Getting Started

Lightwell Documentation

Lightwell is a knowledge operating system: it transforms accumulated human knowledge — books, research, decision records, formal systems — into executable, verifiable intelligence.

What is Lightwell?

Most knowledge tools help you store what you learn. Lightwell helps you run it. A source enters the pipeline as raw material; the platform extracts its underlying machinery, compiles that machinery into a formal model, and projects it into skills and agents that can be executed, tested, and verified.

The system is built around one discipline: no claim ships without verification. Every stage of the pipeline has gates — extraction fidelity checks, formal type checking, execution grading, and independent review — so the output is trustworthy by construction, not by assertion.

The pipeline at a glance

Choose a source

Start from a book, a corpus, a person’s recorded decisions, or any body of knowledge you want to make executable. Lightwell treats sources as input to a compiler, not as static text.

Extract the machinery

The Cangjie Extractor and related pipelines pull out the generative machinery of the source: mental models, decision functions, frameworks, and anti-patterns — not summaries.

Compile to a formal model

Extractions are normalized into typed, machine-checkable structures with explicit invariants, so knowledge can be verified, versioned, and composed.

Generate skills & run them

Compiled models project into executable skills and agents that run against real tasks — with verification gates before anything is trusted.

Quickstart

The fastest path through the platform is the Knowledge Compiler. Start with a single, well-understood source, run the full pipeline, and inspect every intermediate artifact before projecting it into a skill.

quickstart — conceptual flow

1. Open the Knowledge Compiler  ->  /compiler
2. Select or upload a source   ->  book, corpus, or person-corpus
3. Run extraction              ->  review the extracted machinery
4. Compile                     ->  formal model with typed invariants
5. Verify                      ->  run the verification gates
6. Emit a skill                ->  project the model into /skills

Every artifact in the pipeline is inspectable. If a stage fails verification, the failure is surfaced with the evidence that produced it — the platform never silently promotes unverified output.

Core concepts

SourceThe raw material: a text, corpus, or person-corpus of recorded decisions and reasoning.

ExtractionThe front-end pass that pulls cognitive decision functions and generative machinery out of a source.

Formal modelThe typed, machine-checkable representation of what was extracted — the compile target.

InvariantA checkable property the model must hold true. Invariants are the unit of trust in Lightwell.

SkillAn executable projection of a formal model, runnable against real tasks behind verification gates.

Explore the tools

Knowledge Compiler

The core pipeline: source in, executable intelligence out.

OCompressor

Compression and normalization of extracted knowledge.

Cangjie Extractor

Front-end extraction of generative machinery from sources.

Architecture Library

Browse recovered and verified formal architectures.

Leonardo Lab

Evolutionary optimization and experimentation workspace.

Skill Generator

Project compiled models into runnable skills.

Open science, verified in public

Read the research behind the platform, including published whitepapers with formal derivations.

View Research