NovelClaw

NovelClaw is a structured long-form fiction writing workspace designed for chapter-based writing, observability, manuscript review, and memory-aware control. Unlike single-prompt tools, it organizes long-form creation into continuous sessions, run inspections, storyboards, and editable memory banks.

Key components include:

• Writing Workspace: Integrates conversation, chapter writing, run review, and continuation control.
• Manuscript Interface: Provides persistent access to storyboard, manuscript, style, world, and character data.
• Memory Control: Features an editable memory bank and chapter outputs to maintain continuity and facilitate revisions.

NovelClaw provides a structured environment for long-form writing that avoids opaque prompt interactions by using a closed-loop architecture.

Workspace Capabilities

• Writing Sessions: Continuous writing sessions.
• Views: Manuscript and storyboard views.
• Workspaces: Dedicated interfaces for world, character, and style settings.
• Memory Management: Editable memory bank tools.

Inspectable Artifacts

• Logs: Worker logs and progress traces.
• Outputs: Chapter outputs and downloadable artifacts.
• Review: Job details and run review interfaces.

The repository is organized into functional layers:

• apps/auth-portal/: The public entry layer.
• apps/multiagent/: An optional layer for fast ideation.
• apps/novelclaw/: The main writing workspace.
• scripts/: Helpers for local bootstrapping, starting, and stopping services.
• docs/: Screenshots and operator notes.
• infra/: Infrastructure examples (e.g., nginx/ for reverse-proxy, systemd/ for service units, and env/ for safe environment templates).
• state_snapshots/: Intended for public-safe releases (cleared regularly).

NovelClaw is a structured long-form fiction workspace designed for chapter-based writing, memory-aware drafting, and inspectable runs. Unlike typical fiction tools that treat writing as a single large prompt, NovelClaw treats it as an ongoing workspace.

Key Differences

NovelClaw uses an inspectable loop architecture designed to support long-form writing projects. Instead of a single opaque prompt interaction, it links several stages together:

1. Public Entry: The initial interface for users.
2. Writing Sessions: Active sessions where content is generated.
3. Run Execution: The actual processing of writing tasks.
4. Manuscript Review: Surfaces for reviewing generated content.
5. Memory-Aware Continuation: Using memory banks to ensure continuity in long projects.

This loop allows the workspace to maintain context (world, character, and style) across extended writing sessions without losing coherence.

NovelClaw uses a multi-route architecture where different components are exposed at specific URL paths. When deploying, treat /claw/ as the primary application route.

• / -> Portal (public entry point)
• /multiagent/ -> MultiAgent (optional support workspace)
• /claw/ -> NovelClaw (main writing workspace)

The provided systemd units and deployment guides assume a specific directory layout on your server. Ensure your files are organized as follows:

Application Files:

• /opt/novelclaw/.venv/ (Python virtual environment)
• /opt/novelclaw/apps/ (Contains auth-portal/, multiagent/, and novelclaw/)

Configuration Files:

• /etc/novelclaw/ (Contains .env files for each service)

text
/opt/novelclaw/
|-- .venv/
`-- apps/
    |-- auth-portal/
    |-- multiagent/
    `-- novelclaw/

/etc/novelclaw/
|-- auth-portal.env
|-- multiagent.env
`-- novelclaw.env

NovelClaw is organized into three main functional layers, each mapped to a specific URL path. When deploying to a production server, /claw/ is considered the primary application route.

• / -> Portal (Public entry point)
• /multiagent/ -> MultiAgent (Optional auxiliary workspace)
• /claw/ -> NovelClaw (Main writing workspace)

When running locally, the following defaults are applied to simplify the preview experience:

• Embeddings: EMBEDDING_MODEL=none and DISABLE_EMBEDDING_DOWNLOADS=1 are set by default to prevent heavy embedding downloads during the preview phase.
• Authentication: The public version does not expose registration, login, password recovery, or email verification flows.
• Web Server: Local preview does not require Nginx.

NovelClaw uses Docker volumes to ensure data is not lost when containers are stopped or removed. The following directories are mounted for persistence:

• Auth Portal Database: apps/auth-portal/local_web_portal/data
• MultiAgent Data: apps/multiagent/local_web_portal/data
• MultiAgent Runs/Outputs: apps/multiagent/local_web_portal/runs
• NovelClaw Database: apps/novelclaw/local_web_portal/data
• NovelClaw Runs/Outputs: apps/novelclaw/local_web_portal/runs

Create the configuration directory and copy the provided templates to /etc/novelclaw/. You must manually edit these files to include real production values.

Required configuration items to fill in:

• Real domain names and public URLs
• session secret and encryption key
• Database URL
• API keys or provider configurations
• APP_BASE_PATH (must match your deployment routes)

sudo mkdir -p /etc/novelclaw
sudo cp infra/env/auth-portal.env.template /etc/novelclaw/auth-portal.env
sudo cp infra/env/multiagent.env.template /etc/novelclaw/multiagent.env
sudo cp infra/env/novelclaw.env.template /etc/novelclaw/novelclaw.env