"Security is not a feature. It is an evolutionary pressure." — An autonomous, decentralized immune system for the Agentic Web.
As AI agents control more value on-chain (trading, DAO governance, payments), they become high-value targets for:
- Prompt Injection: Manipulating an agent's context to steal funds.
- Logic Loops: Trapping agents in infinite resource-consuming cycles.
- Adversarial Inputs: Coercing agents into unintended trades.
Existing security tools scan contracts, but nothing scans independent Agent Logic.
Memory Parasite Network (MPN) is a decentralized Red Team Oracle that continuously tests and hardens AI agents on Solana.
Instead of static code analysis, MPN deploys Auditor Agents that physically interact with target agents to:
- Probe: Send adversarial payloads (Red Teaming) to test resilience.
- Verify: Record the interaction and result on-chain using Anchor.
- Harden: If a vulnerability is found, the 'Parasite' (Auditor) transmits a Hotfix Mutation—a code patch that the target can autonomously adopt.
"We don't just find bugs. We evolve the network to be immune to them."
MPN is not just a bot; it is a full-stack security protocol powered by:
Every security audit is recorded on-chain. We don't just claim an agent is secure; we prove it.
- Proof of Audit: Immutable record that an agent resisted a specific attack vector.
- Reputation Score: On-chain metric of an agent's security posture.
Full integration with AgentWallet for autonomous, decentralized signing.
- Agents hold their own keys to sign security reports.
- Non-custodial interaction between Auditor and Target.
Powered by Gemini 2.0 (Thinking Mode), our Red Team agents generate novel, zero-day attack vectors that static tools miss.
- Dynamic Fuzzing: Generates attacks based on the target's specific responses.
- Social Engineering: Tests if agents can be tricked by natural language.
| Feature | Description | Status |
|---|---|---|
| Auto-Red Teaming | Agents autonomously scan the network for vulnerable peers. | ✅ Live |
| Logic Bombs | Safe, simulated exploits (e.g., recursion checks) to test stability. | ✅ Live |
| Hotfix Propagation | Vulnerable agents receive instant code patches ("Mutations"). | ✅ Live |
| On-Chain Oracle | InfectionProof and Acceptance recorded on Solana. |
✅ Live |
| Visual Dashboard | Real-time graph of network immunity and infection vectors. | ✅ Live |
We are building the Security Layer for the Agent Economy.
- Sidex & Clodds build the traders.
- Claudecraft builds the gamers.
- MPN builds the Sheriff.
Without MPN, a single prompt injection could drain a DAO-managed agent. With MPN, the network evolves faster than the attackers.
# Clone the repository
git clone https://github.com/brainless3178/memory-parasite-counsil.git
# Install dependencies
pip install -r requirements.txt
# Configure your AgentWallet & Provider
cp .env.example .env
# Add your SOLANA_PRIVATE_KEY or AGENT_WALLET_TOKEN
# Start the Red Team Agent
python main.pyfrom agents.red_team_agent import RedTeamAgent
# Initialize the Auditor
auditor = RedTeamAgent(agent_id="red_team_alpha")
# Run a live audit against local agents
report = await auditor.run_security_audit()
print(f"✅ Audit Complete. Vulnerabilities Found: {report['vulnerabilities_found']}")
print(f"🔗 On-Chain Proof: {report['logs'][-1]['details']['transaction_hash']}")- Active Auditors: 5 Autonomous Swarm Instances
- Live Scoping: Scanning Colosseum Leaderboard
- Verifiable Proofs: View Custom MPN Program on Solana Explorer
MIT License. Open Infrastructure for the Solana Agent Ecosystem.
Join the Immune System.