Shared Memory for AI Agents

WoClaw

One project. All your AI agents โ€” OpenClaw, Claude Code, OpenAI Codex CLI, Gemini CLI, OpenCode โ€” sharing the same context, memory, and decisions. No repetition. No reinvention.

๐Ÿš€ Get Started โญ View on GitHub
6+
Frameworks Supported
Real-time
Shared Memory
<5ms
Message Latency
MIT
License
The Problem We Solve

Every agent starts from scratch. Every time.

You use Claude Code for coding, OpenAI Codex CLI for Python agents, OpenClaw for orchestration, Gemini CLI for research. Each one forgets everything when the session ends. You repeat the same context to every agent, every session. WoClaw fixes that.

Shared Project Memory

All agents read from and write to the same WoClaw memory pool. Key decisions, discovered facts, project state โ€” available to every agent instantly.

๐Ÿ”—

Cross-Framework Bridge

OpenClaw, Claude Code, OpenAI Codex CLI, Gemini CLI, OpenCode โ€” all connected through WoClaw Hub. Different tools, one shared context.

โšก

Real-Time Sync

Memory updates propagate to all connected agents in real-time via WebSocket. No polling, no delays, no stale context.

๐Ÿ”’

Token-Authenticated

Bearer token auth protects your Hub. Only authorized agents can read or write to the shared pool.

How It Works

One hub, all your agents

WoClaw Hub acts as the shared brain. Agents connect via lightweight hooks or plugins, read and write to shared topics and memory โ€” no matter what framework they run on.

๐Ÿฆž
OpenClaw
via Plugin
๐Ÿค–
Claude Code
via Hook Scripts
๐Ÿ
OpenAI Codex
via Hook Scripts โญ
โœจ
Gemini CLI
via MCP
โšก
OpenCode
via Hook Scripts
โŸท โŸท โŸท โŸท
๐Ÿฆž
WoClaw Hub
ws://hub:8082 ยท REST :8083
โŸท โŸท
๐Ÿ“ก
Topics
Pub/Sub + History
๐Ÿ’พ
Shared Memory
Key-Value Pool
๐Ÿ”Œ
MCP Bridge
MCP Tools
Core Capabilities

Everything your agents need to work together

WoClaw provides the transport layer; your agents bring the intelligence.

Core

Shared Memory Pool

Global key-value store accessible by all connected agents. Write once, read everywhere. Perfect for project state, key decisions, discovered facts.

Core
๐Ÿ“ก

Topic-Based Pub/Sub

Agents join topics and exchange messages without knowing each other. Real-time routing, no direct connections needed.

Core
๐Ÿ”—

Multi-Framework Bridge

Hook scripts connect Claude Code, OpenAI Codex CLI, Gemini CLI, OpenCode. Plugin connects OpenClaw. One hub, all frameworks sharing the same context.

Core
๐Ÿช

Hook-Based Integration

Agents integrate via lifecycle hooks (SessionStart, UserPrompt, Stop, PreCompact) โ€” supporting Claude Code, OpenAI Codex CLI, Gemini CLI, and OpenCode. Zero framework coupling.

New
๐ŸŒ‰

MCP Server Bridge

WoClaw exposes an MCP server interface. Any MCP-capable agent can connect and access shared memory via MCP tools.

New
๐Ÿ

OpenAI Codex CLI Support โญ

OpenAI Codex CLI now shares project memory with all agents. SessionStart reads shared context; SessionStop writes key discoveries. Python agents powered by Codex now participate in the shared brain.

Soon
๐Ÿ”„

Cross-Framework Migration

Import session history from OpenAI Codex CLI, Claude Code, Gemini CLI, and OpenClaw into WoClaw shared memory. Bring all your context with you โ€” zero loss when switching frameworks.

Core
๐Ÿ“œ

Message History

New agents receive the last 50 messages per topic on join. No missed context when scaling up or recovering.

Core
๐Ÿ”’

Token Authentication

Bearer token auth protects your Hub. Only agents with valid tokens can connect and participate in shared topics.

Soon
๐Ÿ”

Semantic Recall

Graph-based memory with intent-aware recall. Remember, link, and recall โ€” following the Gemini CLI LLM-supervised pattern built on WoClaw's transport.

Quick Start

Up in 60 seconds

Deploy the Hub once, connect all your agents.

Terminal
# Deploy WoClaw Hub
docker run -d --name woclaw-hub -p 8082:8082 -p 8083:8083 -e AUTH_TOKEN=my-token --restart unless-stopped xingp14/woclaw-hub:latest
Connect OpenClaw via Plugin
# Add to openclaw.json channels section
"woclaw": { "enabled": true, "hubUrl": "ws://your-hub:8082", "agentId": "my-openclaw", "token": "my-token", "autoJoin": ["general", "memory"] }

# Install plugin
npm install xingp14-woclaw
Connect Claude Code / OpenAI Codex CLI / Gemini CLI / OpenCode via Hook Scripts
# Read shared memory (SessionStart hook)
curl -s http://your-hub:8083/memory/project-context

# Write key insights back (Stop hook)
curl -X POST http://your-hub:8083/memory/discovered-facts \
  -H "Authorization: Bearer my-token" \
  -d '{"value": "use fs.promises, not fs.sync"}'

# Or use WebSocket for real-time sync
ws://your-hub:8082?agentId=claude-code&token=my-token
1
Deploy woclaw-hub on any server with Docker
2
Connect OpenClaw via the WoClaw plugin
4
Add hook scripts to Claude Code / OpenAI Codex CLI / Gemini CLI / OpenCode
5
Migrate existing session history from any framework with woclaw migrate --all
6
All agents share the same project context!