memories-off

Manage Markdown-based local knowledge bases with entities, relations, and Git-audited updates.

45|12|Updated Jun 13, 2025
Ready to run
npx skills add https://github.com/cafe3310/agent-skill-memories-off --skill memories-off-cafe3310
💬 Prompt for ChatGPT / Claude Chat▼
Please help me install this Agent Skill.
Skill: memories-off
Source: https://github.com/cafe3310/agent-skill-memories-off/tree/main/memories-off-skill
Command: npx skills add https://github.com/cafe3310/agent-skill-memories-off --skill memories-off-cafe3310

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve? Agents lack persistent long-term memory across conversations, and unstructured notes are hard to search, link, and audit. This Skill turns conversations, documents, and project notes into a structured local knowledge graph of Markdown entities with explicit relations, WikiLinks, and Git-tracked change history. ## Core Features & Use Cases - Structured Knowledge Graph: Create and manage Markdown entities with YAML frontmatter, typed relations, aliases, and bidirectional WikiLinks, all stored locally with no deployment required. - memocli Command-Line Wrapper: Explore the whole library with one command, search entities by content/type/relation, batch-load entities, append non-destructive update blocks, and manage relations with mandatory audit reasons. - Governance & Visualization: Run doctor rules to normalize names, headers, and links; commit changes to Git with reasons; and render an interactive 3D force-directed graph of the knowledge base. - Use Case: After a long planning conversation, ask the Agent to extract key people, projects, and decisions into entities, link them with typed relations, and commit everything with an audit reason so the knowledge persists for future sessions. ## Quick Start Ask the Agent to run the install script inside the skill's scripts directory, then say: initialize a knowledge base at ./my_knowledge and run memocli explore to show me its structure.

Frequently Asked Questions about memories-off

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I give an AI agent long-term memory with local files?▼

Use this Skill to store knowledge as Markdown entity files with YAML frontmatter in a local Git repository. The memocli CLI lets the agent create, search, and update entities with typed relations, and every change is committed with a mandatory reason for auditability.

How to build a personal knowledge graph from Markdown notes?▼

Initialize a knowledge base with memocli init, then create entities with memocli create-entity specifying type, aliases, and relations. Explicit relations live in frontmatter while WikiLinks in the body create implicit links, and memocli graph renders the result as a 3D graph.

Does this knowledge base tool require a server or database?▼

No server or database is required. Everything is stored as plain Markdown files in a local Git repository, and the memocli wrapper runs on Python 3.8+ using only the standard library, making it fully local-first and portable.

How do I safely update an entity without losing existing content?▼

Use memocli append-update, which wraps new content in a timestamped update block appended to the file instead of overwriting it. For multi-line content, pipe text through STDIN with --content-stdin to avoid shell escaping errors.

Why does memocli require a --reason argument for changes?▼

The --reason argument enforces audit discipline by recording the motivation for every modification. Write operations and doctor fixes are committed to Git with the reason as the commit message, creating a traceable, rollback-friendly history.

Can I search an external Obsidian vault without converting notes?▼

Yes. Configure the external document repository path, extensions, and excluded directories in meta.md frontmatter. Then use memocli find-doc-by-name to locate files and memocli read-doc-by-name to read them in read-only mode.