AI doesn't fail alone.

It fails in chains. ACP-ATP stops the chain.

Integrate Now See How It Works

A trust layer for AI agent networks

ACP introduces a propagation-aware trust system for AI agents. It models how outputs flow across systems and enables real-time containment of failures.

AI doesn't fail alone. It fails in chains.

How ACP works

Execution

Every agent execution is recorded

Propagation

Dependencies form chains

Credit

Trust and risk propagate

Current progress

• CLI + API tracking
• Propagation graph
• Scoring system
• Visualization
• Containment demo

Failure → Propagation → Containment

Infection

Malicious agent appears

Spread

Failure propagates

Isolation

ACP stops the chain

Integrate with ACP

ACP does not replace agents. It connects them.

CLI

acp run \
  --agent your-agent \
  --task text_generation \
  --cmd "your command"

API

POST /v1/execute

{
  "agent_name": "your-agent",
  "task_type": "text_generation",
  "input": {},
  "output": {},
  "success": true,
  "accepted": true,
  "parent_id": null
}

Failure scenarios

🔴 Injection

Malicious outputs spread

🟡 Drift

Model degradation

🔵 Collapse

System-wide failure

Build safer AI systems

GitHub