Memory Grain is an open specification designed to provide autonomous systems and AI agents with a standardized approach to memory storage, retrieval, and context assembly. It addresses the challenge of enabling agents to maintain immutable, portable, and verifiable memory, supporting collaboration and reasoning across diverse systems and industries. The specification is structured as a family of three interoperable standards: the Open Memory Specification (OMS), the Context Assembly Language (CAL), and the Semantic Markup Language (SML).
" Each grain is cryptographically signed, uses SHA-256 hashing for content addressing, and supports COSE-based signatures for tamper evidence. These containers are designed to be portable and self-contained, allowing storage in a variety of environments such as S3, Kafka, file systems, or version control systems, without vendor lock-in. Grains are described as audit-ready, supporting features such as DID-scoped consent, jurisdiction-aware retention, and compliance with regulatory frameworks including GDPR, HIPAA, and SOX.
CAL serves as a declarative query language that enables non-destructive, deterministic assembly of context from grain stores. This allows AI agents to assemble relevant memory blocks for tasks such as LLM context consumption, supporting token-budgeted context assembly and multiple output formats, including SML, TOON, Markdown, and JSON. SML, the Semantic Markup Language, is a flat, tag-based output format optimized for LLM context windows and human review, using grain type tags as epistemic signals to inform LLMs about the nature of the information.
The specification outlines three levels of conformance, ranging from minimal reader implementations to full production stores with features like AES-256-GCM encryption, per-user key derivation, encrypted search, hash-chained audit trails, and policy engines for compliance. 0, allowing developers to implement interoperable agent memory solutions in any programming language. This makes it suitable for AI system and agent developers seeking standardized, portable, and secure memory management for autonomous systems.
Memory Grain is an AI & ML product. It focuses on providing a standardized, portable, and verifiable memory format for AI agents to store and share knowledge. It is built as an open-source project for ai developers. Memory Grain is open source under the Open Source license. It runs on the web, and it can be self-hosted.
It is developed by Open Web Foundation (OWF), and the product first shipped in 2026. Development happens publicly on GitHub with 8 commits in the last 90 days. PulseGate's similarity index finds few close equivalents — Memory Grain occupies a relatively distinct niche. Key capabilities include open specification, immutable memory, and content-addressed storage.
Latest indexed changes and source events
memorygrain.org discovered by the PulseGate indexer
Other apps tracked under the same category.
