OMIR
Open Memory Interoperability Resources

Implementations

Who speaks OMIR. This is the scoreboard — and the most important page on the site.

The honest metric. A format authored by exactly one vendor is just that vendor's export format with a logo. The number that matters for OMIR is not spec completeness — it is a second implementer who did not write the spec. That single independent integration is what turns a file format into a standard. If that's you, we want to hear from you.

Conformant implementations

An implementation is a producer (emits .omir), a consumer (reads it), or both. Conformance is graded by the reference validator at the Core level — structural validity, full reference integrity, and the R1 marker. See Conformance.

ImplementationRoleConformanceEncodingsNotes
Veld — Agentic Memory Producer Targeting Core .omir Reference implementation. Veld's internal Memory Interchange Format (MIF) exports to OMIR via an adapter — the way an EHR implements FHIR. Exercises the full R1 core: MemoryRecord (OMM-4), Entity, Relationship, Episode (OMM-3).
Your implementation here — — — Open a PR adding a row, or file it via Feedback.

Veld convenes the Working Group as the reference implementation, but does not own the standard — the spec and schemas are CC-BY-4.0 and the format is vendor-neutral by design.

How to get listed

  1. Implement against the schemas. Use the R1 JSON Schemas in schemas/ as your authoritative field list; build a Bundle exporter or importer. See Guides.
  2. Pass the reference validator at Core. Run omir-validate <your-bundle>.omir --level core until it exits 0. For consumers, also run it against the published R1 example Bundles.
  3. Publish a conformance statement. A short note listing the release you target (R1), your role(s), the conformance level, and the extension URLs you emit.
  4. Claim the badge. A clean Core run makes an artifact or producer eligible for the "Powered by OMIR" badge. The JSON conformance report is the evidence the Working Group accepts.

Why neutrality is the whole game

OMIR is positioned inside the agent-interoperability tent, not against it. MCP and A2A (now under the Linux Foundation) transport memory between agents at runtime; OMIR is the memory at rest — the document on disk, in the archive, in the export. A protocol that ships a memory blob still benefits from a neutral, versioned, schema-validated definition of what that blob contains. That's the role OMIR plays, and the reason a second implementer can adopt it without betting on any one vendor.