Built for production, powered by Rust
Zero-cost abstractions, fearless concurrency, and guaranteed memory safety. Cortex Memory leverages Rust's type system to deliver production-grade performance without sacrificing developer experience.
Enterprise-grade multi-tenancy with strict data isolation. Deploy once, serve thousands of users with tenant-scoped collections and filesystem partitioning.
Integrate via REST API, CLI, MCP protocol, or native Rust library. From chatbots to enterprise systems—Cortex Memory adapts to your stack.
Principles for building scalable, long-running AI agents
Everything is stored as files. Resources, users, sessions, and memories
are organized in a hierarchical filesystem structure using the
cortex:// URI scheme—making context portable, versionable,
and compatible with Git.
cortex:// File System │ ├─ 📁 resources/ │ ├─ 📄 api_docs.md │ └─ 📄 best_practices.md │ ├─ 📁 user/ │ └─ 📁 user_001/ │ └─ 📄 profile.json │ ├─ 📁 agent/ │ └─ 📁 assistant_v2/ │ └─ 📄 profile.json │ ├─ 📁 session/ │ └─ 📁 sess_20260219/ │ ├─ 📄 timeline.md │ ├─ 📄 L1_overview.md │ └─ 📄 L0_abstract.md │ └─ 📁 memories/ ├─ 📄 preferences.json └─ 📄 decisions.json
Progressive context disclosure through a three-tier hierarchy. Load only what's needed—from high-level abstracts (L0) to detailed transcripts (L2)—optimizing LLM token usage while maintaining rich context.
Intelligent context discovery using hierarchical retrieval. Start with directory-level positioning, perform local vector search, and rerank results—ensuring relevant memories surface even in large knowledge bases.
Memory evolves through continuous agent-context interaction. Every session commit triggers automated extraction, profiling, and indexing—building a knowledge base that grows smarter over time without manual intervention.
Built with Rust for production workloads View benchmark results
Modular, scalable, and production-ready
Get started with a single command
Minimal API surface for context operations
use cortex_mem_core::MemoryOperations; // Initialize memory system let memory = MemoryOperations::new(config).await?; // Semantic search let results = memory.search("user preferences", filters).await?; // Access via cortex:// URIs let content = memory.read("cortex://user/001/profile.json").await?;
Modular components for every use case
Core library with memory operations
REST API server (Axum-based)
Command-line interface tool
Web dashboard (Svelte 5)
Model Context Protocol server
Rig framework adapter