prehnite 0.1.0a1__tar.gz

prehnite-0.1.0a1/.gitignore

@@ -0,0 +1,230 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[codz]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#   Usually these files are written by a python script from a template
+#   before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py.cover
+*.lcov
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+# Pipfile.lock
+
+# uv
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+# uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+# poetry.lock
+# poetry.toml
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#   pdm recommends including project-wide configuration in pdm.toml, but excluding .pdm-python.
+#   https://pdm-project.org/en/latest/usage/project/#working-with-version-control
+# pdm.lock
+# pdm.toml
+.pdm-python
+.pdm-build/
+
+# pixi
+#   Similar to Pipfile.lock, it is generally recommended to include pixi.lock in version control.
+# pixi.lock
+#   Pixi creates a virtual environment in the .pixi directory, just like venv module creates one
+#   in the .venv directory. It is recommended not to include this directory in version control.
+.pixi/*
+!.pixi/config.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule*
+celerybeat.pid
+
+# Redis
+*.rdb
+*.aof
+*.pid
+
+# RabbitMQ
+mnesia/
+rabbitmq/
+rabbitmq-data/
+
+# ActiveMQ
+activemq-data/
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.envrc
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs/Zensical documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#   JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#   be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#   and can be added to the global gitignore or merged into this file.  For a more nuclear
+#   option (not recommended) you can uncomment the following to ignore the entire idea folder.
+# .idea/
+
+# Abstra
+#   Abstra is an AI-powered process automation framework.
+#   Ignore directories containing user credentials, local state, and settings.
+#   Learn more at https://abstra.io/docs
+.abstra/
+
+# Visual Studio Code
+#   Visual Studio Code specific template is maintained in a separate VisualStudioCode.gitignore that
+#   can be found at https://github.com/github/gitignore/blob/main/Global/VisualStudioCode.gitignore
+#   and can be added to the global gitignore or merged into this file. However, if you prefer, you
+#   could uncomment the following to ignore the entire vscode folder
+# .vscode/
+# Temporary file for partial code execution
+tempCodeRunnerFile.py
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+# Marimo
+marimo/_static/
+marimo/_lsp/
+__marimo__/
+
+# Streamlit
+.streamlit/secrets.toml
+
+# Private, owner-authorized image corpora and source manifests (never upload)
+/private-r2-pilot/
+/private-r2-*.json
+/prehnite-private-corpus/
+/prehnite-private-reports/
+
+# Private exploratory apparel masks and reports
+/apparel-pilot-qa/
+/private-apparel-pilot/

prehnite-0.1.0a1/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Studio Zettlerei
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

prehnite-0.1.0a1/PKG-INFO

@@ -0,0 +1,70 @@
+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**.

prehnite-0.1.0a1/README.md

@@ -0,0 +1,236 @@
+<div align="center">
+  <img src="assets/prehnite-riso-green-coral.png" alt="Prehnite — Make the batch. Keep the best. Forest-green and coral Riso identity on warm paper." width="680" style="border-radius: 16px; clip-path: inset(0 round 16px);" />
+  <p><strong>Batch image production without the platform tax.</strong><br />
+  An open-source Python engine for generating, checking, refining and exporting visual assets.</p>
+  <p>
+    <a href="https://github.com/interzettel/prehnite/actions/workflows/ci.yml"><img alt="Tests" src="https://github.com/interzettel/prehnite/actions/workflows/ci.yml/badge.svg" /></a>
+    <a href="LICENSE"><img alt="License: MIT" src="https://img.shields.io/badge/license-MIT-075231" /></a>
+    <img alt="Python 3.11 or newer" src="https://img.shields.io/badge/python-3.11%2B-53675a" />
+    <img alt="Development version" src="https://img.shields.io/badge/status-development-a63d35" />
+    <img alt="Mandatory platform fees: zero" src="https://img.shields.io/badge/platform%20fees-%240-075231" />
+  </p>
+  <p>
+    <a href="#try-it-in-two-minutes">Quick start</a> ·
+    <a href="#how-it-works">How it works</a> ·
+    <a href="#where-things-stand">Project status</a> ·
+    <a href="docs/progress/STATUS.md">Progress</a> ·
+    <a href="CHANGELOG.md">Changelog</a> ·
+    <a href="docs/VERSIONING.md">Versioning</a> ·
+    <a href="https://github.com/orgs/interzettel/projects/1">Project board</a>
+  </p>
+</div>
+
+---
+
+## Ninety images is a different problem from one image
+
+Generating a single good picture is easy to demonstrate. Making a *batch* of assets to a brief, with the correct dimensions, cutouts, filenames, and a predictable bill, is a different job.
+
+Prehnite is an attempt to automate the boring, expensive middle: keeping track of prompts and seeds, trying variants, checking what worked, doing targeted repairs instead of blindly starting over, and handing back exactly the files you asked for.
+
+The ambition is simple: **spend less per *usable* asset, not merely less per generation.** It is a design target, not a claim we've already established with real-world benchmarks.
+
+**Free orchestration is non-negotiable.** Prehnite is MIT licensed, runs locally, and has no required account, hosted workflow subscription, or per-seat platform fee. Use a local model where practical; connect to a paid image API *only when you choose to*, with a budget you set.
+
+## Start with a zero-cost local batch
+
+Run `prehnite init --destination my-first-batch` to get two safe sample
+YAML files, then `prehnite doctor` to inspect your Python environment
+without revealing any keys or contacting a service. For details, see
+[the first-run guide](docs/ONBOARDING.md). These commands do not run AI
+image generation or incur any inference costs.
+
+For measured-versus-unknown OS/RAM/GPU inventory, run `prehnite hardware`.
+This **offline, read-only** command can query local `nvidia-smi`; it never
+certifies a GPU memory tier or recipe feasibility. See
+[hardware doctor limitations](docs/HARDWARE_DOCTOR.md) (issue #134).
+
+## Try it in two minutes
+
+You can test the complete workflow without a GPU, credentials, subscription, or inference charge. **The demo creates simple synthetic test images, not AI photographs.**
+
+Clone the repo, create a Python 3.11+ environment, and install:
+
+```bash
+git clone https://github.com/interzettel/prehnite.git
+cd prehnite
+python -m venv .venv
+# Windows PowerShell: .\.venv\Scripts\Activate.ps1
+# macOS/Linux:        source .venv/bin/activate
+python -m pip install -e .
+```
+
+Run a preflight estimate, then the zero-cost sample batch:
+
+```bash
+prehnite validate examples/manifest.yaml --profile examples/profile-demo.yaml
+prehnite estimate examples/manifest.yaml --profile examples/profile-demo.yaml
+prehnite run examples/manifest.yaml --profile examples/profile-demo.yaml --workspace workspace
+```
+
+Look under `workspace/batches/portrait-pilot/`. You'll find the selected images, the state journal, JSON/CSV reports, an [offline visual gallery](docs/OFFLINE_GALLERY.md) (`gallery.html`), a [local provenance record](docs/PROVENANCE.md) (`provenance.json`), and a ZIP with the requested output paths. A second run reuses completed stages.
+
+<details>
+<summary><strong>What the output looks like</strong></summary>
+
+```text
+workspace/
+└── batches/
+    └── portrait-pilot/
+        ├── state.sqlite3
+        ├── intermediates/
+        ├── selected/
+        │   └── photos/
+        │       ├── street-001.jpg
+        │       └── interior-001.jpg
+        ├── report.json
+        ├── operations.csv
+        ├── report.html
+        ├── gallery.html
+        ├── provenance.json
+        └── prehnite-portrait-pilot.zip
+```
+
+The ZIP preserves your requested relative filenames. Accepted cutouts, when configured, are transparent PNGs. Internally generated files stay separate from the deliverables.
+
+</details>
+
+## How it works
+
+| | Stage | What Prehnite does |
+| :--- | :--- | :--- |
+| 01 | **Describe** | Read a structured manifest: prompts or existing images, exact sizes, targets and output filenames |
+| 02 | **Plan** | Choose model providers and variants; estimate request counts and spend before starting |
+| 03 | **Make** | Run generation and optional upscaling; preserve seeds, inputs and intermediate artifacts |
+| 04 | **Check** | Apply technical checks, and optional model-based judgments when configured |
+| 05 | **Repair** | Retry selectively under a hard attempt limit; cutout errors needn't regenerate accepted photos |
+| 06 | **Deliver** | Export accepted images, named cutouts, a ZIP and an inspectable report |
+
+A small manifest can describe a portrait batch with precise delivery requirements:
+
+```yaml
+id: street-pilot
+assets:
+  - id: rainy-street
+    prompt: "A candid smartphone photograph of a rainy street at dusk"
+    aspect_ratio: "9:16"
+    target_size: [2160, 3840]
+    output:
+      image: photos/rainy-street.jpg
+      cutouts: {}
+```
+
+The **manifest** says *what* to make. A separate **profile** says *how*: which provider/model to use, whether to compare 2K-upscaled and 4K variants, what to accept, the retry limit, and whether cloud calls may cost money. See the [working examples](examples/).
+
+### Model Recipe status: micro / macro
+
+Prehnite now has **versioned model-workflow blueprints** rather than a single hard-coded model. `prehnite-micro` targets operator-configured, local-first ComfyUI runs; `prehnite-macro` explores later high-capability/hybrid workflows. The checked-in JSON recipes **do not contain models or execute on their own**.
+
+| Capability | `prehnite-micro` | `prehnite-macro` |
+| --- | --- | --- |
+| Versioned manifest / capability graph | Implemented | Implemented |
+| Local model/input hash resolver | Implemented, operator-supplied | Not implemented |
+| Recipe-bound engine/journal execution path | Experimental, opt-in | Not implemented |
+| Checkpoint installer and rights review | Planned | Planned |
+| Real GPU & human-cutout-quality benchmark | Unverified | Unverified |
+
+**Read the [recipe roadmap](docs/ROADMAP.md#model-recipes-workstream--versioned-micromacro-october-2026) and [recipe architecture](docs/MODEL_RECIPES.md)** for the work completed, outstanding milestones and exact evidence gates. Neither mode claims a tested 4/6/8/12 GB laptop tier or preinstalled models.
+
+### Explicit local ComfyUI recipe execution (experimental)
+
+The existing `prehnite run` command can now pin an operator-supplied micro recipe
+and exact local workflow/checkpoint SHA256 to a batch's SQLite journal using
+`--recipe --bindings --workflows --checkpoints`. Model/graph mismatches and
+unbound resumes fail closed; **actual ComfyUI execution requires the operator's
+separate local ComfyUI process, checkpoints and compatible graphs** and has not
+been GPU-benchmarked. See [recipe execution limitations](docs/RECIPE_EXECUTION.md).
+
+### Local micro recipe input verification (non-executable)
+
+`prehnite recipe-resolve` verifies explicitly supplied local ComfyUI API graphs and
+checkpoint bytes against SHA256 and operator-acknowledged terms without downloads,
+cloud calls, credentials or inferred GPU suitability. This **inspection command**
+produces a read-only, non-executable receipt; use the separate, opt-in
+`prehnite run` recipe bridge above to bind those inputs to a real batch.
+See the [operator setup and limitations](docs/LOCAL_MODEL_RESOLVER.md).
+
+### Bring your own models, not your own orchestration subscription
+
+| Route | What it means today |
+| --- | --- |
+| **Demo (no-cost)** | Synthetic generator + Pillow resize and technical checks. Runs offline and is covered by CI. |
+| **Local ComfyUI** | Adapter for your own ComfyUI API-format workflows on localhost. You supply model weights and real graph/node mappings; no validated GPU workflow is bundled yet. |
+| **fal.ai (optional paid)** | Opt-in hosted image generation/upscaling/segmentation adapter. Interfaces exist, but real endpoint schemas, pricing and image results still require live integration tests. |
+| **Anthropic judge (optional paid)** | An opt-in image comparison/rubric adapter. It is not a calibrated or production-validated quality benchmark. |
+
+No paid stage can be enabled by simply running the demo. Real cloud profiles require explicit `allow_paid` and, where images leave your machine, `allow_cloud_uploads`. They also require per-stage cost estimates and a positive `max_usd` cap. **Reserved amounts are estimates, not provider invoices or guaranteed billing ceilings.**
+
+You can inspect what each adapter is actually prepared to run using
+`prehnite models`. The [typed capability catalogue](docs/PROVIDER_CONTRACTS.md)
+lists operation/input types, supported tiers and formats, upload requirements,
+schema references, licensing cautions, and unimplemented candidates. Preflight
+checks unsupported paid models and incompatible options **before an API call**;
+custom ComfyUI graphs remain user-defined. The catalogue never supplies a
+live price: all paid quotes in example profiles must be verified and updated
+by you.
+
+For the locally hosted route, start with [the ComfyUI example profile](examples/profile-comfyui.yaml). For pay-per-use hosting, review [the fal profile](examples/profile-fal.yaml), verify endpoint contracts and current prices, and provide your own `FAL_KEY`. Do not use the sample prices as a quote.
+
+### Inspect detail without paying a judge
+
+Image previews can hide important 4K defects. You can create **native-pixel detail crops** and an annotated overview entirely offline:
+
+```bash
+prehnite crops my-photo.png --output detail-evidence --edge 1024
+```
+
+The output includes five lossless corner/center crops and a SHA256 manifest. Add `--anchor-x 0.45 --anchor-y 0.35` for one manually chosen area. This creates review evidence, **not an automated quality score**. [Read the crop guide](docs/CROP_EVIDENCE.md).
+
+### Opt-in reuse for repeatable stages
+
+Prehnite can keep locally verified stage outputs in a
+[private, content-addressed cache](docs/CACHING.md). It is disabled unless
+you specify an explicit owner namespace; prompts, seeds, input hashes,
+model options, version and model revision participate in the cache key.
+Cloud-model reuse requires a separate cache opt-in and a pinned model
+revision. A cache hit has no *new* provider request, but it does not erase
+historical charges or prove previously billed amounts.
+
+### Measure cost per usable asset, not raw generations
+
+The offline [economics report](docs/ECONOMICS.md) distinguishes
+**reserved/estimated spend** from **provider-reported partial amounts** and
+**verified invoices** (currently unavailable). It tracks accepted versus
+rejected assets, repeated stage invocations, local timing coverage and
+cost-per-accepted output. Failures still contribute to reserved cost; a
+batch with zero accepted assets has no meaningful cost-per-accepted figure.
+
+## Where things stand
+
+**`0.1.0a1` · first public developer alpha candidate.** Public PyPI installation is available only after the separately gated, verified [publishing workflow](docs/PYPI_PUBLISHING.md) succeeds; a version number or Git tag alone does not prove publication. See [Changelog](CHANGELOG.md), [Versioning](docs/VERSIONING.md) and [Release policy](docs/RELEASING.md). The core synthetic workflow works; **photorealistic quality, 4K GPU feasibility, segmentation accuracy and actual API billing have not been certified**.
+
+The tracker distinguishes implemented scaffolding from independently evidenced completions, and records issues blocked by local hardware, human calibration or paid-call authorization.
+
+[![Prehnite roadmap and atomic-task delivery progress](docs/progress/atomic.svg)](docs/progress/ATOMIC.md)
+
+**Two different measures:** The original 61-unit release roadmap counts complete, evidenced units. The [atomic task tracker](docs/progress/ATOMIC.md) counts smaller PR-sized deliverables **only within the initially unfinished roadmap units**. An atomic task may finish without completing its parent; neither percentage measures image quality or engineering hours. [Full original milestone chart](docs/progress/progress.svg).
+
+[Read the 61-unit release ledger](docs/progress/STATUS.md) · [73 agent-sized tasks and acceptance criteria](docs/progress/ATOMIC.md) · [90-asset no-cost benchmark](docs/OFFLINE_90_BENCHMARK.md) · [Technical release gates](docs/PROGRESS.md) · [Offline production analytics](docs/OFFLINE_ANALYTICS.md) · [Private native source/output inspection](docs/NATIVE_INSPECTION.md) · [Pixel-grid diagnostics](docs/PIXEL_DIAGNOSTICS.md) · [Affine v2 registration contract](docs/AFFINE_CONTRACT.md) · [Experimental Jev typed visual judge](docs/JEV_VISUAL_JUDGE.md) · [Unified decision engine](docs/DECISION_ENGINE.md) · [Attested 4K tile diagnostics](docs/ALIGNMENT_TILES.md) · [Optional Parquet export](docs/PARQUET_EXPORT.md) · [1:1 native image inspection](docs/NATIVE_INSPECTION.md) · [Failure codes and recovery actions](docs/FAILURE_TAXONOMY.md) · [Current sprint](docs/SPRINT.md) · [Roadmap](docs/ROADMAP.md)
+
+The [Produce / Explore / Experiment evaluation RFC](docs/BATCH_OBJECTIVES.md) describes future batch modes and a measurement framework; v0.1 implements neither a general exploration planner nor a Pareto dashboard.
+
+GitHub [Project #1](https://github.com/orgs/interzettel/projects/1) tracks workflow; the version-controlled [ledger](progress/ledger.json) tracks evidence. They answer different questions. Native Project field syncing is still [planned](https://github.com/interzettel/prehnite/issues/44).
+
+## Built for a maintainer and their agents
+
+Prehnite has a CLI first, optional [stdio MCP tools](src/prehnite/mcp_server.py) second. A person can run it directly; agents can submit preconfigured batches, check status, and request exports without being given arbitrary shell access. MCP workers now support [durable reconnect and cooperative pause/cancel/resume](docs/MCP_LIFECYCLE.md) on the main branch; a stop request does not immediately stop an in-flight provider request.
+
+The project itself is managed the same way: small issue scopes, PR-based delivery, offline tests, and a persistent [agent-to-agent handoff thread](https://github.com/interzettel/prehnite/issues/9). One maintainer sets direction and approves spend and releases. Agents do the repeatable engineering work and leave evidence—not inflated status reports.
+
+To contribute or extend the engine, start with [Contributing](CONTRIBUTING.md), [Architecture](docs/ARCHITECTURE.md), [Security](docs/SECURITY.md), and [Versioning](docs/VERSIONING.md). Issues are welcome, particularly for adapters that remain optional and testable without network access.
+
+## License and boundaries
+
+**[MIT](LICENSE)** for Prehnite's own code. Image models, checkpoints, external APIs, datasets, and other installed applications retain **their own** terms; nothing about this repository grants a license to third-party weights. See [Licensing](docs/LICENSING.md) and the [release policy](docs/RELEASING.md).
+
+No model checkpoint ships with the repo. No mandatory hosted product or subscription. No release tags until a version has passed its gates and the maintainer approves it.

prehnite-0.1.0a1/docs/PYPI_README.md

@@ -0,0 +1,38 @@
+# 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**.

prehnite-0.1.0a1/pyproject.toml

@@ -0,0 +1,61 @@
+[build-system]
+requires = ["hatchling>=1.26"]
+build-backend = "hatchling.build"
+
+[project]
+name = "prehnite"
+dynamic = ["version"]
+description = "Local-first, quality-gated visual asset production with model-agnostic providers and MCP."
+readme = "docs/PYPI_README.md"
+license = "MIT"
+requires-python = ">=3.11"
+authors = [{name = "Studio Zettlerei"}]
+dependencies = [
+  "pydantic>=2.9,<3",
+  "pillow>=11,<13",
+  "pyyaml>=6,<7",
+  "typer>=0.15,<1",
+  "httpx>=0.27,<1",
+]
+
+[project.urls]
+Homepage = "https://github.com/interzettel/prehnite"
+Documentation = "https://github.com/interzettel/prehnite/blob/main/README.md"
+Source = "https://github.com/interzettel/prehnite"
+Issues = "https://github.com/interzettel/prehnite/issues"
+Changelog = "https://github.com/interzettel/prehnite/blob/main/CHANGELOG.md"
+
+[project.optional-dependencies]
+fal = ["fal-client>=0.7,<1"]
+anthropic = ["anthropic>=0.69,<1"]
+mcp = ["mcp>=1.28,<3"]
+parquet = ["pyarrow>=18,<27"]
+dev = ["pytest>=8,<10", "ruff>=0.8,<1", "packaging>=24,<27"]
+
+[project.scripts]
+prehnite = "prehnite.cli:app"
+prehnite-mcp = "prehnite.mcp_server:main"
+
+[tool.hatch.version]
+path = "src/prehnite/_version.py"
+
+# Public sdists contain only the package and release metadata. Never ship
+# local .env templates, CI internals, uploaded photos or third-party weights.
+[tool.hatch.build.targets.sdist]
+include = [
+  "/pyproject.toml",
+  "/README.md",
+  "/LICENSE",
+  "/docs/PYPI_README.md",
+  "/src/prehnite",
+]
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/prehnite"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+
+[tool.ruff]
+line-length = 100
+target-version = "py311"

prehnite-0.1.0a1/src/prehnite/__init__.py

@@ -0,0 +1,2 @@
+"""Prehnite: quality-gated visual asset production."""
+from ._version import __version__

prehnite-0.1.0a1/src/prehnite/_version.py

@@ -0,0 +1,7 @@
+"""Authoritative PEP 440 package version for reviewed public alpha builds.
+
+Hatchling discovers this value, and prehnite.__init__ imports it.
+Change only in a dedicated, reviewed version-bump PR.
+"""
+
+__version__ = "0.1.0a1"