Skip to content
MIF

Introduction

Memory Interchange Format

Section titled “Memory Interchange Format”

An open standard for portable AI memory representation.

MIF defines a vendor-neutral data model for AI memories with dual representations: human-readable Markdown files and machine-processable JSON-LD documents. One format, every provider.

Why MIF?

Section titled “Why MIF?”

AI memory is fragmented. Mem0, Zep, Letta, LangMem, and others each use proprietary schemas with no interoperability. Switching providers means losing memories or writing custom migration code.

MIF fixes this.

For memory providers: Adopt a common interchange format. Users gain confidence knowing their data is portable, which lowers switching costs and raises adoption.

For developers: Build on a stable schema with semantic web compatibility. MIF memories are valid JSON-LD documents that work with existing linked data tooling.

For users: Own your memories. MIF files are valid Obsidian notes that work in any Markdown editor. No cloud dependency required.

Quick Start

Section titled “Quick Start”

A minimal MIF memory in Markdown:

---
id: "urn:mif:memory:a1b2c3"
type: "semantic"
title: "User prefers dark mode"
confidence: 0.92
created: "2026-01-15T10:30:00Z"
---


The user consistently selects dark mode across all applications.
They mentioned it reduces eye strain during late-night coding sessions.


## Relationships


- relates_to: [[urn:mif:memory:ui-prefs]] "UI Preferences"

The same memory as JSON-LD:

{
  "@context": "https://mif-spec.dev/schema/context.jsonld",
  "@type": "mif:Memory",
  "mif:id": "urn:mif:memory:a1b2c3",
  "mif:memoryType": "semantic",
  "mif:title": "User prefers dark mode",
  "mif:confidence": 0.92,
  "mif:created": "2026-01-15T10:30:00Z",
  "mif:content": "The user consistently selects dark mode across all applications. They mentioned it reduces eye strain during late-night coding sessions.",
  "mif:relationships": [
    {
      "@type": "mif:Relationship",
      "mif:type": "relates_to",
      "mif:target": "urn:mif:memory:ui-prefs",
      "mif:label": "UI Preferences"
    }
  ]
}

Key Features

Section titled “Key Features”
FeatureDescription
Dual FormatHuman-readable .memory.md and machine-processable .memory.json
Semantic WebJSON-LD with full linked data compatibility
Obsidian NativeMarkdown files work as valid Obsidian vault notes
Bi-TemporalTrack both event time and recording time
Memory DecayBuilt-in confidence decay models (exponential, step, linear)
Entity SystemTyped entities: Person, Organization, Technology, Concept, File
9 Relationship Typesrelates_to, derived_from, supersedes, contradicts, and more
ProvenanceW3C PROV-based origin tracking
NamespacesHierarchical organization with reserved cognitive prefixes
3 Conformance LevelsProgressive adoption from basic to full compliance

Current Status

Section titled “Current Status”

Version: 0.1.0-draft

This is a draft specification. No providers currently implement MIF. The goal is to establish a vendor-neutral interchange format that providers may choose to adopt. Feedback and contributions are welcome.