Every era creates its file format. AI doesn't have one yet.

#ai #python #showdev #opensource

Print gave us PDF. Photography gave us JPEG. Music gave us MP3.

The AI era generates more content than any previous era — but has no format to carry
trust, provenance, or verification with that content.

Today I'm releasing AKF — Agent Knowledge Format. Open source. Open spec.

What is it?

AKF is EXIF for AI. ~15 tokens of JSON that embed directly into any file:

Trust scores — how reliable is this content? (0-1)
Source provenance — where did it come from? (SEC filing → analyst → AI summary)
Security classification — who can see it? (public, internal, confidential)
AI attribution — which model generated it?
Compliance metadata — does it meet EU AI Act, SOX, NIST requirements?

Install

pip install akf          # Python
npm install akf-format   # TypeScript

30-Second Demo

# Stamp a file
akf stamp report.md --agent claude --evidence "generated from quarterly data"

# See what's inside
akf inspect report.md

# Check compliance
akf audit report.md --regulation eu_ai_act

# Embed into Word
akf embed report.docx

Zero Touch

eval "$(akf shell-hook)"
# Now every file Claude, ChatGPT, OpenClaw, Aider, Ollama creates is auto-stamped

Why Now

The EU AI Act Article 50 takes effect August 2, 2026. AI-generated content
published for public interest must carry transparency metadata. Penalties: up to
€35M or 7% of global turnover.

Colorado's SB 205 follows on June 30, 2026. DORA is already enforced.

AKF maps directly to these requirements. One command gives you a compliance report.

Integrations

Ships with integrations for LangChain, LlamaIndex, CrewAI, MCP (Model Context
Protocol), VS Code, and GitHub Actions.

DEV Community