Your AI Agent Memory Is Now Portable: Introducing MIF in ZeroMemory

# Your AI Agent Memory Is Now Portable Agent memory has a lock-in problem. Once your agents' context lives inside a proprietary memory service, moving it out is painful — or impossible. You're betting your product's most valuable asset, its accumulated knowledge, on one vendor never raising prices, never shutting down, and never getting in your way. Today we're fixing that. **ZeroMemory now exports and imports agent memory in open, portable formats** — the **Memory Interchange Format (MIF)** and **Vault-LD**. Your memory is yours: import it from Mem0, Zep, or Letta; export it to git or Obsidian; round-trip it losslessly. No lock-in, ever. ## What is MIF? **MIF (Memory Interchange Format)** is a portable, standards-aligned format for AI agent memory: markdown plus linked-data (JSON-LD) frontmatter, projected into a knowledge-graph ontology. Think of it as the CSV or PDF of AI memory — a neutral format your memory can move in and out of. Alongside it, **Vault-LD** is a git- and Obsidian-friendly dialect: plain markdown notes with YAML frontmatter that you can version-control and browse as files. Here's the strategic part: MIF and Vault-LD are emerging standards that Mem0, Zep, and Letta don't implement yet. ZeroMemory already has the knowledge-graph engine those file formats describe — so it's a **drop-in destination** for memory leaving other tools, and a durable, inspectable home for the memory you keep. ## Three things you can do today ### 1. Migrate off Mem0, Zep, or Letta Map each provider's records to MIF documents and import them with a single API call. Imported memories flow through ZeroMemory's standard write path, so they get embeddings, entity extraction, and semantic recall automatically. ```python import requests def mem0_to_mif(rec): return { "@context": "https://mif.ainative.studio/context/v1", "@type": "Memory", "@id": f"urn:mif:memory:{rec['id']}", "content": rec["memory"], "conceptType": "semantic", } docs = [mem0_to_mif(r) for r in your_mem0_records] requests.post( "https://api.ainative.studio/api/v1/public/memory/v2/import", headers={"Authorization": f"Bearer {TOKEN}"}, json={"format": "mif-json", "namespace": "project:acme", "documents": docs}, ) ``` ### 2. Export a memory (or a whole namespace) Export a single memory as MIF JSON-LD, MIF Markdown, or Vault-LD: ```bash curl -H "Authorization: Bearer $TOKEN" \ "https://api.ainative.studio/api/v1/public/memory/v2/$MEMORY_ID/export?format=mif-json&level=3" ``` Or stream an entire namespace as NDJSON — namespace-isolated, so nothing leaks across projects: ```bash curl -H "Authorization: Bearer $TOKEN" \ "https://api.ainative.studio/api/v1/public/memory/v2/export?namespace=project:acme&format=vault-ld" ``` ### 3. Keep your memory in git or Obsidian Because Vault-LD is just markdown with frontmatter, you can commit your agent memory to a repo, diff it, review it in a PR, or open it in Obsidian and see the whole knowledge graph as linked notes. ## Conformance levels: from content to full graph MIF export supports three levels of fidelity: - **Levels 1–2** — the essentials: content, title, concept type, namespace, tags, and properties. - **Level 3** — the full picture: **relationships** between memories, **bi-temporal validity** windows (when a fact was true), **provenance** (where it came from and how much to trust it), portable **embeddings**, and consolidated **summaries**. That Level-3 graph is exactly what plain files from other tools can't capture — and it round-trips losslessly. ## Why this matters Your agents get smarter the more they remember. That accumulated memory is a moat — but only if you own it. Portability turns memory from a liability (vendor lock-in) into an asset you control: - **Switch without fear.** Try ZeroMemory with your real memory, exported back out any time. - **Own your data.** Git-backable, inspectable, auditable — not trapped in a black box. - **No re-training tax.** Migrate the memory your agents already have instead of starting from zero. ## Get started The interchange endpoints are live on the ZeroMemory API today: - `GET /api/v1/public/memory/v2/{id}/export` — export one memory - `GET /api/v1/public/memory/v2/export?namespace=...` — bulk-export a namespace - `POST /api/v1/public/memory/v2/import` — import MIF/Vault-LD documents See the [Memory Interchange page](https://ainative.studio/memory-interchange) for migration guides from Mem0, Zep, and Letta — and start free, no credit card required. Your memory. Your format. No lock-in.

Your AI Agent Memory Is Now Portable: Introducing MIF in ZeroMemory

Check your site's AX Score