# Agent Memory Needs Its Own OpenTelemetry Moment Fragmented AI agent memory formats require a unified standard for interoperability. - **Source:** [Hacker News (news.ycombinator.com)](https://brifing.si/source/13c923cd-6d99-442b-9b8b-1181f4d8cd91) - **Published:** 2026-11-08 - **Topic:** [technology](https://brifing.si/c/technology) - **Original article:** - **On Brifing:** https://brifing.si/s/d18bc3c9-9777-4f88-a6e2-7590d2d9b940 ## The Fragmentation Crisis Modern coding agents store memory in incompatible formats, preventing tools from sharing learned lessons. Whether it is an editor agent or a CI bot, each tool maintains its own isolated schema. This creates a redundant cycle where agents repeatedly rebuild the same small ideas without being able to read each other's stored data. ## Lessons from Observability Before 2019, the observability industry suffered from similar fragmentation, with vendors forcing users into proprietary instrumentation. The merger of OpenTracing and OpenCensus into OpenTelemetry resolved this by standardizing data formats while allowing vendors to compete on analysis. Agent memory development currently mirrors this pre-2019 state, requiring a similar collective industry agreement. ## Existing Memory Proposals Several projects are attempting to solve memory portability, including MacPaw Research’s Portable Memory and the Open Memory Protocol. These initiatives focus primarily on export functionality, allowing users to move data between platforms. However, they lack a unified standard for live addressing, which is essential for multiple agents working within one repository. ## Core Data Requirements A successful memory specification must answer four fundamental questions: where a memory is true, its unique identity, its content, and its expiration. While tools like LangGraph and Graphiti offer partial solutions, a comprehensive standard must also record the author of the memory to ensure transparency and trust during agentic interactions. ## Proposed Record Structure A potential record structure uses frontmatter to define the envelope, including address, identity, and lifecycle, followed by Markdown content. This approach allows tools to store metadata in attributes while keeping the core memory readable. Such a format remains flexible enough to be serialized as JSON for wire transmission between different agentic systems. ## Addressing Security Risks Shared memory introduces potential prompt-injection channels, as agents may be steered by malicious data. A robust specification must mitigate this by recording the author of each memory and supporting digital signatures. By allowing records to point at specific code dependencies, readers can verify if the underlying code has changed since recording. ## The Path to Agreement The industry should avoid creating yet another competing standard, which often leads to further fragmentation. Instead, the focus must be on merging existing proposals from tools like LoreKit, Mem0, and Letta. By taking the best features from each, the community can build a shared specification that benefits the entire developer ecosystem. ## Future of Agentic Interoperability Standardizing agent memory is essential for creating a cohesive ecosystem where tools can collaborate effectively. By moving beyond vendor-specific silos, developers can ensure that agents learn from each other seamlessly. The goal is to establish a neutral, shared specification that enables true interoperability, mirroring the success seen in the observability sector. ## Sources covering this story - https://www.lorekit.io/blog/open-agent-memory-format — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._