Skip to content

tomazinho/open-ckf-compiled-knowledge-format

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

76 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Open CKF

Standalone, open-source Compiled Knowledge Format (CKF) tools. Two browser-only apps you can host anywhere — GitHub Pages, S3, Netlify, a VPS, or even a USB stick.

This repository contains the same engine that powers compiledknowledgeformat.org, repackaged with zero backend, zero database, zero login.

Downloads

Download Release

Packages

Package What it does Backend?
packages/ckf-viewer Drop a .ckf.json file and explore its sections, rules, definitions, and source traceability. None
packages/ckf-compiler Compile raw text into a CKF package. Heuristic mode runs fully offline; BYOK mode calls OpenAI / Anthropic / Google / DeepSeek / OpenRouter directly from your browser. None — your API key stays in localStorage

Quick start

git clone https://github.com/tomazinho/open-ckf-compiled-knowledge-format
cd open-ckf-compiled-knowledge-format/packages/ckf-viewer
npm install && npm run dev
# or, in packages/ckf-compiler

To build a static bundle:

npm run build
# -> dist/ ready for any static host or `python -m http.server`

When to use Open CKF vs the hosted site

Need Use
Try CKF without installing anything compiledknowledgeformat.org
Run on your own infra / behind a VPN Open CKF
100% offline / air-gapped Open CKF (heuristic mode)
White-label deployment Open CKF (MIT licensed)
Advanced AI free credits Hosted site (admin allowlist)

Research

This directory collects the research papers, benchmark protocols, technical reports, and synthesis documents behind Open CKF.

The materials are organized as a research timeline: from the motivating failure mode, through CKF compiler evolution, robustness studies, model sensitivity analyses, synthesis papers, operational applications, and legal-domain specialization.

Research Timeline

Phase Folder Focus
01 01-foundation-composition-hallucination Defines composition hallucination, the RAG failure mode that motivates CKF
02 02-baseline-format-semantic-preservation Shows why plain JSON/YAML are not enough for semantic preservation
03 03-compiler-evolution Documents the evolution of the CKF compiler architecture
04 04-pipeline-robustness Studies end-to-end effects of PDF extraction, chunking, and pipeline observability
05 05-model-sensitivity Compares how different models affect CKF depth, structure, and semantic preservation
06 06-grand-synthesis Consolidates the research program and CKF's technology maturity
07 07-personal-knowops Applies CKF ideas to personal knowledge operations and Obsidian workflows
08 08-legal-compilation Explores legal-domain compilation for normative, article-level, compliance-oriented knowledge

Research Benchmarks

  • Composition Hallucination Bench
    A general RAG benchmark for testing whether models can compose implicit relations among retrieved fragments, such as exceptions, overrides, scope constraints, preconditions, temporal dependencies, contraindications, and procedural sequences.

  • Education Composition Hallucination Bench
    A pt-BR education benchmark for testing composition hallucination in pedagogical reasoning, assessment design, formative feedback, classroom decision-making, and brain-based learning strategies.

Suggested reading order

Start with 01 to understand the failure mode, then read 02 and 03 to understand why CKF is a compiler problem rather than a serialization problem. Papers 04 and 05 cover operational robustness and model sensitivity. Paper 06 is the synthesis. Papers 07 and 08 show applied and domain-specialized directions.

License

MIT — see LICENSE.

Contributing

Bug reports, PRs, and forks are welcome. The canonical CKF specification lives at compiledknowledgeformat.org/docs.

About

Open-source Compiled Knowledge Format (CKF) tools for schema-stable knowledge compilation, structured retrieval, and RAG infrastructure.

Topics

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors