convez.ai
PluginEscrita & ConteúdoIntermediário
thaisssimoes

Storyteller

thaisssimoes/storyteller

Detector e corretor de inconsistências em ficção com IA

Ver no GitHub
InicialPluginScore · de 0 a 100

CLI em Go que usa LLMs (Claude/Ollama) para detectar, corrigir e garantir consistência de worldbuilding em web novels. Analisa capítulos em três passes (macro/cena/continuidade), reescreve inconsistências e pode gerar histórias originais do zero com pipeline de validação automática.

Instalação

$npx convez-ai thaisssimoes/storyteller

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Validar consistência de mundos em web novels e romances serializados
  • Reescrever capítulos com problemas de continuidade automaticamente
  • Gerar histórias originais com validação de coerência narrativa
  • Análise em larga escala de múltiplos capítulos para detecção de plotholes

Veredito

Ferramenta especializada e bem-arquitetada para autores de ficção seriada. Suporta tanto Claude quanto Ollama local, oferecendo flexibilidade entre qualidade e custo. Ideal para quem trabalha com longas narrativas e precisa manter coerência em worldbuilding complexo.

README

A tradução para o português deste README ainda está na fila.

Storyteller

A Go CLI that detects and fixes worldbuilding inconsistencies in web novels, and can also write a fresh story from scratch and run the same consistency pipeline on the result.

Supports two LLM providers:

  • Anthropic Claude (default)
  • Ollama (local models)

Uses a two-tier model setup: a fast model for per-chapter scans and a robust model for macro analysis, fixer rewrites, and chapter writing.

Commands

CommandWhat it does
scrapeDownload all chapters from a chapter-list URL into output/chapters/
readAnalyse chapters in three passes (macro / scene / continuity) into output/inconsistencies.json
fixRewrite inconsistent chapters into output/story_fixed.txt
writeInteractive interview, write an original story from scratch, then run read + fix on the result
allRun scrape, read, and fix in sequence. Each step is skipped if its output is already on disk (set FORCE=1 to redo).

Setup

Requirements:

  • Go 1.22 or higher
  • An Anthropic API key, or a local Ollama install
git clone https://github.com/thaisssimoes/storyteller
cd storyteller
go mod download
go build

Configuration

The CLI reads from a .env file in the current directory.

NOVEL_URL=https://example.com/novel/chapters
OUTPUT_DIR=./output

PROVIDER=anthropic
ANTHROPIC_API_KEY=sk-ant-...

# Optional model overrides
ANTHROPIC_FAST_MODEL=
ANTHROPIC_ROBUST_MODEL=

# Ollama
OLLAMA_FAST_MODEL=qwen2.5:14b
OLLAMA_ROBUST_MODEL=gemma2:27b
OLLAMA_URL=http://localhost:11434
OLLAMA_NUM_CTX=16384

Recommended Ollama model pairs (by VRAM)

VRAMFastRobustTime on 94 chapters
8 GBqwen2.5:7bgemma2:9b~2 h
12 GBqwen2.5:7bqwen2.5:14b~3 h
16 GBqwen2.5:14bgemma2:27b~5 h
24 GB+qwen2.5:14bgemma2:27b~2 h

Using a model larger than your VRAM forces CPU offload and runs 5 to 10 times slower.

Architecture

internal/
├── scraper/   Download chapters from a URL
├── reader/    Three-pass analysis: macro / scene / continuity
├── fixer/     Rewrite inconsistent passages
├── writer/    Interactive writer (story from scratch)
├── models/    WorldBible, Character, Chapter, etc.
└── claude/    Provider abstraction (Anthropic / Ollama), Fast/Robust pair

The scripts/ directory contains Python utilities for packaging the output as EPUB, PDF, and DOCX, plus a few one-off helpers for renumbering and renaming chapter files.

License

All rights reserved.

This source is published for reference and portfolio purposes only. No permission is granted to use, copy, modify, merge, publish, distribute, sublicense, or sell copies of the code, in whole or in part, without explicit written permission from the author.

Ver o original no GitHub

Bom para

Relacionados