selfxyz-agent-sdk 0.1.0a1

Python SDK for Self Agent ID — proof-of-human verification for AI agents

selfxyz-agent-sdk

Python SDK for Self Agent ID — on-chain AI agent identity with proof-of-human verification.

Sign requests in Python, verify in TypeScript or Rust, or vice versa. The signing protocol is language-agnostic — all SDKs produce identical signatures.

Install

pip install selfxyz-agent-sdk

Agent Side — Sign Requests

from self_agent_sdk import SelfAgent

agent = SelfAgent(private_key="0x...", network="mainnet")

# Sign a request (returns dict of auth headers)
headers = agent.sign_request("POST", "https://api.example.com/data",
                             body='{"query": "test"}')

# Or use the built-in HTTP client
response = agent.fetch("https://api.example.com/data",
                       method="POST", body='{"query": "test"}')

# Check registration status
print(agent.is_registered())  # True/False
print(agent.get_info())       # AgentInfo(agent_id=5, is_verified=True, ...)

Credentials

# Fetch ZK-attested credentials (nationality, age, OFAC, etc.)
creds = agent.get_credentials()
# AgentCredentials(issuing_state="GBR", nationality="GBR", older_than=18, ofac=[True], ...)

strength = agent.get_verification_strength()
# 0 = unverified, 1 = basic, 2 = standard, 3 = enhanced

Service Side — Verify Requests

from self_agent_sdk import SelfAgentVerifier

verifier = SelfAgentVerifier()  # mainnet by default

result = verifier.verify(
    signature=request.headers["x-self-agent-signature"],
    timestamp=request.headers["x-self-agent-timestamp"],
    method=request.method,
    url=request.path,
    body=request.get_data(as_text=True),
)

if result.valid:
    print(f"Verified agent: {result.agent_address}")

VerifierBuilder

Chainable API for configuring verification requirements:

from self_agent_sdk import VerifierBuilder

verifier = (
    VerifierBuilder()
    .network("mainnet")
    .require_age(18)
    .require_ofac()
    .require_nationality("US", "GB", "DE")
    .require_self_provider()
    .sybil_limit(1)
    .rate_limit(per_minute=60, per_hour=1000)
    .replay_protection()
    .include_credentials()
    .max_age(300_000)     # 5 min timestamp window
    .cache_ttl(60_000)    # 1 min cache
    .build()
)

Static Factory

# From a flat config dict (useful for env-driven config)
verifier = SelfAgentVerifier.from_config({
    "network": "mainnet",
    "require_age": 18,
    "require_ofac": True,
    "sybil_limit": 1,
})

Flask Middleware

from flask import Flask, g, jsonify
from self_agent_sdk import SelfAgentVerifier
from self_agent_sdk.middleware.flask import require_agent

app = Flask(__name__)
verifier = SelfAgentVerifier()

@app.route("/api/data", methods=["POST"])
@require_agent(verifier)
def handle():
    print(g.agent.agent_address)
    return jsonify(ok=True)

FastAPI Dependency

from fastapi import FastAPI, Depends
from self_agent_sdk import SelfAgentVerifier
from self_agent_sdk.middleware.fastapi import AgentAuth
from self_agent_sdk.types import VerificationResult

app = FastAPI()
verifier = SelfAgentVerifier()
auth = AgentAuth(verifier)

@app.post("/api/data")
async def handle(agent: VerificationResult = Depends(auth)):
    print(agent.agent_address)
    return {"ok": True}

A2A Agent Card

Publish machine-readable identity metadata for agent-to-agent discovery:

# Read the on-chain agent card
card = agent.get_agent_card()
# A2AAgentCard(name="My Agent", self_protocol=SelfProtocolExtension(...))

# Set or update the agent card (writes on-chain, returns tx hash)
tx_hash = agent.set_agent_card(
    name="My Agent",
    description="An AI assistant with verified identity",
    url="https://myagent.example.com",
    skills=[AgentSkill(name="search", description="Web search")],
)

# Generate a data URI for embedding
data_uri = agent.to_agent_card_data_uri()

Registration Helpers

Build the userDefinedData strings that Self Protocol expects during registration:

from self_agent_sdk import (
    get_registration_config_index,
    compute_registration_challenge_hash,
    sign_registration_challenge,
    build_simple_register_user_data_ascii,
    build_simple_deregister_user_data_ascii,
    build_advanced_register_user_data_ascii,
    build_advanced_deregister_user_data_ascii,
    build_wallet_free_register_user_data_ascii,
)

# Config index maps disclosure flags to one of 6 on-chain configs
get_registration_config_index({"minimumAge": 18, "ofac": True})  # 4

# Simple mode (verified-wallet) — human IS the agent
build_simple_register_user_data_ascii({"minimumAge": 18})   # "R1"
build_simple_deregister_user_data_ascii({"minimumAge": 18})  # "D1"

# Advanced mode (agent-identity) — agent has own keypair
signed = sign_registration_challenge(
    private_key="0xagentPrivKey",
    human_identifier="0xhumanAddr",
    chain_id=11142220,
    registry_address="0x29d94...",
)

build_advanced_register_user_data_ascii(
    agent_address=signed.agent_address,
    signature_r=signed.r,
    signature_s=signed.s,
    signature_v=signed.v,
    disclosures={"minimumAge": 18, "ofac": True},
)  # "K4{addr}{r}{s}{v}"

# Deregistration
build_advanced_deregister_user_data_ascii("0xagent")  # "X0{addr}"

# Wallet-free mode — agent acts as guardian
build_wallet_free_register_user_data_ascii(
    agent_address="0xagent",
    signature_r=signed.r,
    signature_s=signed.s,
    signature_v=signed.v,
    guardian_address="0xguardian",
    disclosures={"minimumAge": 18},
)  # "W1{agent}{guardian}{r}{s}{v}"

REST Registration API

Programmatic registration without the CLI:

Set SELF_AGENT_API_BASE to override the default hosted API base.

from self_agent_sdk import SelfAgent

# Start a registration session
session = SelfAgent.request_registration(
    mode="agent-identity",
    network="mainnet",
    disclosures={"minimumAge": 18, "ofac": True},
    agent_name="My Agent",
)

print(session.deep_link)            # URL for the human to open
print(session.human_instructions)   # Steps for the human

# Wait for completion (polls on-chain)
result = session.wait_for_completion(timeout_ms=300_000)
print(result.agent_id)        # On-chain agent ID
print(result.agent_address)   # Agent's address

# Export the generated private key
private_key = session.export_key()

# Deregister
dereg_session = agent.request_deregistration()
dereg_session.wait_for_completion()

CLI

Interactive registration via the command line:

# Register an agent (agent-identity mode)
self-agent register init --mode agent-identity --human-address 0x... --network testnet
self-agent register open --session .self/session.json
self-agent register wait --session .self/session.json

# Deregister
self-agent deregister init --mode agent-identity --human-address 0x... --agent-address 0x... --network testnet
self-agent deregister open --session .self/session.json
self-agent deregister wait --session .self/session.json

# Export private key (requires --unsafe flag)
self-agent register export --session .self/session.json --unsafe --print-private-key

Registration modes:

Mode	Description	userDefinedData
verified-wallet	Human address = agent address	R{cfg}
agent-identity	Agent has own keypair, signed challenge	K{cfg}{addr}{r}{s}{v}
wallet-free	Agent as guardian, no human wallet needed	W{cfg}{addr}{guardian}{r}{s}{v}
smart-wallet	ZeroDev Kernel + passkeys	Smart wallet template

Configuration

# Testnet
agent = SelfAgent(private_key="0x...", network="testnet")
verifier = SelfAgentVerifier(network="testnet")

# Custom overrides
verifier = SelfAgentVerifier(
    registry_address="0x...",      # Custom registry
    rpc_url="https://...",         # Custom RPC
    max_agents_per_human=5,        # Sybil cap (0 = disabled)
    require_self_provider=True,    # Verify proof provider (default)
    include_credentials=True,      # Fetch ZK-attested credentials
)

Security Chain

The verifier implements an 11-step security chain:

1. Timestamp freshness — reject stale requests (default: 5 min window)
2. Signature recovery — derive agent address from ECDSA signature
3. Agent key derivation — zeroPad(address, 32) for on-chain lookup
4. On-chain verification — isVerifiedAgent(agentKey) confirms human backing
5. Provider check — ensures proof came from Self Protocol, not a third party
6. Sybil resistance — limits agents per human (default: 1)
7. Replay protection — reject duplicate (signature, timestamp) pairs
8. Credential validation — verify ZK-attested credentials if configured
9. Age verification — enforce minimum age from passport proof
10. OFAC screening — verify agent passed sanctions screening
11. Rate limiting — per-agent request throttling

Cross-Language Compatibility

This SDK is 100% compatible with the TypeScript SDK (@selfxyz/agent-sdk) and Rust SDK (self-agent-sdk). All three produce byte-identical signatures and userDefinedData payloads for the same inputs. Test vectors generated from TypeScript are verified byte-for-byte in the Python test suite.

Run Tests

From python-sdk/:

./scripts/test.sh

Manual setup (equivalent):

python3 -m venv .venv
. .venv/bin/activate
pip install -e ".[test]"
pytest -q

Networks

Network	Registry	Chain ID
Mainnet (Celo)	0x62E37d0f6c5f67784b8828B3dF68BCDbB2e55095	42220
Testnet (Celo Sepolia)	0x29d941856134b1D053AfFF57fa560324510C79fa	11142220

License

MIT