About OMIR
OMIR — Open Memory Interoperability Resources (pronounced "OH-meer") — is an open, vendor-neutral, at-rest data format for portable AI agent and cognitive memory. It is a document standard: not a wire protocol, and not a product.
Modeled on FHIR
OMIR is shaped deliberately on HL7 FHIR, carrying that ecosystem's hard-won lessons about resources, references, profiles, and honest maturity grading into the agent-memory domain. FHIR didn't win by building a better hospital — it won by defining the at-rest resources every system reads and writes. OMIR makes the same bet for memory.
| FHIR | OMIR |
|---|---|
| Fast | Open |
| Healthcare | Memory |
| Interoperability Resources | Interoperability Resources |
"Open" leads the acronym deliberately — it's the one positioning a commercial product can't copy without abandoning its moat.
Where OMIR sits: MCP, A2A, and OMIR
Two transport standards already exist for agent systems, both now stewarded by the Linux Foundation: MCP (Model Context Protocol) connects agents to tools and context; A2A (Agent-to-Agent) lets agents talk to each other. Both move data between live endpoints. Neither defines what memory is once it has been written down.
They are complementary, never competing. An agent can serve memory over MCP, hand it to a peer
over A2A, and persist or export it as an .omir Bundle. The transport moves the bytes;
OMIR defines the bytes — the same clean separation FHIR drew between its RESTful API and its
Resources.
How it works
Everything in OMIR is a typed Resource; resources link by typed reference of
the form ResourceType/id. The R1 core is small on purpose —
MemoryRecord,
Entity,
Relationship,
Episode, and a
Bundle container that is the
.omir document.
- 80/20 core + extensions. The core covers the 80% every memory system shares:
content, timestamps, tiers, calibrated Bayesian confidence, multi-time-scale decay and anchoring,
Hebbian relationship strength, temporal invalidation, provenance, and prospective ("intention")
memory. The proprietary 20% rides in a typed
extension[]escape hatch that any reader may safely ignore. - Honest maturity (OMM). Each resource type is graded on the OMIR Maturity
Model, an integer 0–5 surfaced in
meta.maturity. We do not overclaim: MemoryRecord OMM-4 Entity OMM-3 Relationship OMM-3 Episode OMM-3. - Two lossless encodings.
.omiris canonical JSON / JSON-LD;.omirbis a compact binary profile (CBOR) for edge and robotics. The two are lossless round-trips of one logical model.
Governance & licensing
OMIR is stewarded by a vendor-neutral OMIR Working Group in a neutral GitHub organization. Veld convenes the group but does not own the standard. The group operates an open RFC + ballot process, a Technical Steering Committee, semantic release versioning (R1, R2, …), and a published deprecation policy — see GOVERNANCE.md.
Licensing is deliberately decoupled from any implementation:
- The specification and schemas are licensed CC-BY-4.0.
- The reference code (validator, generators) is licensed Apache-2.0.
A standard under a restrictive license is dead on arrival. The OMIR spec is, and will remain, free.
The honest risk
Standards die without a second implementer. A format with one producer and zero independent consumers is documentation for a private file, not a standard. Securing an external adopter is more important than any single feature — which is why Implementations is the page we watch most closely. Status: OMIR R1 (draft).