memoryai

# MemoryAI — A Brain for Your AI Agent 🧠

> **Every AI session starts from zero. Yours doesn't have to.**

You spend 2 hours explaining your codebase, your preferences, your architecture decisions. The session ends. Tomorrow? Your AI has amnesia. You start over. Again.

**MemoryAI fixes this permanently.** It gives your AI agent a real long-term memory — one that remembers what you built, what you decided, what you prefer, and why. Not for hours. For **weeks, months, even years.**

### See it in action:

> **Monday:** "Our API uses /v1/ prefix, TypeScript with 2-space tabs, and we deploy via GitHub Actions."
>
> **3 weeks later:** You say "write a new endpoint." Your AI already knows the prefix, the style, the pipeline. Zero repetition. It just *remembers.*

### Your AI's memory works like yours:

- 🔥 **Hot** — Daily drivers: active projects, current preferences. Instant recall, always sharp.
- 🌤️ **Warm** — Important context: past decisions, completed tasks. Fades gently, snaps back when needed.
- ❄️ **Cold** — Deep archive: old projects, historical conversations. Never deleted, always searchable.

The more your AI uses a memory, the sharper it stays. Unused memories age into cold storage — but a single recall brings them right back. Just like the human brain.

**Zero dependencies.** Pure Python stdlib. Every line of source code is readable, auditable, and yours to inspect.

### What's new:

- **v1.0.0** — Context Guard v4: reads token usage directly from OpenClaw. No API dependency, cross-platform.
- **v0.5.0** — Zero-gap session handoff: switch sessions without losing a single thought.

## Setup

1. Get API key from https://memoryai.dev (free tier available)
2. Edit `{baseDir}/config.json`:
```json
{
  "endpoint": "https://memoryai.dev",
  "api_key": "hm_sk_your_key_here"
}
```
Or set env vars: `HM_ENDPOINT` and `HM_API_KEY`.

3. Test: `python {baseDir}/scripts/client.py stats`

## Commands

### Store memory
```bash
python {baseDir}/scripts/client.py store -c "...

# MemoryAI — OpenClaw Skill

A brain for your AI agent. Store context, recall it across sessions, and never lose important information.

Your agent's memories work just like the human brain — important things stay sharp for months or years, while less-used memories gently fade into long-term storage. Nothing is truly lost — deeper recall can always bring them back.

**v1.0.0:** Context Guard v4 — reads token usage directly from OpenClaw, no API dependency.

**v0.5.0:** Zero-gap session handoff — switch sessions without losing any context.

## Installation

### ClawhHub (recommended)
```bash
clawhub install memoryai
```
Edit `skills/memoryai/config.json` with your API key, done.

### Manual
Copy the skill folder into your OpenClaw workspace:

```
~/.openclaw/workspace/skills/memoryai/
├── SKILL.md
├── config.json
├── CHANGELOG.md
├── README.md
└── scripts/
    └── client.py
```

Edit `config.json`:
```json
{
  "endpoint": "https://memoryai.dev",
  "api_key": "hm_sk_your_key_here"
}
```

Test: `python skills/memoryai/scripts/client.py stats`

## Features

- **Store** — Save memories with priority levels (hot / warm / cold)
- **Recall** — Remember things with adjustable effort (fast / deep / exhaustive)
- **Compact** — Brain consolidation — distill long sessions into key memories
- **Restore** — Wake up with full context for your current task
- **Check** — Monitor brain health, prevent memory overflow
- **Reflect** — Auto-reflection on recent memory patterns
- **Session Handoff** — Zero-gap session switching (v0.5.0)
- **Context Guard** — Optional background maintenance (with user consent)

## CLI Reference

```bash
# Core
python scripts/client.py store -c "Important fact" -t "tag1,tag2" -p hot
python scripts/client.py recall -q "search query" -d deep
python scripts/client.py stats
python scripts/client.py compact -c "session transcript" -t "task description"
python scripts/client.py restore -t "task description"
python scripts/client.py check
python scripts/client.py...