← all repositories
Narcooo/inkos

An AI agent that writes pulp fiction while you sleep

InkOS automates the full novel pipeline—draft, audit, revise—with human approval gates so you don't publish AI slop by accident.

7k stars TypeScript AgentsCreative · Design
inkos
Velocity · 7d
+79
★ / day
Trend
steady
star history

What it does

InkOS is a TypeScript CLI and local web studio that runs autonomous AI agents to write, audit, and revise novels. It targets Chinese web-fiction genres (xuanhuan, xianxia, urban, sci-fi) but recently added native English support via --lang en. The workflow is deliberate: agents draft chapters, other agents audit them, then revisions happen according to your config. Human review gates sit between each stage so you can approve, reject, or redirect before anything ships.

The interesting bit

The project treats LLM configuration as a systems problem, not a footnote. Studio (web UI) and CLI have isolated config paths—Studio uses a visual provider bank and project-scoped secrets, while CLI/daemon environments layer env vars and command-line overrides with a documented precedence chain. There’s even model routing per agent role (writer vs. auditor) and explicit validation that prevents you from accidentally sending a Moonshot model to Google’s endpoint. For a creative-writing tool, the infrastructure rigor is almost suspicious.

Key highlights

  • Three interfaces sharing one execution kernel: TUI (inkos tui), Studio 2.0 (inkos web UI), and OpenClaw Skill (inkos interact --json)
  • Built-in provider bank with compatibility shims for Gemini, Moonshot, MiniMax, DeepSeek, OpenRouter, Ollama, and others
  • Short-form mode generates standalone stories with cover prompts and sales packaging in one shot
  • Exports to EPUB for Kindle/mobile reading
  • AGPL-3.0 licensed

Caveats

  • README is primarily in Chinese; English documentation exists but is secondary
  • The project self-describes as updating frequently with ongoing feature additions, so interfaces may shift
  • “AI detection” and “market radar” features are mentioned but not detailed in the provided sources

Verdict

Worth a look if you’re writing serial fiction at volume and want structured automation without surrendering editorial control. Skip it if you need mature English-language documentation or are hoping for literary fiction nuance—the examples lean heavily toward genre pulp and web-novel conventions.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.