Metadata-Version: 2.5
Name: prehnite
Version: 0.1.0a1
Summary: Local-first, quality-gated visual asset production with model-agnostic providers and MCP.
Project-URL: Homepage, https://github.com/interzettel/prehnite
Project-URL: Documentation, https://github.com/interzettel/prehnite/blob/main/README.md
Project-URL: Source, https://github.com/interzettel/prehnite
Project-URL: Issues, https://github.com/interzettel/prehnite/issues
Project-URL: Changelog, https://github.com/interzettel/prehnite/blob/main/CHANGELOG.md
Author: Studio Zettlerei
License-Expression: MIT
License-File: LICENSE
Requires-Python: >=3.11
Requires-Dist: httpx<1,>=0.27
Requires-Dist: pillow<13,>=11
Requires-Dist: pydantic<3,>=2.9
Requires-Dist: pyyaml<7,>=6
Requires-Dist: typer<1,>=0.15
Provides-Extra: anthropic
Requires-Dist: anthropic<1,>=0.69; extra == 'anthropic'
Provides-Extra: dev
Requires-Dist: packaging<27,>=24; extra == 'dev'
Requires-Dist: pytest<10,>=8; extra == 'dev'
Requires-Dist: ruff<1,>=0.8; extra == 'dev'
Provides-Extra: fal
Requires-Dist: fal-client<1,>=0.7; extra == 'fal'
Provides-Extra: mcp
Requires-Dist: mcp<3,>=1.28; extra == 'mcp'
Provides-Extra: parquet
Requires-Dist: pyarrow<27,>=18; extra == 'parquet'
Description-Content-Type: text/markdown

# Prehnite

**Local-first, quality-gated visual asset production. No mandatory SaaS or platform fees.**

[Source and project README](https://github.com/interzettel/prehnite#readme) · [Roadmap](https://github.com/interzettel/prehnite/blob/main/docs/ROADMAP.md) · [Changelog](https://github.com/interzettel/prehnite/blob/main/CHANGELOG.md) · [Issues](https://github.com/interzettel/prehnite/issues)

Prehnite is an MIT-licensed Python CLI and optional MCP integration for batch image generation, masks/cutouts, technical validation, targeted repair, stateful replay, quality evidence, and reproducible export. It **orchestrates image models**; it does not include model weights, a photorealistic image generator, or a subscription service.

## Quick start: no-cost local test

Requires Python 3.11 or later.

```bash
python -m pip install prehnite
prehnite init --destination my-first-batch
prehnite doctor
prehnite --help
```

The `init` command creates local starter configuration, but does not run model inference or incur fees. For a sample batch, follow the source repository's [two-minute demo](https://github.com/interzettel/prehnite#try-it-in-two-minutes).

For a sample zero-cost batch, follow the [up-to-date project README](https://github.com/interzettel/prehnite#try-it-in-two-minutes) and [onboarding guide](https://github.com/interzettel/prehnite/blob/main/docs/ONBOARDING.md).

## What is actually implemented

- CLI, strict manifest/profile validation, stage journal, bounded attempts and safe archive/export paths.
- Offline synthetic demo flow, reports and reproducible zero-cost validation.
- Experimental local ComfyUI adapter and **opt-in** `prehnite-micro` recipe binding with immutable batch identity. You must separately install ComfyUI and supply licensed checkpoints and compatible graphs. No real GPU model/cutout quality or 4K VRAM tier is certified.
- Optional paid inference connectors require explicit operator-selected profiles, permissions and budgets. **Installing Prehnite does not authorize any cloud uploads or paid API calls.**
- `prehnite-macro` is currently a read-only workflow blueprint, not an installed or benchmarked high-quality image pipeline.

See the [implementation/evidence roadmap](https://github.com/interzettel/prehnite/blob/main/docs/ROADMAP.md), [recipe contract](https://github.com/interzettel/prehnite/blob/main/docs/MODEL_RECIPES.md) and [public release history](https://github.com/interzettel/prehnite/blob/main/CHANGELOG.md).

## Release quality and licensing

The initial **0.1.0 alpha** is an early developer preview, **not production certified**. Passing synthetic CI does not prove realistic image quality, legal rights to separate checkpoints, accepted cutout accuracy, stable public APIs or physical GPU suitability.

Prehnite code: MIT. Checkpoint weights, providers and third-party assets are governed by their own licenses and usage terms. The package contains **no bundled checkpoint/model weights**.
