aiqbench analysis
memorywire addresses the fragmentation in agent-memory ecosystems by standardizing wire formats and governance interfaces across five major frameworks, enabling interoperable memory operations without requiring bespoke integrations. The reference implementation demonstrates high recall performance and resilience against adversarial injection attacks, validating the protocol's utility for building robust, composable agent architectures. This initiative shifts the focus from competing algorithms to establishing a neutral infrastructure layer that facilitates seamless migration and human-in-the-loop oversight.
memorywire: A Vendor-Neutral Wire Format for Agent Memory Operations
arXiv:2606.01138v2 Announce Type: replace-cross Abstract: Agent-memory frameworks -- mem0, Letta/MemGPT, Cognee, Zep/Graphiti, MemoryOS, MemTensor -- each ship their own SDK, storage layout, and operational vocabulary. There is no shared wire format: every integration is bespoke, every migration rebuilds memory from scratch, and no framework ships a governance surface that lets a human review writes before they enter long-term storage. We present memorywire, a JSON-Schema 2020-12 wire format for five memory operations (remember, recall, forget, merge, expire) over four memory types (semantic, episodic, procedural, emotional), with a MemoryStore interface, a fan-out router, and an optional HITL governance channel. We describe an open-source reference implementation with five backend adapters (sqlite-vec, mem0, Letta, Cognee, pgvector); a microbenchmark on a 100-fact / 50-query labelled corpus (42 with non-empty gold ids + 8 no-match probes) achieving recall@5 = 1.000 on the 42 gold-id queries with ingest p50 = 37.8 ms and recall p50 = 40.6 ms; an adversarial-fusion experiment showing Reciprocal Rank Fusion holds recall@5 = 1.000 across a 1-of-N rank-0 injection sweep (K in {0, 5, ..., 50}) where max fusion collapses to 0.500 with 80% leak at K >= 5; and a 16-scenario cross-adapter conformance suite passing 68 of 80 cells with zero failures. The contribution is not a new algorithm; it is a packaging of established components (RRF, FSMs, STM/LTM consolidation, diff-and-approve workflows) into a venue-neutral protocol with an empirically validated reference, positioned to compose with the Model Context Protocol rather than compete with it.
Abstract:Agent-memory frameworks -- mem0, Letta/MemGPT, Cognee, Zep/Graphiti, MemoryOS, MemTensor -- each ship their own SDK, storage layout, and operational vocabulary. There is no shared wire format: every integration is bespoke, every migration rebuilds memory from scratch, and no framework ships a governance surface that lets a human review writes before they enter long-term storage. We present memorywire, a JSON-Schema 2020-12 wire format for five memory operations (remember, recall, forget, merge, expire) over four memory types (semantic, episodic, procedural, emotional), with a MemoryStore interface, a fan-out router, and an optional HITL governance channel. We describe an open-source reference implementation with five backend adapters (sqlite-vec, mem0, Letta, Cognee, pgvector); a microbenchmark on a 100-fact / 50-query labelled corpus (42 with non-empty gold ids + 8 no-match probes) achieving recall@5 = 1.000 on the 42 gold-id queries with ingest p50 = 37.8 ms and recall p50 = 40.6 ms; an adversarial-fusion experiment showing Reciprocal Rank Fusion holds recall@5 = 1.000 across a 1-of-N rank-0 injection sweep (K in {0, 5, ..., 50}) where max fusion collapses to 0.500 with 80% leak at K >= 5; and a 16-scenario cross-adapter conformance suite passing 68 of 80 cells with zero failures. The contribution is not a new algorithm; it is a packaging of established components (RRF, FSMs, STM/LTM consolidation, diff-and-approve workflows) into a venue-neutral protocol with an empirically validated reference, positioned to compose with the Model Context Protocol rather than compete with it.
v2: title corrected from pre-launch name "AMP" to "memorywire"; abstract clarifies recall@5 = 1.000 is on the 42 gold-id queries (50 total; 8 no-match probes excluded). 17 pages, 1 figure, 6 tables. Code: this http URL. Companion to arXiv:2604.18248 (Prompt Injection Detection)
Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC)
arXiv:2606.01138 [cs.CR]
(or arXiv:2606.01138v2 [cs.CR] for this version)
Full text under the publisher's open licence · 301 words
Related coverage
memorywire: A Vendor-Neutral Wire Format for Agent Memory Operations
arXiv · Jul 20, 2026
Presentation: Architecting AI Systems for the Messy Reality of Enterprises: Why Agentic Compute is the Missing Layer
InfoQ · Aug 3, 2026
Spur: Modern GPU Job Scheduling for HPC and AI Workloads
AMD · Jul 22, 2026
Article: The Agent Harness: What it is and Two Ways to Build One
InfoQ · Sep 25, 2026