Why Velith?

Most AI "writing tools" are a single prompt and a blank page — they give you disconnected chapters, drifting voice, and no structure. Velith is the opposite: a plan-then-execute pipeline that validates before it writes, gates quality at every stage, and keeps the whole manuscript coherent end to end.

You wouldn't ship code without a spec, tests, and review — so don't ship a book (or an RFC, or a whitepaper) without an outline, continuity checks, and an edit pass. Velith brings software-engineering discipline to long-form knowledge.

Benchmark

What the pipeline does to unstructured input — try it yourself →

One pipeline, many artifacts

Velith ships as a book pipeline — but the same 6 phases apply to any long-form structured knowledge. It doesn't matter whether the artifact is a 300-page novel or a 12-page RFC; the plan-then-execute flow, quality gates, and agents are identical.

Comparison

Installation

Claude Code

/plugin marketplace add epicsagas/plugins
/plugin install velith@epicsagas

All 17 skills and 8 agents are available immediately. No further steps needed.

Updates with /plugin update velith@epicsagas.

Prerequisites: Claude Code CLI installed and authenticated.

Codex CLI (OpenAI)

codex plugin marketplace add epicsagas/plugins

Velith provides 17 skills (via .agents/skills/) and 8 custom subagents (via .codex/agents/):

Codex auto-discovers skills from .agents/skills/ and subagents from .codex/agents/*.toml. No extra configuration needed.

Updates with codex plugin update velith@epicsagas.

Prerequisites: Codex CLI installed and configured with an OpenAI API key.

Agy (Antigravity)

agy plugin install https://github.com/epicsagas/Velith

Agy auto-discovers skills and agents from the repository root. No extra configuration needed.

Prerequisites: Agy installed and configured.

Cursor

Velith provides context rules in .cursor/rules/ that give Cursor's agent full awareness of the book publishing pipeline, genre patterns, and editing standards.

Rules are automatically loaded when you open a Velith book project in Cursor. No installation needed — just clone or copy the .cursor/rules/ directory into your project.

Prerequisites: Cursor installed.

Cline

Velith provides project-level instructions in .clinerules at the repository root. Cline reads this file automatically when working in the project directory — no extra configuration needed.

Prerequisites: Cline extension installed in VS Code or JetBrains.

Aider

Velith provides writing conventions in CONVENTIONS.md, auto-loaded via .aider.conf.yml.

aider  # CONVENTIONS.md is auto-loaded

Prerequisites: Aider installed and configured with an API key.

Quick Start

# Start a new book project
> /book-init

# Auto-detect your current phase and continue
> /loom

The plugin guides you through:

1. Onboarding — Genre, audience, language, source material, style guide
2. Ideation — Market research, concept distillation, competing titles
3. Outlining — Full chapter outline with specs, dependencies, cross-references
4. Drafting — Chapter-by-chapter generation with parallel subagents
5. Editing — 5-stage pipeline: Assessment → Developmental → Line → Copy → Proofread
6. Publishing — EPUB/PDF/MOBI conversion, metadata, marketing plan

Skills

Agents

Visual Dashboard

/book-status --ui opens a Svelte-based progress dashboard in your browser. The dashboard auto-refreshes every 5 seconds:

• 6-phase pipeline tracker (Onboarding → Ideation → Outlining → Drafting → Editing → Publishing)
• 8 agent status cards (book-architect, chapter-writer, continuity-editor, cover-designer, illustrator, marketing-expert, scene-generator, style-doctor)
• Chapter outline, drafts table, and 5-stage editing kanban
• Output file status (EPUB/PDF/MOBI/TXT/MD) with publish checklist
• Project settings and command reference

The dashboard reads from per-project status.json files dynamically. The pre-built dist/ is included — no build step required for plugin users.

To run locally for development:

cd dashboard
npm install
npm run dev     # http://localhost:5173
npm run build   # rebuild dist/

Design Principles

• Plan-Then-Execute — Outline first, validate, then write
• Idempotent — Skip completed chapters, resume from where you left off
• Token Efficient — Summary-based context, not full text
• Genre-Aware — Different structures, templates, and validation per genre
• Quality Gated — Each phase must pass criteria before proceeding

External Dependencies

For EPUB/PDF output (Phase 5):

brew install pandoc        # EPUB/PDF conversion
brew install texlive       # PDF with CJK/Korean support
brew install --cask calibre  # MOBI (Kindle) conversion — optional

Troubleshooting

brew install pandoc

brew install texlive
# Or for minimal install:
brew install basictex && sudo tlmgr install collection-langkorean

claude restart

Project Structure

When you create a book project, Velith sets up:

{project-dir}/
├── PRD.md          # Book requirements
├── STYLE.md        # Voice, tone, conventions
├── ideation.md     # Ideas, market research
├── outline.md      # Full chapter outline
├── drafts/         # Chapter drafts
│   ├── ch00-foreword.md
│   ├── ch01-xxx.md
│   └── ...
├── edits/          # Editing reports
│   └── editorial-report.md
├── publish/        # Final outputs
│   ├── book.epub
│   ├── book.pdf
│   ├── book.mobi
│   └── metadata.yaml
└── sources/        # Source material references

Integration

Built-in agent workflows

No extra setup — these run automatically during the pipeline:

• discover — During /book-outline, book-architect probes for blind spots and contradictions in the book concept before the structure is locked
• council — During /book-outline and /book-edit, multiple editorial perspectives (developmental, structural, line-edit) are weighed for outline and revision decisions

alcove — Research vault as source material

alcove is a private document server that lets Velith agents read your existing notes, research, and project docs as source material during drafting.

When it helps:

• You have years of research notes, interview transcripts, or reference docs you want the agent to cite from
• You're writing non-fiction and need agents to pull facts from structured project documentation
• You maintain a knowledge base with terminology, timelines, or world-building details the agent should respect

How to use:

1. Install and configure alcove as an MCP server in your Claude Code settings
2. During /book-init, point to your alcove project as a source
3. Agents will query alcove automatically when drafting chapters that reference your research

obsidian-forge — Write where you think

obsidian-forge bridges your Obsidian vault to Velith, so you can research in Obsidian and write with Velith without copying files manually.

When it helps:

• Your research, character profiles, and reference notes already live in an Obsidian vault
• You want to iterate on outlines in Obsidian's linked-note environment before committing to Velith
• You collaborate with co-authors who prefer Obsidian for brainstorming

How to use:

# Create a book project inside your Obsidian vault (01-Projects/)
of book init my-book --genre non-fiction --lang ko

# Work in Obsidian: research notes, character profiles, references
# Tag notes with book/my-book to link them as source material
of book sync my-book

# Export to a standalone directory when ready to write
of book export my-book --output ~/projects/my-book

# Now run velith on the exported project
> /loom

Both alcove and obsidian-forge are optional — Velith works standalone.

Contributing

See CONTRIBUTING.md. PRs welcome — check open issues labeled good first issue.

License

Apache-2.0