🧠 Memory Sync

An OpenClaw skill that syncs context across channels — Discord, Telegram, webchat, and more.

The Problem

Every channel in OpenClaw gets its own session. Say something on Discord, switch to Telegram, and it's gone. You end up repeating yourself.

How It Works

Memory Sync uses shared memory files as a bridge between isolated sessions:

Discord ──┐                  ┌── Discord
Telegram ─┤── writes to ──►  │
Webchat ──┘   memory files   ├── reads from ──► All sessions stay in sync
              (daily + long   │
               term)          └── Telegram, Webchat, etc.

Three mechanisms work together:

Mechanism	How	When
Auto Sync	Cron job scans all sessions, extracts key context, writes to shared memory	Every 2 hours (configurable)
Immediate Write	Agent writes important info to memory files in the same turn	Real-time, every session
Handoff Detection	Agent checks other sessions when it detects a channel switch	On demand

Installation

Copy into your OpenClaw workspace skills directory:

git clone https://github.com/rajwaitforit/memory-sync.git ~/.openclaw/workspace/skills/memory-sync

Or into managed skills (shared across agents):

git clone https://github.com/rajwaitforit/memory-sync.git ~/.openclaw/skills/memory-sync

Setup

1. Add the sync cron job

Use the OpenClaw CLI or cron tool to add a job:

openclaw cron add --name memory-sync \
  --schedule "0 */2 * * *" \
  --tz "America/Los_Angeles" \
  --session-target isolated \
  --agent-turn \
  --message "Read /path/to/skills/memory-sync/SKILL.md and run the auto-sync workflow."

Or via the gateway cron API — see OpenClaw cron docs for details.

Recommended model: Use a cheap, fast model for the sync job — it's just reading and summarizing. Examples: google/gemini-2.0-flash, openai/gpt-4o-mini, anthropic/claude-3-haiku.

2. Update your AGENTS.md

Add these two sections to your workspace AGENTS.md:

Immediate Write Rule:

### ✍️ Immediate Write Rule
When the user tells you anything meaningful — a preference, decision, task, request,
personal info, opinion, or new context — write it to memory RIGHT NOW in the same turn.
Sessions are siloed per channel. If you don't write it down, other sessions won't know.

Channel Handoff Awareness:

### 🔄 Channel Handoff Awareness
When you detect the user is coming from a different channel (or mentions something
you don't have context for), proactively check:
1. memory/YYYY-MM-DD.md (today + yesterday)
2. Recent session histories via sessions_list + sessions_history

3. Create the memory directory

mkdir -p ~/.openclaw/workspace/memory

Verify It Works

1. Say something on one channel: "I'm working on Project Alpha"
2. Wait for sync (or say "sync my memory")
3. Switch channels and ask: "What am I working on?"
4. The agent should know about Project Alpha

Configuration

Sync frequency

Use Case	Schedule	~Cost/day
Heavy multi-channel	Every hour	$0.50
Default	Every 2 hours	$0.25
Light use	Every 4 hours	$0.12
Budget	3x daily	$0.06

Costs are estimates based on typical session sizes with a cheap model.

What gets synced

The sync job extracts:

• Decisions and preferences
• Tasks and action items
• Personal info and context
• Project updates and blockers

It skips: credentials, routine chatter, greetings, duplicate info.

File Structure

memory-sync/
├── SKILL.md           # Skill definition and sync workflows
├── README.md          # This file
├── LICENSE            # MIT
└── references/
    └── setup-guide.md # Troubleshooting and advanced config

Compatibility

• OpenClaw 2026.2.x+
• Requires: sessions_list, sessions_history, cron tools
• Works with any LLM provider configured in OpenClaw

Contributing

PRs welcome. Ideas:

• Smarter deduplication
• Memory file pruning (auto-archive old daily files)
• Conflict resolution when sessions disagree

License

MIT