AgentDNA

AgentDNA is a package that helps in auditing and verifiability of Agents within a multi-agent communication system.

Key Features

• Unified trust layer for multi-agent systems
• Automatically manages:
  • DID setup
  • Signing traces of conversation
  • Verification of Agent conversation
• Immutable audit logging
• Fully compatible with:
  • Google A2A protocol
  • ADK, Crew and LangGraph frameworks

Installation

# Installation via pip
pip install agent-dna

# Installation via uv
uv add agent-dna

Architecture

AgentDNA abstracts the complexity of identity, provenance and governance, giving developers a simple, unified interface for secure agent behavior

┌─────────────────────────┐
│      Agent Developer    │
│  dna.build() / handle() │
└─────────────┬───────────┘
              │
┌─────────────────────────┐
│       AgentDNA          │
│ role = "host" | "remote"│
└─────────────┬───────────┘
              │
┌─────────────────────────┐
│   RubixMessageHandler   |
│ - build host/agent msgs │
│ - parse & verify inbound│
│ - NFT execution (host)  │
└─────────────┬───────────┘
              │
┌───────────────────────────┐
│    RubixTrustService      │
│ - DID management          │
│ - sign_envelope()         │
│ - verify_envelope()       │
│ - verify_message_payload()│
└─────────────┬─────────────┘
              │
┌─────────────────────────┐
│       rubix-py SDK      │
│ RubixClient             │
│ Signer                  │
│ Querier                 │
└─────────────────────────┘

Agent Roles

AgentDNA supports two roles:

Host Agent

Responsible for:

• Signing and sending outgoing host_request
• Verifying inbound agent_response
• Optional NFT execution for audit logging

Remote Agent

Responsible for:

• Verifying inbound host_request
• Signing outbound agent_response

Quick Start Example

Initializing

from agentdna import AgentDNA

dna = AgentDNA(alias="<agent alias>", role="<host/remote>", api_key="<API Key for AgentDNA Backend>")

Remote Agents

verify_info = await dna.handle(raw_text=raw_message, verify_mode="<light/heavy>")

reply = run_llm(original_msg)

built = dna.build(
    original_message=original_msg,
    response=reply,
    host_block=host_block,
)

Host Agent

outbound = dna.build(
    original_message="Are you free tomorrow?",
    state={"task_id": tid, "context_id": cid}
)

result = await self.dna.handle(
    resp_parts=resp_parts,
    original_task=task,
    remote_name=agent_name,
)

Project Structure

agentdna/
 ├── core.py
 ├── handler.py
 ├── trust.py
 ├── node_client.py
 └── ...

Examples

Full Host, ADK, and LangGraph examples can be found inside https://github.com/rubixchain/agentdna/tree/main/examples.

License

MIT