maludb

Maludb Core — AI skill for Claude Code

AI community

A memory DBMS for long-term institutional memory, human-AI knowledge sharing, and contextual recall.

How to install Maludb Core

This entry records only its repository, not the path inside it, so there is no exact command to give. Open maludb/maludb-core and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Maludb Core does

A memory DBMS for long-term institutional memory, human-AI knowledge sharing, and contextual recall. PostgreSQL 17 extension.

Alternatives in AI

Hindsight — State-of-the-art long-term memory for AI agents by Vectorize 6.7k ★
Deepseek Pp — DeepSeek Web browser extension: AI agent workspace with MCP tools, memory, Skills, automation, web search, and 1.7k ★
Nocturne Memory — A lightweight, rollbackable, and visual Long-Term Memory Server for MCP Agents 1.3k ★

README

MaluDB

MaluDB is a memory DBMS for long-term institutional memory, human-AI knowledge sharing, and contextual recall. Built in **C** as PostgreSQL extensions on **Ubuntu 24.04 LTS**, with **PostgreSQL 17** (PGDG) as the foundation.

**New here?** Start with the [executive summary](executive-summary.md) — what MaluDB is, the memory model, and why retrieval is relational/graph-first with vector search reserved for the query classes that genuinely need it.

Status

Version 0.106.0 (extension) — the engine learns who is asking: tenant-scoped principals and scope grants, three session settings (maludb_core.principal_ref / principal_scopes / principal_readonly; unset = unrestricted, as before), principal_ref + scope on source packages, documents, memories, episodes, chat sessions and pools, enforced for reads by a RESTRICTIVE policy and inside the SECURITY DEFINER search workers, for writes by trigger, with sensitivity checked against the principal's ceiling; forgetting works — tombstones are honoured on the exact and exact_parallel search paths (only local_ann filtered them, so a deleted memory stayed recallable), maludb_forget_document() / maludb_forget_chunk(), and a deleted document takes its chunks with it; skills gain review_state, principal grants and load events; pools gain presence facades; maludb_memory_ingest_extraction takes a namespace. Re-run enable_memory_schema per tenant after upgrading. On 0.105.3 text comparisons use their indexes inside functions called with a name argument (#35): PL/pgSQL gave such functions collation "C", so comparisons against the extension's default-collation indexes fell back to scanning — the dirty-queue purge fired per deleted statement read the whole queue, and memory search filtered every compartment of a schema; 17 functions now compare with COLLATE "default", and the collation_index_use regress test lists any comparison with that shape. On 0.105.2 **


Version	0.106.0 (extension) — the engine learns who is asking: tenant-scoped principals and scope grants, three session settings (`maludb_core.principal_ref` / `principal_scopes` / `principal_readonly`; unset = unrestricted, as before), `principal_ref` + `scope` on source packages, documents, memories, episodes, chat sessions and pools, enforced for reads by a RESTRICTIVE policy and inside the SECURITY DEFINER search workers, for writes by trigger, with `sensitivity` checked against the principal's ceiling; forgetting works — tombstones are honoured on the `exact` and `exact_parallel` search paths (only `local_ann` filtered them, so a deleted memory stayed recallable), `maludb_forget_document()` / `maludb_forget_chunk()`, and a deleted document takes its chunks with it; skills gain `review_state`, principal grants and load events; pools gain presence facades; `maludb_memory_ingest_extraction` takes a namespace. Re-run `enable_memory_schema` per tenant after upgrading. On 0.105.3 text comparisons use their indexes inside functions called with a `name` argument (#35): PL/pgSQL gave such functions collation "C", so comparisons against the extension's default-collation indexes fell back to scanning — the dirty-queue purge fired per deleted statement read the whole queue, and memory search filtered every compartment of a schema; 17 functions now compare with `COLLATE "default"`, and the `collation_index_use` regress test lists any comparison with that shape. On 0.105.2 **