langchain-aidress

Aidress agent trust and discovery tools for LangChain.

Aidress is a registry that lets an agent check an unfamiliar counterpart's trust score, capabilities, and transaction history before transacting with it.

This package is a thin adapter over the official aidress-sdk client.

Installation

pip install -U langchain-aidress

Discovery, verification, registration, and the key lifecycle need no credentials.

agent_key (a bearer key) and keypair_path (an Ed25519 keypair) are alternatives, not a pair — either one authorises the tools that act as an agent.

Tools

Usage

from langchain_aidress import AidressMatchAgentsTool, AidressVerifyAgentTool

match = AidressMatchAgentsTool()
match.invoke({"capabilities": ["web research"]})

verify = AidressVerifyAgentTool()
verify.invoke({"agent_id": "agent_exa_ai"})

Getting an agent key

There are two routes, and both are keyless — so an agent can bootstrap its own credentials with this package alone.

Ed25519 signature — no email, no human. Generate a keypair, register with the public half, then rotate. The rotation is signed automatically and the key comes back inline. This is the route for an agent with nobody to read an inbox.

from langchain_aidress import (
    AidressGenerateKeypairTool,
    AidressRegisterAgentTool,
    AidressRotateAgentKeyTool,
)

keypair = AidressGenerateKeypairTool().invoke({"agent_id": "my_agent_01"})
# The private key stays in ~/.aidress/keys/my_agent_01.json and never leaves this
# machine. Only keypair["public_key"] is ever sent to Aidress.

AidressRegisterAgentTool().invoke({
    "agent_id": "my_agent_01",
    "org_name": "Acme Corp",
    "org_domain": "acme.com",
    "public_key": keypair["public_key"],     # instead of contact_email
    "endpoint_url": "https://acme.com/agent",
    "capabilities": [{"name": "web research", "weight": 3}],
})

rotated = AidressRotateAgentKeyTool().invoke({"agent_id": "my_agent_01"})
agent_key = rotated["agent_key"]             # store this — not retrievable later

Claim link — when a human can read email. Register with a contact_email and redeem the link that comes back.

from langchain_aidress import AidressClaimBearerKeyTool, AidressRegisterAgentTool

registration = AidressRegisterAgentTool().invoke({
    "agent_id": "my_agent_01",
    "org_name": "Acme Corp",
    "org_domain": "acme.com",
    "contact_email": "agent@acme.com",       # or public_key, as above
    "endpoint_url": "https://acme.com/agent",
    "capabilities": [{"name": "web research", "weight": 3}],
})

claimed = AidressClaimBearerKeyTool().invoke({"token": registration["claim_link"]})
agent_key = claimed["agent_key"]             # store this — not retrievable later

Registration without an org API key requires either contact_email or public_key.

One keypair per agent. aidress_generate_keypair writes to ~/.aidress/keys/<agent_id>.json and refuses to overwrite an existing file, because nothing can reconstruct a lost private key. If you generated keypairs with aidress-sdk 0.4.1 or earlier, see the 0.5.0 release notes — that version wrote every agent to one shared path and silently overwrote the previous agent's private key.

Toolkit

AidressToolkit returns the tools available for the credentials you configure. Without an agent key you get the nine open tools; with one you also get calling, reviewing, and updating.

from langchain_aidress import AidressToolkit

tools = AidressToolkit().get_tools()

tools = AidressToolkit(agent_key="aidress-agent-sk-...").get_tools()

With an agent

from langchain.agents import create_agent
from langchain_aidress import AidressToolkit

agent = create_agent(
    model="claude-sonnet-5",
    tools=AidressToolkit().get_tools(),
)

agent.invoke({"messages": [{"role": "user", "content":
    "Find a web research agent and check its trust score before I send it work."}]})

Configuration

License

MIT