MaluDB Core

What MaluDB Core is

Not just “what do we have?” — also “how did we get here?”

MaluDB is built on PostgreSQL 17 and is designed for long-term institutional memory and scalability.

It is a database management system for both people and AI agents that stores memories and historical knowledge alongside data — delivered as a set of extensions on PostgreSQL, the most trusted enterprise database.

MaluDB doesn't only store your workflows and knowledge: it turns them into skills that AI agents can use to solve tasks while saving tokens.

The MaluDB system: people reach shared project memory pools through SQL and an API, AI agents through native MCP, and all of them draw on the same deep memories in MaluDB. — People and agents work from the same project memory pools: people through SQL and the API, agents through native MCP.

Our core premises

Four beliefs that shape MaluDB

Memories, not just facts

A database must store memories — context, provenance and meaning — not just isolated facts.

One memory system

Developers should not have to stitch several databases together to get a working memory system.

Real-time shared memory

Agents need to share memory in real time — collaborating on the same fresh, evolving context.

Database-governed access

Access to LLMs and MCP servers should be governed by database security — not bolted on afterwards.

Architecture

An extension layer on a PostgreSQL foundation

Memories, knowledge, workflows and skills for people and AI agents — in one enterprise database management system.

MaluDB architecture: an extension layer of knowledge ingestion, hybrid search across relational, vector, graph, document and temporal modes, database-governed inference with local LLMs and Claude, ChatGPT and Gemini, database-governed MCP, and managed skills, on a PostgreSQL foundation with verbatim storage and shared memory pools.

Shared memory for agent teams

Memories, knowledge, facts

Compartmentalized hybrid search

MC2DB — database-driven MCP

Built-in LLM control

Skills objects

Knowledge ingestion

Memory pipeline

What MaluDB does

An enterprise database for teams of people and AI agents

LLM access and MCP server security belong in the enterprise database. With MaluDB you control the prompts executed, the MCP tools and the data AI uses — at the database level, down to row-level security.

Historical knowledge ingestion

Emails, chat transcripts, tickets, meeting notes, documents, source control, operational logs, time series and relational databases — each artifact preserved verbatim as a Source Package. Decades of scattered history become an addressable, reprocessable base for memory.

Real-time context sharing

Active Memory Pools give people, agents and tools a shared, task-scoped working space: load authorized memories, publish observations, mark pending claims and promote validated knowledge — so every agent works from the same fresh, permission-filtered context.

One platform for memory search

Four search modes working together, with permissions enforced inside candidate generation — no more stitching three products together to answer one question.

Relational Vector Graph Temporal

Why a memory DBMS

Organizations store data. They keep losing the story behind it.

Records, tickets, messages, logs, dashboards and documents show what happened. The reasons, assumptions, evidence and lessons stay scattered across tools — and are rebuilt by hand every time a team changes or a system migrates.

MaluDB treats this as a database concern. Memory Objects are governed by the system: identity, provenance, validity, permissions, lifecycle and audit live in the catalog — not in application code. Storage engines, embedding models, graph indexes and LLMs remain replaceable mechanisms attached through controlled interfaces.

White paper · Zenodo

A Scalable Database Management System for Long-Term Institutional Memory, Human-AI Knowledge Sharing, and Contextual Recall

Read the white paper