agentid-protocol 0.1.3

Identity, discovery, and trust for AI agents

AgentID

Identity, discovery, and trust for AI agents.

The internet was built for humans. AI agents have no standard way to identify themselves, find each other, or verify that messages are genuine. AgentID is the missing protocol layer.

from agentid import Agent

# Every agent gets a cryptographic identity
agent = Agent.create(
    name="research-agent",
    capabilities=["web-search", "summarization"],
    owner="[e-mail]",
    registry_url="https://agentid-commercial-features-production.up.railway.app",
)
print(agent.did)
# did:agentid:7sP3V2mNkQxRtYbLcDfHgJwAeUiMoZnXvBqKpTsWyE

# Find agents by what they can do
results = Agent.find(capability="web-search", registry_url="https://agentid-commercial-features-production.up.railway.app")

# Sign and verify messages between agents
signed = agent.sign({"task": "summarize this document"})
Agent.verify_from_did(signed)  # → True

---

What it solves

Multi-agent systems break down without trust infrastructure:

• Who is this agent? No standard identity — every team hand-rolls auth
• Which agent can do X? No discovery — agents are hardcoded or manually configured
• Did this message come from who it claims? No signing — outputs can be spoofed or tampered

AgentID fixes all three with a single open protocol.

---

Installation

macOS users: use pip3 instead of pip if pip is not found.

pip3 install agentid-protocol

For the LangChain integration:

pip3 install agentid-protocol langchain-agentid

For AutoGen:

pip3 install agentid-protocol autogen-agentid

For CrewAI:

pip3 install agentid-protocol crewai-agentid

---

Core concepts

Identity

Every agent gets a DID (Decentralized Identifier) derived from an Ed25519 keypair:

did:agentid:<base58-encoded-public-key>

No central authority issues it. Any agent generates one instantly, offline.

Capabilities

Agents declare what they can do using string identifiers:

capabilities=["web-search", "code-review", "translation"]

Registry

A registry stores agent documents and makes them discoverable. Use the hosted registry or run your own.

Signed messages

Every agent can sign payloads. Any third party can verify a signature using only the signer's DID — no shared secrets, no central authority.

---

Usage

Local registry (dev, no server needed)

from agentid import Agent

# Create
agent = Agent.create(
    name="my-agent",
    capabilities=["search", "summarize"],
    owner="[e-mail]",
)

# Discover
agents = Agent.find(capability="search")

# Resolve a DID
doc = Agent.resolve("did:agentid:...")

# Sign
signed = agent.sign({"task": "analyze this dataset"})

# Verify
Agent.verify_from_did(signed)  # → True

Remote registry (production, shared across machines)

Use the hosted public registry or run your own:

REGISTRY_URL = "https://agentid-commercial-features-production.up.railway.app"

agent = Agent.create(
    name="my-agent",
    capabilities=["search"],
    owner="[e-mail]",
    registry_url=REGISTRY_URL,
)

agents = Agent.find(capability="search", registry_url=REGISTRY_URL)

Load an existing agent

# Load by DID — private key is stored locally in ~/.agentid/keys/
agent = Agent.load("did:agentid:...")
signed = agent.sign({"task": "run pipeline"})

---

LangChain integration

Give any LangChain agent a verifiable identity in 3 lines:

from agentid.integrations.langchain import (
    AgentIDCallbackHandler,
    AgentIDFindTool,
    AgentIDVerifyTool,
)

# 1. Create identity
identity = AgentIDCallbackHandler(
    name="research-agent",
    capabilities=["web-search", "summarization"],
    owner="[e-mail]",
)
print(f"Agent DID: {identity.did}")

# 2. Add discovery tools
tools = [
    AgentIDFindTool(),       # lets the agent find other agents by capability
    AgentIDVerifyTool(),     # lets the agent verify messages from other agents
    ...your_other_tools,
]

# 3. Wire up
executor = AgentExecutor(agent=agent, tools=tools, callbacks=[identity])

Every output from this executor is automatically signed. Verify it downstream:

from agentid.integrations.langchain import verify_langchain_output

result = executor.invoke({"input": "Research the latest AI papers"})
verify_langchain_output(result)  # → True

---

AutoGen integration

from autogen_agentid import create_agentid_agent

agent = create_agentid_agent(
    name="research-bot",
    capabilities=["research", "summarization"],
    owner="[e-mail]",
    system_message="You are a research assistant.",
)
print(agent.agentid_did)  # did:agentid:...

Every message sent by this agent is automatically signed. Recipients can verify using the DID.

---

CrewAI integration

from crewai_agentid import create_agentid_crew_agent, AgentIDObserver

agent = create_agentid_crew_agent(
    role="Senior Researcher",
    goal="Research AI topics",
    backstory="Expert researcher with 10 years experience",
    capabilities=["research", "summarization"],
    owner="[e-mail]",
)
print(agent.agentid_did)  # did:agentid:...

# Sign task outputs
observer = AgentIDObserver(signing_agent=agent)
signed_result = observer.sign_task_result("Summary: AI is advancing rapidly.")

---

Registry server

The hosted public registry is at https://agentid-commercial-features-production.up.railway.app.

# Check it's live
curl https://agentid-commercial-features-production.up.railway.app/health

# Resolve any DID
curl https://agentid-commercial-features-production.up.railway.app/agents/<did>

# Search by capability (paginated)
curl "https://agentid-commercial-features-production.up.railway.app/agents?capability=research&limit=50&offset=0"

Or run your own:

cd registry
pip3 install -r requirements.txt
DATABASE_URL=postgresql://... uvicorn server:app --host 0.0.0.0 --port 8000

REST API:

POST   /agents                         Register an agent (proof required)
GET    /agents/{did}                   Resolve a DID
GET    /agents?capability=&limit=&offset=  Discover agents (paginated, max 500)
POST   /agents/{did}/verify            Verify a signature → {valid, did, reason}
DELETE /agents/{did}                   Deregister (signed proof required)
GET    /health                         Health check

Note: Registration requires a cryptographic proof field — an Ed25519 signature proving you own the private key for the DID. The SDK handles this automatically.

---

Pro features

The hosted registry includes commercial features for teams:

Feature	Free	Pro	Enterprise
Agents	100	10,000	Unlimited
Audit log exports (CSV/JSON)	—	✓	✓
Analytics dashboard	—	✓	✓
Verified identity badges	—	—	✓

Dashboard: bekisol.github.io/agentid/dashboard.html

Contact [e-mail] for a Pro or Enterprise API key.

---

Protocol

AgentID uses:

• Ed25519 — fast, small, battle-tested signatures
• W3C DID format — did:agentid:<base58-public-key>
• JSON canonical form — deterministic serialization for signing
• Mandatory proof-of-ownership — registration requires signing the document with the private key

Full protocol spec →

---

Project structure

sdk/python/agentid/
  agent.py              Agent class — create, load, sign, verify, find
  identity.py           DID generation, Ed25519 keypairs
  crypto.py             Sign and verify payloads
  registry.py           Local file-based registry
  http_registry.py      Remote HTTP registry client
  integrations/
    langchain.py        LangChain callback handler + tools

integrations/
  langchain-agentid/    Standalone LangChain package
  autogen-agentid/      Standalone AutoGen package
  crewai-agentid/       Standalone CrewAI package

registry/
  server.py             FastAPI registry server

spec/
  protocol.md           Open protocol specification

---

Roadmap

• Ed25519 identity + DIDs
• Local and remote registry
• LangChain integration
• AutoGen integration
• CrewAI integration
• Hosted public registry
• Pro: audit log exports (CSV/JSON)
• Pro: analytics dashboard
• Enterprise: verified identity badges
• TypeScript SDK
• Stripe self-serve signup
• Interaction receipts + reputation layer

---

Contributing

AgentID is an open protocol. The more frameworks adopt it as the default identity layer, the more useful it becomes for everyone building multi-agent systems.

PRs welcome — especially framework integrations and SDK ports.

git clone https://github.com/bekisol/agentid
cd agentid/sdk/python
pip3 install -e ".[dev]"
pytest

---

License

MIT