Carry AI memory between tools with portability and neutrality.

A producer can emit an OMI-AI file and a consumer can read it without sharing a storage engine, query language, governance model, ontology, vector index, or vendor-specific runtime.

Read the docs Explore the schemas

.omi.json · application/vnd.omi+json

Vendor-neutral portability

A producer emits an OMI-AI file; any consumer reads it without sharing a storage engine, query language, governance model, ontology, vector index, or vendor runtime.

01

Legible from content alone

The floor is just id + content + created. A floor-only OMI-AI record remains legible from content alone, so memory survives tool changes.

02

Extensible via profiles

Integrity, retrieval, and sync stay out of core, carried in the ext object via named profiles, so the floor stays small and neutral.

03

A floor, not a framework.

OMI-AI is snapshot-first and deliberately minimal. The exclusions are a commitment, not an omission — a small floor that parties who disagree about everything above it can still share.

Snapshot-first, human-legible JSON
Two conformance levels — L0 Core and L1 Interchange
Two serializations — .omi.json and .omi.jsonl

basic-l1.omi.json

1{
2  "format": "open-memory-interchange",
3  "version": "0.1",
4  "serialization": "json",
5  "subject": {
6    "id": "user-123",
7    "type": "person",
8    "label": "Freddy"
9  },
10  "generated_at": "2026-06-06T09:00:00Z",
11  "generator": "example-tool/1.4",
12  "memories": [
13    {
14      "id": "01JZ0WFR4K2Q6N7S8T9V0ABCDF",
15      "content": "Freddy prefers direct critical pushback.",
16      "type": "semantic",
17      "created": "2026-05-01T09:02:11Z",
18      "confidence": 0.96,
19      "lang": "en",
20      "tags": ["communication", "preference"],
21      "source": {
22        "platform": "example-chat",
23        "ref": "session-882",
24        "method": "extracted"
25      },
26      "valid_from": "2026-05-01",
27      "valid_to": null
28    }
29  ]
30}

ECOSYSTEM

Tooling, coming soon.

The floor is small on purpose, so adapters, servers, CLIs, and SDKs can be built on top without changing the format.

Coming soon

Adapters

Import memory from existing assistant and agent exports into OMI-AI, and export back out — mapping common export fields by concept onto the OMI-AI floor.

ChatGPT
Claude
Mem0
LangChain

Coming soon

MCP servers

Read and write OMI-AI records over the Model Context Protocol, so agents can carry memory between tools without bespoke integrations.

Coming soon

CLIs

Validate, convert, and inspect .omi.json and .omi.jsonl from your terminal — L0/L1 validation and JSON to JSONL.

Coming soon

Language implementations

Typed libraries for reading and writing OMI-AI, so producers and consumers stay in sync across languages.

TypeScript
Python
Go
Rust

UPDATES

Follow the spec as it develops.

Get notified about releases, schema changes, and public review drafts.

GET INVOLVED

Building on OMI, or want to help shape it?

Tell us what you're working on and we'll be in touch.