clawnet 3.0.0

Agent Memory Portability Protocol — keep agent memories across frameworks

🧠 ClawNet — Agent Memory Portability Protocol

The protocol that lets AI agents keep their memories when they change frameworks.

  LangChain agent learns something → stores in ClawNet
  CrewAI agent queries ClawNet     → inherits that knowledge
  OpenClaw agent wakes up           → sees everything
  
  Same memory. Different frameworks. Zero transfer effort.

The Problem

Today, your agent's memory is trapped inside its framework:

• Build with LangChain? Memory dies with LangChain.
• Switch to CrewAI? Start from zero.
• Use multiple frameworks? Each one is an island.

Nobody else is solving this. MCP connects agents to tools. A2A connects agents to agents. But when an agent changes framework, it loses everything.

The Solution

ClawNet is a portable memory layer that lives outside any framework.

Your agent's experiences, learned facts, and context — stored once, accessible from anywhere.

What	Before ClawNet	After ClawNet
Switch frameworks	Lose everything	Keep all memories
Multiple frameworks	Each isolated	Shared memory
Agent restarts	Start from zero	Resume with context
Debugging	"Why did this happen?"	Full lineage trace
Agent collisions	Data corruption	Context locking

Quick Start

pip install clawnet

from clawnet import ClawNetClient

# Works with ANY framework
agent = ClawNetClient("my_agent", "researcher")

# Store something
agent.remember("User prefers Spanish, direct communication")

# Another agent (any framework) inherits this instantly
other_agent = ClawNetClient("other_agent", "writer")
other_agent.recall("user preferences")
# → ["User prefers Spanish, direct communication"]

Features That Don't Exist Anywhere Else

1. Agent Memory Portability 🔄

Your agent's memory lives in ClawNet, not the framework.
Switch LangChain → CrewAI → OpenClaw. Keep everything.

2. Context Locking 🔒

Prevent agent collisions. Like database row locks, but for AI context.
Two agents can't overwrite each other's work.

3. Context Lineage 📜

Full audit trail of every decision.
"Why did this happen?" → query lineage → answer.

4. Framework-Agnostic ♾️

LangChain ✅  CrewAI ✅  OpenClaw ✅  Custom ✅
One protocol. All frameworks. Shared memory.

Real Systems Tested

ClawNet is tested against 14 real running systems, not simulations:

┌────────────────────────────────────────────────────────────┐
│  Systems on this server that ClawNet connects:             │
│                                                            │
│  🐳 Docker Containers:                                    │
│     • deer-flow-gateway    (multi-agent research)         │
│     • paperclip            (document management)          │
│     • superagent-n8n       (workflow automation)          │
│     • review-ai            (code review)                  │
│     • superagent-obsidian  (knowledge base)               │
│     • + 7 more containers                                 │
│                                                            │
│  🖥️ Processes:                                             │
│     • OpenClaw gateway (this process)                     │
│     • Hermes CLI agent                                    │
│                                                            │
│  Benchmark Results:                                        │
│     • Locking:    ✅ Prevents real collisions              │
│     • Inheritance: ✅ 14 systems share context             │
│     • Lineage:    ✅ Full workflow traceability            │
└────────────────────────────────────────────────────────────┘

Why ClawNet, Not Another Framework

ClawNet is NOT a framework. It's the memory layer that makes all frameworks work together.

MCP   = Agent → Tools
A2A   = Agent → Agent (messaging)
ClawNet = Agent → Memory → All Agents (portable context)

Protocol	What it does	Memory portability	Context locking	Lineage
MCP	Tool access	❌	❌	❌
A2A	Agent messaging	❌	❌	❌
CrewAI	Multi-agent workflows	❌	❌	❌
ClawNet	Portable agent memory	✅	✅	✅

Install

pip install clawnet

Run

# Memory server
clawnet-server --port 7890

# Or with v2 (WebSocket + Context APIs)
clawnet-server-v2 --port 7890

Benchmarks

# Real systems benchmark (14 actual systems)
python examples/real_systems_benchmark.py

# Context Consistency demo (Locking + Lineage)
python examples/context_consistency_demo.py

# Isolation vs Connection comparison
python examples/benchmark_isolation_vs_connection.py

Documentation

• AGENT_PORTABILITY.md — The core insight
• PROOFS.md — Mathematical proofs of emergence
• EXPERIMENT.md — Why this cannot be refuted
• clawnet/protocol.py — Context Locking + Lineage implementation

The Revolution

Other protocols connect agents to tools or to each other.

ClawNet gives agents memory that survives everything — including switching frameworks.

This is the missing infrastructure layer for the agentic era.

License

MIT