control.slnx -c Release
dotnet test AIKernel.Control.slnx -c Release --no-build- Install the matching
AIKernel.Control.*packages. - For v0.1.2 development, use local NuGet package versions such as
0.1.2-dev{buildNumber}and local Python wheel versions such as0.1.2.dev{buildNumber}. - Mount model assets through VFS/ROM instead of local ad hoc paths.
- Use CPU/Emulator packages for deterministic validation before binding GPU execution.
- Keep
aikernel-governancethin over the managed assemblies and publish it only through the v0.1.2 Trusted Publishing flow when stable publication is explicitly opened.