Skip to content
TakoAPI
Back to skills

muninn-memory

by phillipneho

Search & Research
—Not yet rated
Your rating

Brief

Memory system for AI agents.

Description

What This Skill Does

Local-first memory system for AI agents that stores episodic, semantic, and procedural memories in a SQLite database. Uses Ollama for local embeddings to enable hybrid BM25 and semantic search. All data stays on the user's machine with no external network calls.

Runs entirely locally with no API keys or cloud dependencies, and scores 93% on the LOCOMO benchmark compared to 66.9% for Mem0.

When to Use It

  • Persisting user preferences across agent sessions
  • Recalling past meeting notes by topic or person mentioned
  • Building a knowledge base from ongoing conversations
  • Tracking project context across long-running tasks
  • Storing step-by-step workflows for later retrieval

Example Workflow

INPUT

User asks: 'Remember that I met with Sarah about the Q3 roadmap today'

AGENT
  1. 1

    Initializes MemoryStore with the local SQLite database

  2. 2

    Calls remember() with type 'episodic' and entities ['Sarah', 'Q3 roadmap']

  3. 3

    Assigns a salience score and stores the memory

  4. 4

    On follow-up 'What did I discuss with Sarah?', calls recall() with hybrid BM25 and semantic search

  5. 5

    Returns ranked results with entity-boosted relevance

OUTPUT

The Q3 roadmap meeting memory is retrieved and surfaced to the agent

Requirements

  • Ollama installed and running locally

Installation

clawhub install phillipneho/muninn-memory

Run the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.

Links