The Autonomous AI Writing Agent — Built for Authors

AuthorClaw is a security-hardened AI agent purpose-built for fiction and nonfiction authors. It doesn't just write — it runs the entire book production pipeline autonomously, from first idea to KDP-ready manuscript.

Give it an idea and a pen name. It plans, writes, revises, formats, and launches. Pipeline mode chains 6 production phases automatically. Author personas manage multiple pen names with distinct voices. Deep revision runs 21 editing passes. Export produces professional DOCX and EPUB ready for self-publishing.

Tell it what you want. It figures out the steps, picks the right skills, and executes.

"It's not just a writing tool. It's a writing partner, research assistant, editor, and marketing team rolled into one."

• Pipeline — Turn one idea + one pen name into a finished book across 6 automated phases
• Write — Draft scenes, chapters, and full manuscripts in your persona's voice
• Revise — 21-step deep revision: 3 passes (structural → scene-level → line-level) + AI beta readers
• Plan — 6 project templates: Book Planning, Book Bible, Book Production, Deep Revision, Format & Export, Book Launch
• Personas — Manage multiple pen names with distinct genres, voices, style markers, and bios
• Research — Deep dives into genres, markets, historical periods, craft techniques
• Beta Read — AI beta reader panel (romance super-reader, harsh critic, casual reader)
• Market — Blurbs, ad copy, Amazon descriptions, keywords, social media launch posts
• Format — KDP-ready DOCX (trim sizes, front/back matter) and valid EPUB3 export
• Manage — Track projects, pipelines, word counts across pen names
• Listen — Neural TTS voice engine with 9 author-optimized presets — hear your writing read aloud

1. You say what you want — via Telegram, dashboard, or API
2. AuthorClaw plans the steps — AI dynamically decomposes your task into executable steps
3. Skills are auto-selected — 19 focused writing skills get injected into each step's context
4. Work happens autonomously — each step runs through the AI, output saved to files
5. Everything is logged — universal activity feed tracks all agent actions in real-time

User: "/novel a small-town romance under pen name Lily Hart"

AuthorClaw: "Pipeline created — 6 phases, 48 steps total"
  Phase 1: Book Planning    (6 steps)  — market analysis, premise, characters, outline, synopsis
  Phase 2: Book Bible       (5 steps)  — world-building, character bible, continuity, themes, style
  Phase 3: Book Production  (20 steps) — write + self-review per chapter
  Phase 4: Deep Revision    (21 steps) — 3-pass editing + AI beta readers
  Phase 5: Format & Export  (4 steps)  — front matter, back matter, DOCX, EPUB
  Phase 6: Book Launch      (6 steps)  — blurb, Amazon description, keywords, ad copy, social posts

  "Phase 1 started. Persona 'Lily Hart' context injected."

# 1. Clone and install
git clone https://github.com/Ckokoski/authorclaw.git
cd authorclaw
npm install

# 2. Start AuthorClaw (auto-generates vault key on first run)
npx tsx gateway/src/index.ts

# 3. Open dashboard: http://localhost:3847
#    Settings (sidebar) → paste your Gemini API key → Save
#    (Free tier — the whole book costs $0)

# 4. Home → chat: "Write me a thriller about rogue AI" → Send
#    OR send /project to your Telegram bot

First run? AuthorClaw auto-generates a vault encryption key and saves it to .env. Your API keys will persist across restarts. For a guided setup, run bash scripts/setup-wizard.sh.

See QUICKSTART.md for the full setup guide.

┌─────────────────────────────────────────────────────────────┐
│                    AUTHORCLAW v4 ARCHITECTURE                │
│                                                             │
│  ┌───────────┐   ┌─────────────────┐   ┌────────────────┐  │
│  │ Channels  │   │    Gateway       │   │  AI Router     │  │
│  │           │   │                  │   │                │  │
│  │ Telegram  │──▶│ Auth + Sandbox   │──▶│ Ollama (free)  │  │
│  │ Dashboard │   │ Rate Limiting    │   │ Gemini (free)  │  │
│  │ API       │   │ Injection Detect │   │ DeepSeek ($)   │  │
│  │ WebSocket │   │ Audit Logging    │   │ Claude ($$)    │  │
│  └───────────┘   └─────────────────┘   │ OpenAI ($$)    │  │
│                                         └────────────────┘  │
│  ┌───────────┐   ┌─────────────────┐   ┌────────────────┐  │
│  │ Soul      │   │ Project Engine  │   │ Skills (19)    │  │
│  │           │   │                  │   │                │  │
│  │ SOUL.md   │   │ 6 Templates     │   │ Core (4)       │  │
│  │ STYLE.md  │   │ Pipeline Mode   │   │ Author (13)    │  │
│  │ VOICE.md  │   │ Author Personas │   │ Marketing (2)  │  │
│  │           │   │ DOCX + EPUB     │   │                │  │
│  └───────────┘   └─────────────────┘   └────────────────┘  │
│                                                             │
│  ┌───────────┐   ┌─────────────────┐   ┌────────────────┐  │
│  │ Security  │   │ Smart Agent     │   │ Research Gate  │  │
│  │           │   │                  │   │                │  │
│  │ Vault     │   │ Priority Scoring│   │ Web Search     │  │
│  │ Sandbox   │   │ Self-Improve    │   │ HTML Extraction│  │
│  │ Audit     │   │ Agent Journal   │   │ Domain Allowlist│  │
│  │ Injection │   │ Sub-Projects    │   │ Rate Limiting  │  │
│  └───────────┘   └─────────────────┘   └────────────────┘  │
└─────────────────────────────────────────────────────────────┘

AuthorClaw supports 5 AI providers with tiered routing:

Task routing is automatic — planning and research use free models, creative writing uses mid-tier, final editing uses premium (when available).

Connect a Telegram bot to control AuthorClaw from your phone:

You:        /write a short snarky YouTube intro for my channel
AuthorClaw: 📝 On it. Planning "a short snarky YouTube intro"...
AuthorClaw: ✅ Planned 2 steps. Running autonomously...
AuthorClaw: ✅ 1/2: Draft the intro (~400 words)
AuthorClaw: 🎉 All 2 steps complete!

You:        /novel a sci-fi thriller about rogue AI in aviation
AuthorClaw: 📖 Novel pipeline created: 32 steps
            (premise → bible → outline → chapters → revision → assembly)
AuthorClaw: ✅ 1/32: Develop premise (~800 words)
            ⏭ Next: Refine premise...
You:        /stop
AuthorClaw: ⏸ Paused at step 4/32. Say "continue" to resume.

You:        /clean
AuthorClaw: 📊 Workspace Usage: 2.1 MB (67 files)
            📁 projects: 43 files (1.8 MB)
            📁 exports: 2 files (21 KB)
            🧹 /clean projects — delete all project files

Open http://localhost:3847 to access the web dashboard — a sidebar-driven interface with 5 panels:

• Home — Quick stats (words today, active projects, heartbeat status, personas), active project cards, full chat interface with slash command parity (all Telegram commands work in chat), today's writing progress bar, idle task count
• Projects — 7 template tiles (Book Planning, Book Bible, Book Production, Deep Revision, Format & Export, Book Launch, Full Novel Pipeline) + Custom AI-planned. Projects auto-execute on creation — no manual start needed. Project list with status filters, inline detail views with step progress, file downloads (MD + DOCX), and compile controls
• Personas — Author persona card grid with pen names, genres, style tags, and TTS voice. Create manually or generate with AI. Assign personas to projects for voice-consistent writing. Personas persist across updates with auto-backup
• Library — Document uploads and compiled manuscripts. Download DOCX and EPUB exports
• Settings — API keys (vault-encrypted), Telegram bot config, voice/TTS presets, research domain allowlist, autonomous heartbeat mode with configurable word goal, editable idle task queue (CRUD)

AuthorClaw includes a built-in neural voice engine powered by Microsoft Edge TTS — no API keys, no binary installation, no cost.

9 author-optimized voice presets:

Telegram voice features:

• /speak Hello world — Generate and send a voice message
• /speak narrator_deep In a world... — Use a specific voice
• /speak 3 — Read file #3 from your last /files listing aloud
• /voice on — Toggle voice mode (all chat replies become voice + text)
• /voice narrator_deep — Set voice mode with a specific preset
• "Read that back" — Re-read the last response as voice

API: POST /api/audio/generate with { text, voice, rate, pitch, volume }

⚠️ Audio files are automatically deleted after 24 hours. If you generate a voice file you want to keep (e.g., a narration of your chapter), save or download it before the auto-cleanup runs. Use /clean audio to clear them manually, or find them in workspace/audio/.

AuthorClaw supports uploading manuscripts of any size — from short stories to 100K+ word novels.

Two-tier upload system:

How smart excerpts work for large manuscripts:

• The first ~4,000 words (setup, voice, style) and last ~1,000 words (current state) are sent to the AI
• A truncation marker tells the AI the full document is available on disk
• This keeps AI context manageable while giving it enough to work with
• The full manuscript is always saved in workspace/documents/ for reference

Document Library API:

• GET /api/documents — List all documents in the library
• POST /api/documents/upload — Upload directly to the library (up to 50MB)
• DELETE /api/documents/:filename — Remove a document

Dashboard: Upload files via the Projects tab (Upload button). Large files are automatically saved to both the project and the central library.

When you give AuthorClaw a task, it doesn't use hardcoded templates. Instead:

1. The AI receives a catalog of all available skills (with descriptions and triggers)
2. The AI receives the list of Author OS tools
3. The AI dynamically plans the right number of steps, picks the right skills (19 focused) for each
4. Each step is executed with that skill's full content injected into the AI's context
5. Results from earlier steps are chained into later steps for continuity

If AI planning fails, the system falls back to template-based planning (6 project types with pre-built step sequences). For pipeline mode, AuthorClaw chains all 6 phases (Planning → Bible → Production → Revision → Format → Launch) into a single automated workflow, passing outputs forward between phases.

Skills are markdown files that teach the AI how to handle specific writing tasks. V4 ships with 19 focused, author-centric skills:

Core Skills (4): self-improve, after-action-review, prompt-optimizer, error-recovery

Author Skills (13): premise, outline, book-bible, write, revise, dialogue, style-clone, beta-reader, format, research, nonfiction-research, manuscript-hub, ingest-tool

Marketing Skills (2): blurb-writer, ad-copy

Tool Ingestion: AuthorClaw can read source code of any tool and generate a new skill from it. Just say "create a skill from this code" or use POST /api/tools/ingest.

Skills are automatically matched by keyword triggers and injected into the AI's context. A full reference with descriptions and example trigger keywords is available in workspace/SKILLS.txt.

authorclaw/
├── gateway/src/          # Core application
│   ├── index.ts          # Main entry point (gateway, handlers, bridges)
│   ├── ai/router.ts      # Multi-provider AI routing
│   ├── api/routes.ts     # REST API endpoints (projects, personas, pipeline, export)
│   ├── bridges/          # Telegram, Discord bridges
│   ├── security/         # Vault, audit, sandbox, injection detection
│   ├── services/         # Memory, soul, projects, personas, research, heartbeat
│   │   ├── projects.ts   # Project engine (6 templates, pipeline mode)
│   │   ├── personas.ts   # Author persona management
│   │   ├── docx-export.ts # KDP-ready DOCX generation
│   │   └── epub-export.ts # EPUB3 generation
│   └── skills/loader.ts  # Skill loading and matching
├── skills/               # Skill definitions (SKILL.md files)
│   ├── core/             # System skills (4)
│   ├── author/           # Writing skills (13)
│   ├── marketing/        # Marketing skills (2)
│   └── _archived/        # Deprecated V3 skills (reference only)
├── dashboard/dist/       # Web dashboard (single HTML file, sidebar layout)
├── workspace/            # Working directory
│   ├── soul/             # SOUL.md, STYLE-GUIDE.md, VOICE-PROFILE.md
│   ├── memory/           # Conversations, book bible, summaries
│   ├── projects/         # Project output files organized by project
│   ├── documents/        # Document library (large manuscripts, novels)
│   ├── research/         # Research output files
│   ├── .config/          # Persona data, pipeline state
│   ├── .agent/           # Agent journal, self-improve logs
│   ├── audio/            # Generated TTS voice files (auto-cleaned after 24hr)
│   ├── SKILLS.txt        # Full skill reference (auto-generated on startup)
│   ├── .activity/        # Universal activity log (JSONL)
│   └── .audit/           # Security audit log (JSONL)
├── config/               # Configuration files
│   ├── default.json      # Main config
│   ├── .vault/           # Encrypted API key storage
│   └── research-allowlist.json  # Approved research domains
└── scripts/              # Utility scripts

AuthorClaw security features:

• Vault: AES-256-GCM encrypted credential storage (scrypt KDF)
• Sandbox: Workspace-only file access enforcement
• Audit: Daily JSONL logs with categories (message, security, error, connection)
• Injection Detection: Pattern matching for prompt injection attempts
• Rate Limiting: Per-channel rate limits
• Research Gate: Real web search + HTML extraction, 50+ allowlisted domains, 60 req/hr rate limit
• Localhost Only: Server binds to 127.0.0.1 (no external access)

We strongly recommend running AuthorClaw inside a VM or VPS with Docker. Your API keys, manuscripts, and creative work deserve real protection. Defense in depth means multiple security layers — not just application-level security.

This is the gold standard for always-on, secure operation:

1. Rent a VPS ($5-6/month) — Hetzner, DigitalOcean, or Linode
2. Install Docker — containerizes AuthorClaw with strict resource limits
3. Install Tailscale — free mesh VPN, no public ports exposed
4. Deploy AuthorClaw — docker compose up -d

# On your VPS:
curl -fsSL https://get.docker.com | sh
curl -fsSL https://tailscale.com/install.sh | sh
tailscale up

# Clone and deploy:
git clone https://github.com/Ckokoski/authorclaw.git
cd authorclaw/docker
docker compose up -d

Why this matters:

• VPS isolates AuthorClaw from your personal machine
• Docker containers limit file access and resource usage
• Tailscale VPN means zero public ports — only your devices can connect
• Telegram works 24/7 even when your computer is off
• Your manuscripts and API keys never leave the VPS

If you prefer running locally:

1. VirtualBox/UTM — free VM software
2. Ubuntu 24.04 — lightweight Linux inside the VM
3. Run AuthorClaw natively or with Docker inside the VM

# In your VM:
bash /media/sf_authorclaw-transfer/run.sh

Why a VM helps:

• Isolates AuthorClaw from your host OS
• If something goes wrong, the VM is disposable
• Shared folders let you copy files in/out safely
• Snapshots let you roll back to a known-good state

Running directly on your machine works fine for development and testing:

git clone https://github.com/Ckokoski/authorclaw.git
cd authorclaw && npm install
npx tsx gateway/src/index.ts

AuthorClaw binds to localhost:3847 only — not exposed to the internet. But your API keys and manuscripts live on your main OS with no isolation layer.

For a guided setup experience, run the interactive wizard:

bash scripts/setup-wizard.sh

It walks you through everything: OS detection, Node.js installation, Ollama setup, API key configuration, vault passphrase creation, and personalization (genre, word goals). It even generates a troubleshooting prompt you can paste into any AI chatbot if you get stuck.

AuthorClaw is open source and contributions are welcome! Whether you're an author with ideas for new skills, a developer who wants to improve the codebase, or a tinkerer who built a cool integration — we'd love your help.

• New Skills — Create SKILL.md files for writing tasks we haven't covered yet
• Bug Fixes — Find and fix issues in the gateway, dashboard, or bridges
• New AI Providers — Add support for additional AI services
• New Bridges — Build integrations for Slack, WhatsApp, Matrix, etc.
• Dashboard Improvements — The dashboard is a single HTML file — lots of room to grow
• Documentation — Better guides, tutorials, and examples

1. Fork the repository
2. Create a feature branch (git checkout -b feature/my-new-skill)
3. Make your changes
4. Test locally (npx tsx gateway/src/index.ts)
5. Submit a Pull Request with a clear description

For new skills, create a folder in skills/author/, skills/marketing/, or skills/core/ with a SKILL.md file following the existing format (YAML frontmatter + markdown body).

This software is provided "as is" without warranty of any kind. Use at your own risk. AuthorClaw is an experimental AI writing tool — some configuration and code tinkering may be required to get the agent working exactly the way you want it. AI outputs should always be reviewed by a human before publishing. The authors are not responsible for any content generated by the AI or any consequences of using this software.

AuthorClaw relies on third-party AI providers (Gemini, Claude, OpenAI, DeepSeek, Ollama). Usage of those services is subject to their respective terms and pricing. API costs are your responsibility.

MIT License. See LICENSE for details.

Built with love for writers by an author who believes AI should amplify creativity, not replace it.