atomicstrata/atomicmemory-sdk

AtomicMemory SDK is a TypeScript library that gives AI agents a memory layer, the ability to remember things about users across conversations and sessions. The core use case is building AI applications where the assistant needs to retain information beyond the current session, such as user preferences, past interactions, or uploaded documents. The SDK provides two main namespaces: memory, for storing, ingesting, and searching through past interactions, and storage, for managing file artifacts like PDFs or other content. Memory search works using semantic search, meaning you can query for things conceptually related to a topic rather than matching exact words. Embeddings, numeric representations of text that capture meaning, can be generated locally within the SDK without sending data to an external service. The design is provider agnostic: the SDK ships built-in adapters for AtomicMemory's own self-hosted backend and for Mem0, another open-source memory system, and you can implement the MemoryProvider interface to plug in any backend you want. A quick-start example in the README shows loading the client, ingesting a user message, and then searching for a related preference by query, where the memory layer finds the relevant stored fact. Storage operations work similarly: you can save a pointer to an external file and retrieve its artifact ID later. The SDK is written in TypeScript, installable via pnpm, npm, or yarn, and is intended to run server-side only. The README notes that browser code must not instantiate the full client directly. It is open-source under the Apache 2.0 license.