inkos and ainovel-cli
Both implement multi-agent orchestration for autonomous novel generation with human review gates, but A uses an English-language workflow with audit/revise cycles while B uses a Chinese-language implementation—making them **competitors** serving different language markets with similar architectural approaches.
About inkos
Narcooo/inkos
Autonomous novel writing CLI AI Agent — agents write, audit, and revise novels with human review gates
Implements a 10-agent pipeline with structured state management (JSON delta-based updates with Zod validation) and selective context retrieval via SQLite time-series memory to handle multi-chapter scaling, while featuring built-in anti-AI-detection patterns, hook/debt tracking with explicit agenda planning, and conservative word-count normalization with rollback snapshots. Integrates with OpenAI/Anthropic/custom OpenAI-compatible providers, serves as an OpenClaw skill, and supports local model inference via proxy endpoints with automatic stream degradation.
About ainovel-cli
voocel/ainovel-cli
多agent实现全自动AI小说生成
Multi-agent architecture with specialized roles (Coordinator, Architect, Writer, Editor) that orchestrate long-form novel generation through a Scaffolding + Harness runtime—separating startup assembly from execution control, enabling deterministic state management, chapter-level checkpointing, and recovery across planning/writing/review phases. Implements progressive arc-based rolling planning (initial 2-arc skeleton + first arc detail, expanding on-demand) with three-tier context compression (chapter/arc/volume summaries), intelligent chapter relevance recommendation across four dimensions (foreshadowing, character emergence, state changes, relationships), and seven-dimensional quality review with evidence citation. Integrates with OpenRouter, Anthropic, Gemini, OpenAI, DeepSeek, Qwen, Ollama, and custom proxies; supports real-time user intervention injection during authoring with automatic impact assessment and chapter rewriting.
Scores updated daily from GitHub, PyPI, and npm data. How scores work