Introduction

What is Cerememory?

Cerememory is a living memory database for the age of AI. Built in Rust, it provides persistent, brain-inspired memory that any LLM can use -- regardless of provider. Memories stored in Cerememory are not static rows in a table; they decay, evolve, form associations, and respond to emotion, just as biological memory does.

The Problem: LLM Amnesia

Every time an LLM conversation ends, everything learned is lost. Context windows are finite and ephemeral. This creates a fundamental limitation for AI applications that need to remember users, learn from past interactions, or build long-term understanding.

Current workarounds -- RAG databases, vector stores, chat history logs -- treat memory as a search problem. They return exact copies of stored text ranked by similarity. Biological memory works nothing like this. It is reconstructive, associative, and alive.

Core Principles

Alive

Memories in Cerememory are living entities. They decay over time following a power-law model, accumulate noise, and can be reactivated through retrieval. Emotional memories persist longer. Frequently accessed memories grow more stable. This is not a feature -- it is the foundation.

LLM-Agnostic

Cerememory works with any LLM provider through the Cerememory Protocol (CMP). Dedicated adapters for Claude, GPT, and Gemini serialize memory context into each model's preferred format, but the core engine is completely provider-independent.

User-Sovereign

All memory data belongs to the user. Cerememory runs on your infrastructure, stores data locally using redb (an embedded key-value store), and supports encrypted export/import via CMA archives. There is no cloud dependency, no telemetry, and no vendor lock-in.

Brain-Inspired Design

Cerememory models its architecture after the human brain's memory systems:

Each store has distinct characteristics for decay rates, consolidation behavior, and retrieval patterns. The hippocampal coordinator manages cross-store operations like spreading activation and memory consolidation.

What Can You Do With Cerememory?

• Store memories with automatic store classification, embedding generation, and association inference
• Attach structured meta-memory to every record — intent, rationale, evidence, alternatives, decisions, and a typed context graph that explains why a memory exists
• Preserve verbatim conversations in the raw journal for forensic recall
• Recall memories with human-like fidelity (noise, decay) or perfect retrieval mode
• Search the "why" plane with recall.query — meta-memory text is indexed alongside content
• Run dream processing to summarize raw journal entries into curated episodic and semantic memories
• Traverse association graphs across all memory stores via spreading activation
• Run background decay ticks that naturally age and prune stale memories
• Consolidate episodic memories into semantic knowledge over time
• Export and import encrypted memory archives (including raw journal) for portability
• Integrate with any LLM via HTTP, gRPC, or MCP transports — the MCP proxy works with Claude Code, OpenAI Codex CLI, Cursor, Cline, Windsurf, Zed, Continue, and any other MCP-compatible client

How It Is Distributed

Cerememory is distributed as source. You clone the repository and run cargo build -p cerememory-cli --release to produce the single cerememory binary. There are no published binary artifacts, no crates.io / PyPI / npm packages, and no prebuilt Docker images — every deployment runs a binary you built and verified yourself.

Next Steps

Installation

Build Cerememory from source and verify the binary

Quick Start

Start the shared server and connect an MCP client