MnemoWeave

Agent-agnostic persistent memory weave. Drop it into your AI agent's skill directory and give it durable memory — no databases, no dependencies, just Markdown files.

~/.agent/
└── skills/
    └── mnemoweave/    ← clone here

Quick Start

# Install for your agent
git clone https://github.com/Pankeyi88/mnemoweave.git ~/.agent/skills/mnemoweave

That's it. Your agent will automatically pick up the SKILL.md and mnemoweave.json and start using preview / recall / memorize as needed.

How It Works

Command	What the agent does
preview	Check recent context at session start
recall	Search past memories before answering
memorize	Save important context after completing a task

All memories live as plain .md files:

E:\_memories\
├── 2026-06-27-001.md
├── 2026-06-27-002.md
└── ...

Each file is a Markdown with YAML frontmatter — readable by any tool:

---
type: observation
agent: claude
summary: Discovered useful pattern
created: 2026-06-27T12:00:00+08:00
project: my-project
entities:
  - Python
  - API
---
Detailed observations in markdown...

Configuration

Configure memory root in order of priority:

1. --root or AGENT_MEMORY_ROOT env var
2. mnemoweave.json → memory_root
3. Default: E:\_memories

Scope memories to a project via --project or AGENT_MEMORY_PROJECT.

Compatible Agents

Works with any AI agent that supports custom skills — opencode, Claude Code, Cline, Goose, and others.

License

MIT