Metadata-Version: 2.4
Name: apron
Version: 0.0.0
Summary: Apron, by Mondegreens — name reserved; implementation has not started
Keywords: inference,deployment,vllm,evaluation,evidence
Author: Vlad Ryzhkov
License-Expression: Apache-2.0
Classifier: Development Status :: 1 - Planning
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.14
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: System :: Systems Administration
Requires-Python: >=3.11
Project-URL: Homepage, https://apron.dev
Project-URL: Source, https://github.com/mondegreens/apron
Description-Content-Type: text/markdown

# Apron

**Apron finds, deploys and maintains the inference solution that actually works
for a real workload.**

## Status: not implemented

This release reserves the name. There is no software here yet — no command, no
library, nothing to run. Installing it does nothing.

What exists today is the specification: the architecture, the decision records,
the invariants and the delivery plan.

- Specification and design record: <https://github.com/mondegreens/apron>
- Documentation: <https://apron.dev>

## What it will do

Given a real task and an accepted quality floor, serving requirement, policy and
budget, Apron compares managed APIs, self-hosted models and compound systems;
predicts resource fit before anything is rented or downloaded; deploys the
selected solution; measures whether the task outcome and the serving contract
actually hold; diagnoses failures to a corrected plan; and keeps the result
qualified as models, engines and providers change.

Every result declares how it is known. A calculation is labelled a prediction
until something executes on the stated hardware. A measurement belongs only to
the exact execution that produced it. What is unknown is reported as unknown.

## Licence

Apache-2.0 for code and documentation. Evidence records are published separately
under CDLA-Permissive-2.0.
