Overview
Ashlar is a local-first .NET runtime for AI workflows that need to be embedded, inspected, and operated on self-controlled infrastructure. It records workflow activity so teams can connect a submitted task with the corresponding trust-log entry and review what ran.
The project is self-hosted rather than offered as a hosted Ashlar service. Local model routing through Ollama is the default, while cloud providers are opt-in targets.
Key Features
- Auditable workflows store tasks under identifiers and associate them with entries in the trust log.
- A certification gate evaluates proposed code through analyzer checks, correctness witnessing, mutation testing, determinism checks, and dependency analysis.
- Admission and rejection decisions proven by the gate are retained as certification evidence.
- The runtime can operate through a command-line interface, HTTP API, containers, or an application host.
- Network exposure profiles require authentication before the API will start.
Integration Options
Ashlar supports embedding through NuGet packages, an HTTP client, and SDK integration. A host can also register its own bricks with the SDK; host-registered bricks are instantiated directly in the brick registry and are not evaluated through the certification path used for proposed code.
This distinction lets integrators choose how the runtime fits their application while preserving a defined gate for generated artifacts.
Development and Testing
The introductory workflow builds the .NET kernel solution, starts the API on loopback, submits a task, and examines its audit trail. It then runs the certification gate to observe both admission and rejection behavior. This basic path does not require Docker or API keys.
The repository also provides separate paths for trying a portal in Docker, developing with a dev container and CLI, and preparing deployments.