alex663028/alex-novel-platform — explained in plain English
Analysis updated 2026-05-18
Generate a long-form novel chapter by chapter while keeping characters and plot threads consistent.
Track planted plot points across a story and get flagged when one goes unresolved.
Detect when a generated chapter drifts from the story's established writing style and trigger a targeted rewrite.
Build a custom app or editor plugin on top of the engine's REST API.
| alex663028/alex-novel-platform | gixsy95/abap_wiki | hao0321/video-autopilot-kit | |
|---|---|---|---|
| Stars | 37 | 37 | 37 |
| Language | Python | Python | Python |
| Setup difficulty | hard | hard | moderate |
| Complexity | 5/5 | 4/5 | 3/5 |
| Audience | developer | developer | general |
Figures from each repo's GitHub metadata at analysis time.
Requires Python 3.14, Node.js 18+, and at least one LLM API key, a one-click Windows launcher is provided as an alternative.
Alex is an open source engine for keeping long AI-written stories consistent over hundreds of thousands of words. The README is explicit that it is not a chat style writing assistant and not a set of prompt templates. Most AI writing tools are good at producing one passage of text, but this project targets a harder problem: keeping characters consistent, keeping cause and effect logical, and making sure planted plot threads actually get resolved, across a very long story that keeps advancing without someone watching every step. The author frames this as an engineering problem rather than a prompting problem, solved by building infrastructure that tracks the story's state, so the underlying AI model only has to do the part it is good at, writing a strong passage once given the right context. The engine has five cooperating parts. A narrative state machine holds a full snapshot of the story at all times: character profiles, locations, world building facts, chapter summaries, a timeline of key events, a map of separate storylines, and a registry tracking every planted plot thread from setup to payoff. A retrieval layer keeps a searchable index of everything already written, so relevant past scenes get pulled back in automatically instead of relying on the underlying model's own limited memory. A runtime component drives the process, planning at the level of parts and chapters, then generating chapter by chapter through a ten step pipeline that includes writing, checking for style drift, updating tracked state, and scoring narrative tension, with automatic pausing if something keeps failing. A prompt layer exposes over 20 separate, swappable prompt configurations for planning, generation, reference lookups, and analysis. A quality monitor scores narrative tension after each chapter, flags chapters that drift from the story's established style, and triggers a targeted rewrite rather than discarding progress when that happens. The project exposes its capabilities through a REST API, with an official web based workbench built on Vue as one interface among possibly others. It uses FastAPI on the backend, ChromaDB or FAISS for vector search, and SQLite for its database. Running it requires Python 3.14 and Node.js, plus at least one AI provider API key such as Anthropic, OpenAI, or Gemini. It ships under an Apache 2.0 license with an added Commons Clause restriction, meaning reselling it as a commercial product or service requires separate compliance with that clause.
An open source engine that keeps long, AI-generated stories consistent by tracking characters, plot threads, and story state across hundreds of thousands of words.
Mainly Python. The stack also includes Python, FastAPI, Vue.
Apache 2.0 with an added Commons Clause, which blocks selling the software itself as a commercial product or service without separate compliance.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.