AuthorAgent

The free, open-source AI writing agent that runs the entire book pipeline on your own machine.

Every author wants an agent. This one works for your book — not for 15%.

AuthorAgent takes an idea (or your existing manuscript) through research → outline → write → revise → format → publish-ready in one autonomous pipeline, on your computer, with your own AI keys. What a $2,000/year stack of subscriptions tries to do, one free tool does — without your manuscript ever leaving your machine.

Three things that make it different:

1. Your book never leaves your machine. Local-first by design. The only network calls are the AI API calls you configure with your keys. No cloud account, no manuscript upload, no "we may use your content to improve our services."
2. One free tool, not five subscriptions. Writing assistant + developmental editor + formatter + market research + launch planner, in one open-source agent. Pay only your own API usage — and the router spends it wisely (free models for structure, premium only where prose quality pays).
3. Built to fight AI slop, not produce more of it. A quality engine no other tool has: evidence-chained contradiction detection, per-character voice critics, anti-slop screening, a judge→reflect→revise evolution loop — and it learns from its own findings, injecting durable lessons into future writing.

The Conductor's Studio

Open your book and watch the agent work:

• ✨ Start a Book wizard — idea or upload → genre, persona, length → mode: Full Book, or Just Outline / Just Write / Just Polish / Just Format
• The Book View — every book gets a journey bar (Plan → Write → Revise → Polish → Publish) showing honest progress, chapter-by-chapter
• The Orchestra — while the pipeline runs, watch the conductor dispatch its workers live: 🔍 researching, 🗺 outlining, ✍️ drafting Chapter 3, 🧐 auditing pacing…
• Zero-paste quality checks — every chapter gets a "Check ▾" menu: contradictions, character critique, revision analysis — pre-filled from the manuscript the agent already holds
• 💬 Chat drawer — talk to your agent from anywhere in the app

The Quality Engine

This is the part built to make books better, not just faster:

Long-Book Memory

The unsolved problem of AI writing is book-length consistency. AuthorAgent ships a tiered memory system:

• CORE — a budgeted, always-in-prompt digest: active chapter state, the characters this scene touches, open plot threads, your style fingerprint
• ARCHIVAL — full-text search over the entire manuscript and past work, pulled in on demand ("what happened in chapter 12?" actually works)
• RECALL — raw drafts on disk, the source of truth
• Sleep-time consolidation — a nightly job (free-tier models only) re-summarizes character arcs, resolves plot threads, distills your voice, and prunes noise — so the agent wakes up knowing your book

Everything Else in the Box

• Author personas — multiple pen names with distinct voice fingerprints and drift detection
• Publishing suite — 90-day launch orchestrator, AMS ads optimizer, BookBub deal builder, translation planner + executor, author-site deploys, blog drafter
• Craft tools — 10 story structures with outline checking, plot-promise (Chekhov's gun) tracking, series bible for multi-book continuity
• Export — KDP-ready DOCX and valid EPUB3
• Audiobook prep — script cleanup, multi-voice speaker attribution, free neural TTS (9 presets) or ElevenLabs
• Covers — image generation chain (OpenAI → Gemini → Flux) with cover-set output (ebook/print/audiobook/social)
• Telegram bridge — run your whole pipeline from your phone
• Writing momentum — streaks and word counts tracked automatically as the agent writes

Quick Start

git clone https://github.com/Ckokoski/authoragent.git
cd authoragent
npm install
npm start

# Open http://localhost:3847
# Settings → Connections → paste a Gemini API key (free tier — a whole book can cost $0)
# Click ✨ Start a Book

First run: AuthorAgent generates a vault encryption key and stores it outside the repo (%LOCALAPPDATA%\AuthorClaw\vault.key on Windows, ~/.authorclaw/vault.key elsewhere) so your encrypted API keys never sync to cloud drives. The Getting Started checklist on the home screen walks you through the rest.

See QUICKSTART.md for the full guide.

AI Providers — bring your own model

Per-stage cost routing is automatic: outlines and research use free models, drafting uses mid-tier, final polish and judging use premium — and every provider's model is a setting, not code (Settings → AI Models), so new models are a dropdown away with cost tracking that follows the model.

Dashboard

Six panels, no clutter:

• Home — Start a Book, writing momentum (streak/words), your active books with "Continue writing →", getting-started checklist
• Books — your library; each book opens the journey view
• Personas — pen names, voices, per-persona activity
• Library — uploads and compiled outputs
• Tools — every tool available standalone (Revise & Critique / Craft / Publishing & Marketing) for one-off use outside a book
• Settings — Connections, Integrations, Automation, Data & Memory, Preferences

Security

• Vault: AES-256-GCM encrypted keys (scrypt KDF), key stored outside the repo and outside cloud-sync folders
• Path sandbox: single hardened path-safety module on every file route
• Injection detection: scoped so your fiction never gets blocked ("You are now in the throne room" is prose, not an attack)
• Audit log: JSONL trail of every agent action
• Localhost only: binds to 127.0.0.1 — nothing exposed
• Skill/write protections: confirmation gates on external actions

For VM/VPS/Docker deployment (recommended for always-on Telegram use), see SECURITY.md and the docker/ directory.

Development

npm run check   # typecheck + 410 tests
npm run dev     # auto-reload dev server

The codebase is modular TypeScript: domain-split API routes (gateway/src/api/routes/), service container, message pipeline, and a vitest suite covering the security, routing, memory, and quality subsystems.

Contributing

Contributions welcome — new skills (skills/*/SKILL.md, YAML frontmatter + markdown), bug fixes, new providers, bridges, dashboard improvements, docs. Fork → feature branch → npm run check → PR.

Disclaimer

Provided "as is," use at your own risk. AI output should always be reviewed by a human before publishing. Third-party AI usage is subject to those providers' terms and pricing; API costs are your responsibility.

License

MIT. See LICENSE.

Built by an author, for authors — on the belief that AI should amplify creativity, not replace it, and that your agent should work for your book, not for a subscription.