SpineDigest addresses the challenge of distilling long-form books into digestible formats, overcoming LLM context window limitations. It targets engineers, researchers, and power users by providing not just text summaries but also structured outputs like chapter topologies and knowledge graphs, making complex information navigable and adaptable to specific user intents.

How It Works

SpineDigest employs a multi-stage pipeline. Initially, an LLM processes the source text section by section, extracting discrete "chunks" of information, with attention guided by user-defined goals. Subsequently, a classical algorithm constructs a knowledge graph where chunks are nodes connected by conceptual relevance. Graph traversal and community detection then cluster related chunks into "snakes"—threaded knowledge chains. The final summarization phase utilizes an adversarial Multi-Agent LLM framework, pitting a respondent against a panel of "professors" (each holding a snake), simulating a dissertation defense to refine the summary based on the initial extraction goal and ensure comprehensive representation of the source material.

Quick Start & Requirements

Installation can be done globally via npm install -g spinedigest or locally using npx spinedigest. Prerequisites include Node.js version 22.12.0 or higher. Source digestion requires a supported LLM provider with valid credentials; however, .sdpub file re-export or inspection does not necessitate LLM access. Extensive CLI guidance is available via spinedigest --help and related help subcommands.

Highlighted Details

• .sdpub Format: Generates a processed archive (.sdpub) containing the summary, chapter topology, knowledge graph, and snakes, enabling re-export to EPUB, Markdown, or TXT without re-running the LLM pipeline.
• Inkora Viewer: A free application, Inkora, is available for visualizing .sdpub files, offering interactive chapter topology and knowledge graph views.
• Intent-Driven Processing: User-specified goals shape both the initial chunk extraction and the adversarial summarization process, ensuring outputs align with specific reading or analysis purposes.
• CLI-First Design: The project prioritizes a robust command-line interface for integration and discovery, with comprehensive --help documentation for all commands and features.

Maintenance & Community

No specific details regarding maintainers, community channels (e.g., Discord, Slack), or project roadmaps were found in the provided documentation.

Licensing & Compatibility

The provided README does not specify a software license. This absence may present compatibility issues for commercial use or integration into closed-source projects.

Limitations & Caveats

Initial book digestion requires configuration and access to an external LLM provider, potentially incurring costs and setup time. Exporting to formats like EPUB or Markdown results in the loss of the rich structural data (topology, graph, snakes) preserved within the .sdpub archive. The effectiveness of the summarization is dependent on the chosen LLM provider's capabilities.