ADNovelFactory

KO | EN

gitlite — search

#react #go #android #machine-learning #docker #macos #nodejs #java #windows #cli #awesome-list #vue #pytorch #ios

(★ 111)

Native macOS long-form novel workflow from S0 specification to L6 audit and TXT export

ADNovelFactory 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# ADNovelFactory

[English](README.md) | [简体中文](README.zh-CN.md)

[![CI](https://github.com/n9896027/ADNovelFactory/actions/workflows/ci.yml/badge.svg)](https://github.com/n9896027/ADNovelFactory/actions/workflows/ci.yml)
[![CodeQL](https://github.com/n9896027/ADNovelFactory/actions/workflows/codeql.yml/badge.svg)](https://github.com/n9896027/ADNovelFactory/actions/workflows/codeql.yml)
[![License: Apache-2.0](https://img.shields.io/badge/License-Apache--2.0-blue.svg)](https://raw.githubusercontent.com/n9896027/ADNovelFactory/main/LICENSE)

Native macOS workspace for producing long-form fiction through a recoverable, file-backed pipeline. It keeps research, story planning, chapter work, quality gates, cross-chapter facts, and chapter-by-chapter TXT export in one local project directory.

> **0.1 alpha, source-only.** ADNovelFactory is for technically comfortable macOS users. It is not a hosted writing service, an unattended publishing tool, or a promise that generated text is publishable. Start with a short copied project and validate your model, prompts, cost policy, and editorial standard.

![ADNovelFactory icon](https://raw.githubusercontent.com/n9896027/ADNovelFactory/main/Resources/AppIcon.png)

## See It Running

The screenshot below is a real local desktop run captured on 2026-08-08. It contains no API key or manuscript body: an 8-source male-channel public-ranking snapshot, 160 ranking observations, 102 deduplicated titles, local trend statistics, and the entry point for three original topic candidates.

![A live ADNovelFactory S-1 topic-research workspace showing an 8-source channel scan, local statistics, and the candidate-generation action](https://raw.githubusercontent.com/n9896027/ADNovelFactory/main/docs/images/runtime-topic-research.jpg)

The ranking snapshot is an observation of public metadata, **not** evidence of demand, signing probability, traffic, revenue, or output quality. Candidate generation requires a model you configure; the screenshot does not imply a particular provider, model, speed, or cost. See the [reproducible runtime walkthrough](docs/DEMO.md).

## What It Does

```text
S-1 research -> human adoption -> S0 specification -> L0 bible
-> L0.5/L0.6 character ledgers -> L2 outline -> structural gate
-> L2.7 beats -> L3 draft -> L4 expansion -> L5 repair/review
-> L6 acceptance -> final audit -> TXT folder + SHA-256 manifest
```

- **S-1 topic research:** bounded public Qimao ranking matrices (1, 8, or 16 pages), local statistics, and three originality-checked topic candidates.
- **S0 to L2.7 planning:** channel, genre, target length, chapter count, bible, role-led character records, outline, a hash-bound structure gate, and scene beats.
- **L3 to L6 production:** resumable batch work, retries for failed chapters only, L4 expansion, L5 quality repair, L6 acceptance, and a final repair queue.
- **Continuity controls:** `truth.json`, world state, foreshadowing records, and consistency audits are persisted beside the manuscript.
- **Deliverable-first export:** chapter TXT files, a readable export note, and a SHA-256 manifest. Export refuses to claim completion when official L6 files are missing.
- **Model routing:** NorthDipper OpenAI-compatible routes with multiple Keys/models, or a local MLX server. Credentials are stored in macOS Keychain, never in project JSON.

After a human explicitly adopts a candidate, the S0 form remains editable and shows the length and chapter-count constraints that the later stages will consume. This is a sanitized sample, not a generated manuscript.

![Sanitized S0 project specification screen with channel, genre, tags, and length constraints](https://raw.githubusercontent.com/n9896027/ADNovelFactory/main/docs/images/adnovelfactory-s0.jpg)

## Quick Start

### 1. Build the app

```bash
git clone https://github.com/n9896027/ADNovelFactory.git
cd ADNovelFactory

swift build -c debug
./build.sh debug
open ADNovelFactory.app
```

The default build uses ad-hoc signing. It does not create certificates, modify system trust, or qualify as a notarized public macOS release. App projects live in `~/ADNovelFactory`, outside this source checkout.

### 2. Choose a model route

Use the app's **Settings** page to configure a NorthDipper route and save each Key to Keychain, or start local MLX:

```bash
uv venv ~/.mlx-env --python 3.12
source ~/.mlx-env/bin/activate
uv pip install mlx mlx-lm

MLX_MODEL=/absolute/path/to/model scripts/mlx_server.sh start
scripts/mlx_server.sh status
```

The app defaults to `http://127.0.0.1:8080/v1` for MLX. See [provider setup](docs/PROVIDERS.md) before sending unpublished material to any remote service.

### 3. Run a small project first

1. Create a project and save S0 with a small target length.
2. Optionally open **S-1 Research**, scan a bounded ranking scope, generate candidates, then explicitly adopt one into the editable S0 form.
3. Run L0, character records, outline, gate, and beats. Read every gate report before allowing downstream work.
4. Generate a few L3 chapters, inspect them, then continue with L4-L6 and export.

The application does not auto-publish, auto-spend beyond your provider policy, or replace editorial approval.

## Verification You Can Reproduce

The repository's local checks do not call a paid model:

```bash
python3 scripts/check_repository.py
python3 -m py_compile engine/novel_engine.py engine/roster.py engine/truth.py engine/story_state.py
python3 tests/run_local_regression.py
tests/run_swift_smoke_tests.sh
swift build -c debug
bash -n build.sh scripts/mlx_server.sh
```

The latest public branch is additionally checked by GitHub Actions on Apple Silicon and Intel macOS, CodeQL, and dependency review. Local or CI success validates the repository contracts, not model quality or third-party billing.

## Privacy and Boundaries

- Remote providers receive the stage context needed for a selected action. That can include unpublished seeds, plans, or chapter text.
- The S-1 collector requests only fixed public Qimao ranking pages, stores bounded metadata, and excludes blurbs, chapter text, and raw HTML.
- API keys are written through Security.framework to macOS Keychain. Do not commit environment variables, project folders, or screenshots without reviewing them.
- The app is local-first: it does not operate a cloud project store, account service, analytics service, or telemetry service.

Read [Privacy](PRIVACY.md), [Permissions and security boundaries](docs/PERMISSIONS.md), and the [threat model](docs/THREAT_MODEL.md) before sharing a project or using a remote model.

## Documentation

- [Runtime walkthrough](docs/DEMO.md)
- [Chinese README](README.zh-CN.md)
- [User guide](USER_GUIDE.md)
- [Architecture](docs/ARCHITECTURE.md)
- [Project format](docs/PROJECT_FORMAT.md)
- [Troubleshooting](docs/TROUBLESHOOTING.md)
- [Release policy](docs/RELEASING.md)
- [Roadmap](ROADMAP.md)
- [Contributing](CONTRIBUTING.md)
- [Security policy](SECURITY.md)

## Current Limits

- The UI and state reader are file-system based. Very long projects can still make chapter lists and journals feel heavy.
- Production routes are currently limited to NorthDipper OpenAI-compatible routing and local MLX.
- Cost figures are estimates. Provider billing is authoritative.
- GitHub builds are source builds with ad-hoc signing; no Developer ID notarized binary is currently published.

## License

[Apache License 2.0](LICENSE). Third-party and reference notices are listed in [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md).

ADNovelFactory

ADNovelFactory 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

ADNovelFactory

ADNovelFactory 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README