scrypster/muninndb

The cognitive database. A new class of data storage. Not a vector store, not a graph DB, not a RAG wrapper. Ebbinghaus decay, Hebbian learning, and Bayesian confidence are engine-native primitives. Memories evolve on their own. MCP-native. Single binary.

What it solves

MuninnDB is a cognitive memory database designed to provide AI agents and applications with persistent, human-like memory. Unlike traditional databases that store data statically, MuninnDB manages "memory traces" (engrams) that evolve over time, strengthening when used and fading when ignored, mimicking biological memory processes.

How it works

The system uses a 6-phase activation pipeline to retrieve relevant memories based on context. It combines full-text and vector search with several cognitive primitives:

  • Hebbian Learning: Automatically forms associations between memories that are activated together without requiring a predefined schema.
  • Temporal Weighting: Recalculates relevance based on how recently and frequently a memory has been accessed (Ebbinghaus decay).
  • Predictive Activation: Tracks sequential patterns to surface the next likely memory before it is explicitly requested.
  • Bayesian Confidence: Tracks the certainty of each memory, increasing with reinforcement and decreasing with contradictions.

It is delivered as a single binary with zero dependencies and supports multiple protocols including MCP (Model Context Protocol), REST, gRPC, and a proprietary binary protocol (MBP).

Who it’s for

Developers building AI agents, LLM-powered applications, and tool-integrated AI environments (like Cursor, Claude Desktop, or Windsurf) who need a memory system that handles association and temporal relevance automatically.

Highlights

  • Plug-and-Play AI Integration: Built-in support for MCP, allowing it to be added to AI tools with minimal configuration.
  • Zero Dependency: Single binary installation with a bundled local embedder for offline use.
  • Proactive Memory: Features semantic triggers that push relevant memories to the user or agent when relevance changes.
  • Retroactive Enrichment: Ability to upgrade existing memories with new embeddings or enrichment plugins without migrations.
  • ** IllegalArgument Exception: Bulk Insert**: Supports batching up to 50 memories in a single call.

Related

  • Project
    laminlabs/lamindb

    LaminDB is an open‑source, Git‑like data‑management system for AI/ML, especially multimodal life‑science data. It version‑controls files, tables, and array formats, tracks code and environment provenance, supports branching/merging, and offers ACID‑safe metadata stored in SQLite/Postgres. Integrated with Python/R tools (Polars, DuckDB), bio‑ontologies, and workflow managers, it enables traceable, FAIR‑compliant datasets for research and biotech.

  • Project
    nuclia/nucliadb

    NucliaDB is a hybrid AI search database that combines vector, full-text, and graph indexes to store and search unstructured data.

  • Project
    rtk-ai/icm

    ICM (Infinite Context Memory) is a Rust‑based, zero‑dependency binary that provides a shared, searchable SQLite memory store for AI agents. It supports episodic memories with importance‑based decay, permanent knowledge‑graph memoirs, feedback recording, and verbatim session transcripts. Integrated via MCP hooks or a CLI, it works with many popular assistants (Claude, Gemini, Codex, Copilot, etc.) and offers optional ONNX‑Runtime semantic search. Install via Homebrew, a one‑liner script, or Cargo; configure globally or per‑project, then store/recall context across tools without re‑explaining. Experimental (pre‑1.0) but usable today.

  • Project
    mnemosyne-oss/mnemosyne

    Mnemosyne is a local-first, SQLite-backed memory layer for AI agents. It gives assistants persistent memory across sessions via a three-tier architecture (working memory, episodic memory, temporal knowledge graph), with hybrid semantic/keyword search, heavy vector compression, a Python SDK, CLI, and built-in MCP server. It's privacy-focused: no telemetry, local by default, with optional client-side-encrypted sync.