Format Adapters

MIF tools include adapters for converting between popular AI memory formats and MIF v2.

---

Shodh / MIF Native (shodh)

The native MIF v2 format. Also handles backward-compatible import of MIF v1 documents.

Detection: JSON object with "mif_version" or "shodh-memory" key.

{
  "mif_version": "2.0",
  "memories": [
    { "id": "...", "content": "...", "created_at": "..." }
  ]
}

---

mem0 (mem0)

Converts mem0 JSON array exports.

Detection: JSON array where items have a "memory" field.

[
  {
    "id": "...",
    "memory": "User prefers TypeScript",
    "user_id": "user-42",
    "created_at": "2025-06-15T12:00:00Z",
    "metadata": { "category": "preference" }
  }
]

Field mapping:

mem0	MIF
memory	content
id	external_id + id (preserved if valid UUID)
metadata.category	memory_type (via mapping table)
metadata.tags	tags (comma-separated string or array)
user_id	export_meta.user_id

---

CrewAI (crewai)

Converts CrewAI long-term memory exports from LTMSQLiteStorage.

Detection: JSON array where items have a "task_description" field.

[
  {
    "task_description": "User prefers dark mode for all IDEs",
    "metadata": "{\"category\": \"preference\"}",
    "datetime": "1718452800.0",
    "score": 0.95
  }
]

Field mapping:

CrewAI	MIF
task_description	content
metadata (JSON string)	metadata (parsed to object)
datetime (Unix timestamp)	created_at (ISO 8601)
score	metadata.score

---

LangChain / LangMem (langchain)

Converts LangChain and LangMem Item objects.

Detection: JSON array where items have "namespace" and "value" fields.

[
  {
    "namespace": ["memories", "user-prefs"],
    "key": "pref-dark-mode",
    "value": { "kind": "Memory", "content": "User prefers dark mode" },
    "created_at": "2025-06-15T12:00:00Z",
    "score": 0.9
  }
]

Field mapping:

LangChain	MIF
value.content	content
value.kind	memory_type (lowercased, mapped)
namespace	tags (flattened)
key	external_id
created_at	created_at
updated_at	updated_at
score	metadata.score

---

Generic JSON (generic)

Fallback adapter for any JSON array with a content field.

Detection: JSON array where items have a "content" field.

[
  {
    "content": "Remember to check logs daily",
    "type": "task",
    "timestamp": "2025-06-15T12:00:00Z",
    "tags": ["ops"]
  }
]

Accepts timestamp, created_at, or date for the creation time. Accepts type or memory_type for the memory type.

---

Markdown (markdown)

YAML frontmatter + body format. Each memory is a --- delimited block.

Detection: Input starts with ---.

---
type: observation
created_at: 2025-06-15T12:00:00Z
tags: [python, testing]
---
User prefers pytest over unittest.

---

Writing Custom Adapters

Python

from mif.adapters import MifAdapter
from mif.models import MifDocument

class MyAdapter(MifAdapter):
    def name(self) -> str:
        return "My Format"

    def format_id(self) -> str:
        return "myformat"

    def detect(self, data: str) -> bool:
        return '"my_marker"' in data

    def to_mif(self, data: str) -> MifDocument:
        # Parse data and return MifDocument
        ...

    def from_mif(self, doc: MifDocument) -> str:
        # Serialize MifDocument to your format
        ...

TypeScript

import { MifAdapter, MifDocument } from "@varunshodh/mif-tools";

class MyAdapter implements MifAdapter {
  name() { return "My Format"; }
  formatId() { return "myformat"; }
  detect(data: string) { return data.includes('"my_marker"'); }
  toMif(data: string): MifDocument { /* ... */ }
  fromMif(doc: MifDocument): string { /* ... */ }
}

Register your adapter:

from mif.registry import AdapterRegistry
registry = AdapterRegistry()
registry.adapters.insert(0, MyAdapter())  # highest priority

import { AdapterRegistry } from "@varunshodh/mif-tools";
const registry = new AdapterRegistry();
registry.register(new MyAdapter()); // prepended for highest priority