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.
- Source ZIP: https://github.com/tomazinho/open-ckf-compiled-knowledge-format/archive/refs/tags/v1.0.0.zip
- Releases page: https://github.com/tomazinho/open-ckf-compiled-knowledge-format/releases
| 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 |
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-compilerTo build a static bundle:
npm run build
# -> dist/ ready for any static host or `python -m http.server`| 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) |
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.
| 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 |
-
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.
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.
MIT — see LICENSE.
Bug reports, PRs, and forks are welcome. The canonical CKF specification lives at compiledknowledgeformat.org/docs.