Rememori

Rememori is an embedded memory engine written entirely in TypeScript that gives AI agents and applications the ability to remember information across sessions. The core library has zero dependencies, requires no external servers, and avoids native bindings altogether. It compresses down to roughly 8 KB and runs anywhere JavaScript executes, whether that's Node, Bun, a browser tab, an edge worker, or an Electron app. For developers building AI-powered tools that need persistent context without the overhead of standing up dedicated infrastructure, Rememori offers a straightforward way to add semantic memory to their stack. The appeal lies in treating memory as a library concern rather than an infrastructure concern, which dramatically simplifies the deployment story for applications that need to remember things.

The system operates around three fundamental actions that map naturally to how memory works. The remember function stores text along with embeddings, tags, and automatically extracted entities, creating rich representations that support multiple retrieval strategies. The recall function retrieves memories by ranking them according to similarity, shared entities, user feedback, importance scores, and time decay, combining these signals to surface the most relevant context for any given query. The forget function removes memories when they're no longer needed. Recent updates introduced reinforce and demote functions that let developers validate memory accuracy based on how well stored information actually performs during real usage. When a memory proves helpful, reinforcing it increases its future retrieval weight. When stored information turns out to be misleading, demoting it reduces its influence. This feedback loop means the memory layer can improve over time rather than accumulating stale context.

Storage adapts to the runtime environment automatically without requiring configuration. In browsers, Rememori uses IndexedDB to persist memories locally, giving web applications genuine persistent memory without server round trips. On servers, it writes to the filesystem. The embedder component is pluggable, so developers can connect Ollama for local embeddings, any OpenAI-compatible endpoint for cloud-based embeddings, or a custom embedding function of their own design. This flexibility means you're not locked into any particular embedding provider. An automatic knowledge graph handles entity extraction and linking, which helps the recall process surface related memories even when the query text doesn't match verbatim. If you ask about a person, the system can retrieve memories about projects they worked on or organizations they belong to.

The target audience includes developers building customer support bots that need to retain conversation history across sessions. Coding agents benefit from cross-session memory of project context, recalling architecture decisions and code patterns across different conversations. Local-first applications requiring semantic search without cloud dependencies can use Rememori to build privacy-preserving search features that work entirely offline. Compliance-sensitive systems in legal, healthcare, or finance sectors can keep memory entirely on-premise, satisfying data residency requirements. Game developers can give NPCs persistent behavior that evolves based on player interactions. Home automation builders can create assistants that learn household patterns over time. The common thread is any application where an AI component benefits from remembering what happened before.

What sets Rememori apart from larger memory solutions is the zero-infrastructure requirement combined with genuine portability across JavaScript runtimes. There's no database to provision, no vector store to configure, and no server to maintain. The same code runs in a serverless function, a desktop app, or a browser extension without modification. The entity-aware recall goes beyond simple vector similarity by considering relationships between concepts, which improves relevance when queries are indirect or contextual. The feedback mechanisms for reinforcing and demoting memories let the system learn which stored information actually proves useful in practice. The library also works as an MCP server, making it accessible to Claude and other AI tools that support the Model Context Protocol.

Rememori is open source under the MIT license, so there's no pricing to consider and no vendor relationship to manage. Developers can install it from npm and start adding memory capabilities immediately. The project maintains documentation and examples covering common integration patterns. For teams already working with TypeScript codebases who want to give their AI features persistent memory without adopting heavyweight infrastructure or depending on external services, Rememori provides a focused solution that stays out of the way while handling the underlying complexity of embeddings, storage, and retrieval.