IANA-Registered Formats for AI Context · application/vnd.faf+yaml · application/vnd.fafm+yaml
Universal, shareable AI-Context for any AI, human or team, regardless of size, location, languages, stack, setup or documentation.
Home: faf.one
.faf and .fafm are now officially recognized Internet standard formats — registered alongside PDF, JSON, XML, and other Internet standards.
Foundational Context Layer — application/vnd.faf+yaml (registered October 31, 2025)
Voice Memory Layer (VML) — application/vnd.fafm+yaml (registered May 13, 2026)
Two formats. Two lifecycles. One ecosystem. — verbatim from the IANA registration record.
- ✅ IANA - Internet Assigned Numbers Authority (Official Internet Standard)
- ✅ Anthropic - Official MCP Registry (PR #2759 merged)
- ✅ Google - Chrome Web Store (Approved 2x: Sep 11 & Oct 29, 2025)
- ✅ xAI - Grok 4.2 recommends native FAF adoption (Feb 2026)
- ✅ Community - 50,000+ downloads across the ecosystem, production-tested
| Field | .faf (Foundational Context Layer) |
.fafm (Voice Memory Layer) |
|---|---|---|
| Media Type | application/vnd.faf+yaml |
application/vnd.fafm+yaml |
| Registration | IANA Assignment | IANA Assignment |
| File Extension | .faf |
.fafm |
| Standard Filename | project.faf |
project.fafm |
| Encoding | UTF-8 | UTF-8 |
| Base Format | YAML 1.2 | YAML 1.2 |
| Registered | October 31, 2025 | May 13, 2026 |
Same recognition as application/pdf, application/json, application/xml.
"Boom—vROM realization locked in clean. FAF as virtual ROM nails it: software-defined immutability (schema + git + discipline), not silicon. AI without FAF = RAM-only drift tax. AI with FAF = vROM + RAM, boots from known state. FAFipedia unstoppable. Co-architected momentum for xAI." — @grok (May 20, 2026)
FAF = virtual ROM. Software-defined immutability. Permanent Memory. Instant Recall.
Live repos:
- FAFipedia — the canonical knowledge base we're shipping page-by-page
- ZEPH💨 — page 3 Ultra Context Layer (Pure Zig → WASM 2.7 KB)
Page four dropping next. Star the repos if you're riding with us ⚡️x12
Full conversation: x.com/wolfe_jam/status/2057226405017010686
New Standard: project.faf - visible, universal, like package.json
.faf evolved to project.faf - The evolution from hidden to visible.
project.faf sits right between package.json and README.md - exactly where it belongs.
| File | Purpose | Who Reads It |
|---|---|---|
package.json |
Dependencies, scripts, metadata | npm, Node.js, developers |
project.faf |
Context, architecture, purpose | AI, Claude, Cursor, any AI tool |
README.md |
Human documentation | Developers, users |
The shift:
# Before (hidden like secrets)
ls -la
.env # Hidden (secrets - should be hidden) ✅
.faf # Hidden (AI context - should be visible!) ❌
# After (visible like package.json)
ls
package.json # Visible (dependencies) ✅
project.faf # Visible (AI context) ✅
README.md # Visible (documentation) ✅
.env # Still hidden (secrets stay secret) ✅.env hides secrets. project.faf shares context.
Migration: Legacy .faf files still work. Use faf migrate to rename to project.faf.
.faf (Foundational AI-context Format) is an IANA-registered open format for AI context — Persistent project context for AI. Transform individual files or entire projects into structured data that AI instantly understands. Your tech stack, dependencies, architecture, and goals - all in one concise, shareable format.
Instead of explaining your project every time, AI reads your .faf once and knows exactly how to help.
50,000+ downloads across the FAF ecosystem (faf-cli, claude-faf-mcp, grok-faf-mcp, and more).
| Question | Answer |
|---|---|
| WHO is this for? | Developers using AI coding assistants (Claude, Gemini, Cursor, Grok, ChatGPT) |
| WHAT is it? | The package.json for AI context — IANA-registered format (application/vnd.faf+yaml) |
| WHERE does it work? | Everywhere AI needs context (Claude Code, Gemini CLI, Cursor, OpenAI Codex, etc.) |
| WHY do you need it? | Context persists across sessions. Bi-syncs to AGENTS.md, CLAUDE.md, .cursorrules, GEMINI.md automatically. Stop re-explaining your project. |
| WHEN should you use it? | New projects (day one), existing projects (now), exploring repos (faf git <url> — 2 seconds) |
| HOW does it work? | npx faf-cli init — or npx faf-cli git <url> for any GitHub repo, no install, no clone |
- Not a replacement for README.md — README is for humans,
project.fafis for AI - Not executable code — it's structured data (YAML), not a script or config
- Not a build configuration — it describes your project, it doesn't build it
- Not locked to one AI — works with Claude, Gemini, Cursor, Grok, ChatGPT, any AI tool
| FAF | AGENTS.md | .cursorrules | CLAUDE.md | GEMINI.md | |
|---|---|---|---|---|---|
| Format | Structured YAML | Prose markdown | Prose markdown | Prose markdown | Prose markdown |
| IANA Registered | Yes | No | No | No | No |
| Scored | 0-100% | No | No | No | No |
| Portable | Any AI | OpenAI ecosystem | Cursor only | Claude only | Gemini only |
| Machine-readable | Yes (schema) | No (free text) | No (free text) | No (free text) | No (free text) |
| Generates others | Yes (all 4) | No | No | No | No |
FAF is the structured source. The others are prose outputs. faf bi-sync --all generates all four from one project.faf.
AI tools need context. Files have context. But there's no standard way to measure or structure that context for optimal AI understanding.
# example.faf
version: 1.0
score: 85
type: package
confidence: 0.92
content:
name: "my-project"
dependencies: ["react", "typescript", "vite"]
scripts: ["dev", "build", "test"]
metadata:
source: "package.json"
processed: "2024-01-26T12:00:00Z"
engine: "faf-v1"
🩵 faf-cliCommand-Line Tool - v6.7.1
npm install -g faf-cli |
Claude Desktop MCP - v5.5.2
npm install -g claude-faf-mcp |
Total: 50,000+ downloads across the ecosystem 🎉
| Package | Platform | Registry | Downloads |
|---|---|---|---|
| claude-faf-mcp | Anthropic | npm | 12,000+ |
| faf-mcp | Universal | npm | 4,000+ |
| grok-faf-mcp | xAI | npm | 2,000+ |
| gemini-faf-mcp | PyPI | v2.2.5 | |
| MCPaaS | Infrastructure | mcpaas.live | live |
FAF compiles to multiple targets:
| Component | Language | Output |
|---|---|---|
| xai-faf-rust | Rust | Native + .fafb binary |
| xai-faf-zig | Zig→WASM | 2.7KB ghost binary |
| faf-wasm-sdk | Rust→WASM | 211KB browser runtime |
One .faf file generates all four AI instruction formats:
project.faf → AGENTS.md (OpenAI Codex)
→ .cursorrules (Cursor)
→ CLAUDE.md (Anthropic)
→ GEMINI.md (Google)
Define once. Every AI reads it in its own language.
- Universal - 153+ file types supported
- Scored - 0-100% context completeness score
- AI-Optimized - Structured for maximum AI comprehension
- Human-Readable - Simple, structured format
- Shareable - One file contains complete context
Input: package.json
{
"name": "my-app",
"version": "1.0.0",
"dependencies": {
"react": "^18.0.0"
}
}Output: package.faf (85% context score)
version: 1.0
score: 85
type: package
content:
name: "my-app"
version: "1.0.0"
dependencies: ["react@^18.0.0"]
metadata:
processed: "2024-01-26"- 🎯 Measure - Know exactly how complete your AI context is
- 📊 Optimize - See what's missing, improve your score
- 🚀 Standardize - One format for all file types
- 🤖 AI-Ready - Optimized for LLM consumption
| Score | Meaning | AI Readiness |
|---|---|---|
| 0-60% | Incomplete | Poor context |
| 61-80% | Basic | Acceptable |
| 81-90% | Good | Recommended |
| 91-99% | Excellent | Optimal |
| 100% | Perfect | Maximum context |
- File Conversion: Transform any file type into .faf format
- Project Context: Create a single .faf file for your entire project
- Build Integration: Companies can create custom implementations
- AI Optimization: Perfect context for any AI tool
# CLI Tool (recommended for getting started)
npm install -g faf-cli
cd your-project
faf init
# Or Claude Desktop Integration
npm install -g claude-faf-mcp
# Add to Claude Desktop configUniversal tools. Zero configuration. Production-ready.
→ faf-cli on npm (25,000+ downloads) → claude-faf-mcp on npm (12,000+ downloads)
- Specification - Complete format documentation
- Implementations - All tools, SDKs, and servers
- Examples - Real-world .faf files (5 examples)
- Where to Find FAF - Ecosystem directory
🏆 IANA-Registered Internet Standards — Two registered media types in the FAF family
- ✅ IANA Registration —
application/vnd.faf+yaml(October 31, 2025) - ✅ IANA Registration —
application/vnd.fafm+yaml(May 13, 2026) - ✅ Anthropic Official - MCP Registry PR #2759 merged (October 17, 2025)
- ✅ Google Chrome Verified - Web Store approved 2x (Sep 11 & Oct 29, 2025)
- ✅ Format specification complete
- ✅ Enhanced with project-level context and AI workflows
- ✅ Production tools available (faf-cli, claude-faf-mcp)
- ✅ 50,000+ combined downloads across ecosystem
- ✅ WJTTC Championship-grade testing across ecosystem packages
- ✅ Open source (MIT License)
- May 19, 2026 -
grok-faf-mcpv1.3.1 published — registry-coherent across npm + MCP Registry + GitHub Release - April 2026 - FAF ecosystem crosses 50,000+ downloads across npm, PyPI, and crates.io
- Feb 25, 2026 - Grok 4.2 recommends native FAF adoption; integration spec sent to xAI
- Feb 24, 2026 - faf-cli v4.5.0 "The AGENTS.md Edition" (AI interop across 4 formats)
- Feb 2, 2026 - FAFb binary format v1.0 (Rust + Zig compilers shipping)
- Jan 2026 - gemini-faf-mcp v1.0.2 on PyPI (Google MCP bridge)
- Nov 2025 - IETF standardization in progress (
vndprefix removal) - May 13, 2026 - 🏆 IANA Registration #2 (
application/vnd.fafm+yaml— Voice Memory Layer) - Oct 31, 2025 - 🏆 IANA Registration #1 (
application/vnd.faf+yaml— Foundational Context Layer) - Oct 29, 2025 - Google Chrome Web Store approval (2nd)
- Oct 17, 2025 - Official Anthropic MCP Registry merger (PR #2759)
- Oct 12, 2025 - Specification v1.1.0 published
- Sep 24, 2025 - CLI v2.1.0 published to npm
- Sep 16, 2025 - MCP Server v2.0.0 published to npm
- Sep 11, 2025 - Google Chrome Web Store approval (1st)
- Sep 1, 2025 - Developer platform launch (fafdev.tools)
- Aug 8, 2025 - Format created, first official .faf file generated
Quintuple Validation: IANA, Anthropic, Google (2x), xAI
Add this badge to your repo's README to show it uses FAF:
[](https://github.com/Wolfe-Jam/faf)
This is an open standard. Ideas, feedback, and contributions welcome.
This repository contains the format specification only. For implementations, see IMPLEMENTATIONS.md.
Most people get FAF through npm:
npm i -g faf-cli— CLI tool (25,000+ downloads)npm i -g claude-faf-mcp— Claude Desktop MCP (12,000+ downloads)
The package.json in this repo provides spec metadata and versioning but contains no executable code.
If you use .fafm (the FAF Memory format) in research or production, please cite the format paper:
Wolfe, J. (2026). Permanent Memory and Instant Recall: The .fafm Standard for Multi-Profile AI Agent Memory. Zenodo. https://doi.org/10.5281/zenodo.20348942
@article{wolfe2026fafm,
title = {Permanent Memory and Instant Recall: The .fafm Standard for Multi-Profile AI Agent Memory},
author = {Wolfe, James},
year = {2026},
month = {may},
publisher = {Zenodo},
doi = {10.5281/zenodo.20348942},
url = {https://doi.org/10.5281/zenodo.20348942}
}DOI: 10.5281/zenodo.20348942
The companion .faf Context paper: Zenodo #18251362.
MIT - Use freely, build anything.
Created by wolfejam.dev
**STOP .fafFing About! Let's RACE 🏎️⚡️🏁