CKF
Compiler Pro · v1.0

CKF Compiler Pro

Industrial pipeline — chunk · extract · reduce · validate. BYOK with OpenAI, Anthropic, Google, DeepSeek or OpenRouter. Handles up to ~960k characters per run. · reference compiler v1.3.1

Try the public demo →
Open to anyone — bring your own API key. No login required for BYOK (OpenAI, Anthropic, Google, DeepSeek, OpenRouter). Advanced AI (project credits, no key) is reserved for admins and the allowlist. Just want a quick taste? Try the public demo — heuristic compile, instant, 60k characters, no key needed.

Engine

OpenAI API key. Get a key. Stored only in this browser tab (sessionStorage).

Curious how the LLM extraction differs from the heuristic one? See side-by-side →

Auto upgrades to complete for record-oriented sources (JSONL, FAQ, normative texts).

Sources

0 chars

Pipeline

  1. 1
    Extract
    Read all sources, normalize whitespace and headings.
  2. 2
    Chunk
    Split semantically by markdown sections (~12k chars each).
  3. 3
    Map · Extract
    Per-chunk extraction via tool-calling — strict JSON schema.
  4. 4
    Reduce · Merge
    Deduplicate concepts, merge facts/relations, score utility.
  5. 5
    Validate
    Check required fields, types, and counts.
  6. 6
    Serialize
    Emit canonical Markdown + JSON CKF package.

Output

Add sources, choose an engine, then compile.
How it works

From human file to AI agent, in 4 steps

CKF translates human-oriented documents into structured knowledge that agents can reason over — without hallucinating and with full traceability.

01
Human file

PDF, DOCX, MD, TXT, transcripts — anything humans read.

02
CKF Compiler

Extracts entities, concepts, rules, principles and procedures into 22 typed sections.

03
.ckf Package

Markdown / JSON / YAML — portable, versionable, auditable, traceable to source.

04
AI Agents

LLMs, RAG and agents consume pre-compiled knowledge — fewer tokens, less hallucination, higher precision.

See .ckf anatomyPDF vs CKFOpen Viewer