Apron¶
Decision and evidence system for LLM inference deployment.
Apron connects model choice and serving configuration with evidence scoped to the artifact, engine, hardware, and workload that produced it.
Start here¶
- Inspect measured deployment records: predictions and observations, with links to the stored records.
- Read the implementation and tests.
- Read the current project status: the repository README describes implemented phases and current work.
- Explore findings: technical notes linked to evidence.
Project stage¶
Early implementation. The repository contains the implementation, specification, and published measurement records. The PyPI package names are placeholders and are not the installation route for the current code.
A memory prediction, a healthy engine boot, and a serving result establish different things. Records keep their own scope and provenance; support for one environment does not establish support for another.
Design and contracts¶
Start with the product definition, then the target architecture and phase plan. The framework specification defines the extension points, invariants, and metrics. The architecture decision records record design decisions and their consequences.
For contributions, read how development works.
Maintained by Vlad Ryzhkov.