# Cerememory > Cerememory is an open-source, neuroscience-inspired memory database that gives AI systems persistent, evolving memory across sessions. It is LLM-agnostic (Claude, GPT, Gemini, and others), user-sovereign, local-first, and built in Rust. ## Core concepts (read this first) - **CMP (Cerememory Protocol)** is the single, transport-agnostic protocol for every operation against Cerememory. All messages are versioned and organized into four categories: `encode.*`, `recall.*`, `lifecycle.*`, `introspect.*`. - **HTTP, gRPC, and MCP are transport bindings for CMP**, not separate APIs. HTTP and gRPC expose the full CMP surface. MCP exposes a curated 15-tool subset designed for LLM agents. When an LLM agent calls an MCP tool, the tool invokes CMP under the hood. - **Five memory stores**, each modeled after a brain region: Episodic (hippocampus), Semantic (neocortex), Procedural (basal ganglia), Emotional (amygdala), Working (prefrontal cortex). - **Living memory dynamics**: memories decay over time on a power-law curve, accumulate interference noise, are modulated by an 8-dimensional emotion vector, and can reactivate through spreading activation. - **Two recall modes**: `Human` (realistic recall with fidelity-weighted noise) and `Perfect` (complete retrieval of original data). - **User-sovereign**: local-first, fully exportable to a single-file JSON Lines CMA archive (with optional ChaCha20-Poly1305 + Argon2id encryption), MIT-licensed. - **Implementation**: Core engine in Rust (redb, Tantivy, MessagePack, Axum + Tonic + rmcp). Vector search uses a deterministic redb-backed exact cosine scan. Distributed as source only — you clone the repository and build the `cerememory` binary with `cargo build -p cerememory-cli --release`. There are no published GitHub Releases, crates.io, Docker, PyPI, or npm artifacts. - **MCP operation mode**: run one long-lived `cerememory serve` process per data directory and point every MCP client at it with `cerememory mcp --server-url http://127.0.0.1:8420`. `--server-url` is required — there is no embedded-store MCP mode. - Website: https://co-r-e.github.io/cerememory-docs/ - Full documentation (single file for LLMs): https://co-r-e.github.io/cerememory-docs/llms-full.txt - Source: https://github.com/co-r-e/cerememory ## Getting Started - [Introduction](https://co-r-e.github.io/cerememory-docs/docs/getting-started/introduction): Learn what Cerememory is and how it gives AI systems persistent, living memory. - [Installation](https://co-r-e.github.io/cerememory-docs/docs/getting-started/installation): Build Cerememory from source and verify the binary. - [Quick Start](https://co-r-e.github.io/cerememory-docs/docs/getting-started/quick-start): Start the shared Cerememory server, connect an MCP client, and store your first memory. - [MCP & Agent Setup](https://co-r-e.github.io/cerememory-docs/docs/getting-started/agent-memory): Run one Cerememory server and point every MCP client at it over the supported shared-server mode. ## Architecture - [Architecture Overview](https://co-r-e.github.io/cerememory-docs/docs/architecture/overview): Understand Cerememory's layered architecture and crate structure. - [Five Memory Stores](https://co-r-e.github.io/cerememory-docs/docs/architecture/five-stores): Detailed explanation of Cerememory's five brain-inspired memory stores. - [Living Dynamics](https://co-r-e.github.io/cerememory-docs/docs/architecture/living-dynamics): How memories decay, accumulate noise, respond to emotion, and evolve over time. - [Meta-Memory Plane](https://co-r-e.github.io/cerememory-docs/docs/architecture/meta-memory): The structured 'why' plane attached to every record — intent, rationale, evidence, decisions, and a typed context graph. - [Data Model](https://co-r-e.github.io/cerememory-docs/docs/architecture/data-model): Core data structures: MemoryRecord, FidelityState, EmotionVector, and more. ## Protocol - [CMP Protocol Overview](https://co-r-e.github.io/cerememory-docs/docs/protocol/overview): Introduction to the Cerememory Protocol (CMP) -- the single transport-agnostic API for living memory. MCP is a transport for CMP, not a separate protocol. - [Encode Operations](https://co-r-e.github.io/cerememory-docs/docs/protocol/encode): Store and update memories using CMP encode operations. - [Recall Operations](https://co-r-e.github.io/cerememory-docs/docs/protocol/recall): Query and retrieve memories using multimodal cues, association traversal, timelines, graph subgraphs, and raw journal forensic recall. - [Lifecycle Operations](https://co-r-e.github.io/cerememory-docs/docs/protocol/lifecycle): Manage memory dynamics: consolidation, decay, mode switching, forgetting, export, and import. - [Introspect Operations](https://co-r-e.github.io/cerememory-docs/docs/protocol/introspect): Observe system state, inspect records, forecast decay, and monitor evolution. ## Deployment - [Configuration](https://co-r-e.github.io/cerememory-docs/docs/deployment/configuration): Configure Cerememory via TOML file, environment variables, and CLI flags. - [Operations](https://co-r-e.github.io/cerememory-docs/docs/deployment/operations): Run the Cerememory binary as a long-lived service: process supervision, health probes, metrics, and backups. ## Reference - [API Endpoints](https://co-r-e.github.io/cerememory-docs/docs/reference/api-endpoints): Complete HTTP REST endpoint reference for the Cerememory Protocol. - [CLI Reference](https://co-r-e.github.io/cerememory-docs/docs/reference/cli): Complete reference for all Cerememory CLI commands and flags. - [LLM Adapters](https://co-r-e.github.io/cerememory-docs/docs/reference/llm-adapters): Configure and use Claude, GPT, and Gemini adapters for embedding, summarization, and relation extraction. ## Contributing - [Contributing Guide](https://co-r-e.github.io/cerememory-docs/docs/contributing/guide): How to contribute to Cerememory: PR process, commit conventions, and code style. - [Development Setup](https://co-r-e.github.io/cerememory-docs/docs/contributing/development): Set up your local development environment for Cerememory development.