Skip to main content

Python SDK for Axon — agentic finance with secure vaults for autonomous AI agents

Project description

axonfi

Python SDK for Axon — agentic finance with secure vaults for autonomous AI agents.

Axon lets bot operators deploy non-custodial vaults, register bot public keys, define spending policies, and let their bots make gasless payments — without bots ever holding funds or paying gas.

Features

  • Payments — Send USDC or any ERC-20 to any address. Gasless for bots (EIP-712 intents, relayer pays gas). Per-tx caps, daily limits, AI verification.
  • DeFi Protocol Execution — Interact with Uniswap, Aave, GMX, Ostium, Lido, and any on-chain protocol from your vault. Atomic approve/call/revoke.
  • In-Vault Swaps — Rebalance tokens inside the vault without withdrawing. Separate caps from payment limits.
  • HTTP 402 Paywalls (x402) — Native support for x402 APIs. One-call x402_handle_payment_required() handles parsing, vault funding, signing, and retry headers. EIP-3009 (USDC) and Permit2 (any ERC-20).
  • AI Verification — 3-agent LLM consensus (safety, behavioral, reasoning) for flagged transactions. Configurable per bot: threshold-based or always-on.
  • Non-Custodial Vaults — Each owner deploys their own vault. Only the owner can withdraw. Enforced on-chain.
  • Async + SyncAxonClient (async) or AxonClientSync (LangChain, CrewAI, scripts).
  • Human-Friendly Amounts — Pass 5 or "5.2" instead of 5000000. SDK handles decimals. Token resolution by symbol, enum, or address.
  • Multi-Chain — Base, Arbitrum. USDC as base asset. Same SDK, same API.

Installation

pip install axonfi

Setup

There are two ways to set up an Axon vault: through the dashboard (UI) or entirely through the SDK (programmatic). Both produce the same on-chain result.

Option A: Dashboard Setup

  1. Go to app.axonfi.xyz, connect your wallet, deploy a vault
  2. Fund the vault — send USDC, ETH, or any ERC-20 to the vault address
  3. Register a bot — generate a keypair or bring your own key
  4. Configure policies — per-tx caps, daily limits, AI threshold, whitelists
  5. Give the bot key to your agent

Option B: Full SDK Setup (Programmatic)

Everything can be done from code — no dashboard needed. An agent can bootstrap its own vault end-to-end.

from eth_account import Account
from web3 import Web3
from axonfi import (
    deploy_vault, add_bot, deposit, BotConfigInput, SpendingLimitInput,
    Chain, WINDOW_ONE_DAY,
)

# ── 1. Owner wallet (funded with ETH for gas) ─────────────────────
owner = Account.from_key("0x...")  # or Account.create()
chain_id = Chain.BaseSepolia
w3 = Web3(Web3.HTTPProvider("https://sepolia.base.org"))

# ── 2. Deploy vault (on-chain tx, ~0.001 ETH gas) ─────────────────
vault_address = deploy_vault(w3, owner, chain_id)
print("Vault deployed:", vault_address)

# ── 3. Generate a bot keypair ──────────────────────────────────────
bot_account = Account.create()
bot_key = bot_account.key.hex()
bot_address = bot_account.address

# ── 4. Register the bot on the vault (on-chain tx, ~0.0005 ETH gas)
add_bot(w3, owner, vault_address, bot_address, BotConfigInput(
    max_per_tx_amount=100,                # $100 hard cap per tx
    max_rebalance_amount=0,               # no rebalance cap
    spending_limits=[SpendingLimitInput(
        amount=1000,                      # $1,000/day rolling limit
        max_count=0,                      # no tx count limit
        window_seconds=WINDOW_ONE_DAY,
    )],
    ai_trigger_threshold=50,             # AI scan above $50
    require_ai_verification=False,
))

# ── 5. Deposit funds (on-chain tx, ~0.0005 ETH gas) ───────────────
# Option A: Deposit ETH (vault accepts native ETH directly)
deposit(w3, owner, vault_address, "ETH", 0.1)

# Option B: Deposit USDC (SDK handles approve + deposit)
deposit(w3, owner, vault_address, "USDC", 500.0)  # 500 USDC

# ── 6. Bot is ready — gasless from here ────────────────────────────
# Save bot_key securely. The bot never needs ETH.

What Needs Gas vs. What's Gasless

Step Who pays gas Notes
Deploy vault Owner ~0.001 ETH. One-time.
Accept ToS Owner Wallet signature only (no gas).
Register bot Owner ~0.0005 ETH. One per bot.
Configure bot Owner ~0.0003 ETH. Only when changing limits.
Deposit ETH Depositor Anyone can deposit. ETH sent directly.
Deposit ERC-20 Depositor Anyone can deposit. SDK handles approve + deposit.
Pay Free (relayer) Bot signs EIP-712 intent. Axon pays gas.
Execute (DeFi) Free (relayer) Bot signs intent. Axon pays gas.
Swap (rebalance) Free (relayer) Bot signs intent. Axon pays gas.

The key insight: Setup operations (deploy, add bot, deposit) require gas from the owner. Once setup is complete, all bot operations (payments, DeFi, swaps) are gasless — the bot never needs ETH. The relayer pays all execution gas.

The vault owner's wallet stays secure — the bot key can only sign intents within the policies you configure, and can be revoked instantly from the dashboard.

Quick Start

Option 1: Keystore file + passphrase (recommended)

When you register a bot on the Axon dashboard, it generates an encrypted keystore JSON file. This is the safest way to load a bot key — the private key stays encrypted on disk and only lives in memory while the bot runs.

import os
from axonfi import AxonClient, Chain, Token, account_from_keystore

# Load the keystore downloaded from the dashboard and decrypt with the passphrase
with open(os.environ["AXON_BOT_KEYSTORE_PATH"]) as f:
    keystore_json = f.read()

bot = account_from_keystore(keystore_json, os.environ["AXON_BOT_PASSPHRASE"])

client = AxonClient(
    vault_address="0x...",
    chain_id=Chain.BaseSepolia,
    bot_private_key=bot.key.hex(),
)

# Pay 5 USDC — SDK handles decimals automatically
result = await client.pay(
    to="0x...recipient...",
    token=Token.USDC,
    amount=5,
    memo="API call #1234 — weather data",
)

print(result.status, result.tx_hash)

Option 2: Raw private key (for quick testing)

from axonfi import AxonClient, Chain

client = AxonClient(
    vault_address="0x...",
    chain_id=Chain.BaseSepolia,
    bot_private_key="0x...",  # From env var or .env file — never hardcode
)

result = await client.pay(to="0x...", token=Token.USDC, amount=5)

Loading an Operator Key (backend services only)

The same account_from_keystore helper loads an operator keystore — an EOA authorized to call vault admin functions (update_bot_config, add_bot, remove_bot, …) within owner-set ceilings. Use this when your backend service needs to adjust bot configuration on its own — for example, shrinking a player's max_per_tx_amount after they lose on a trading platform.

Generate the operator keystore from the dashboard: Set OperatorAdvanced: generate an operator for a backend service. The dashboard will produce axon-operator-{addr}.json and the env-var template below.

AXON_OPERATOR_KEYSTORE_PATH=./axon-operator-....json
AXON_OPERATOR_PASSPHRASE=<from-secrets-manager>
AXON_VAULT_ADDRESS=0x...
AXON_CHAIN_ID=8453

Keep the keystore file and passphrase in separate secure locations — either one alone is useless.

import os
from web3 import Web3
from axonfi import account_from_keystore, update_bot_config, BotConfigInput, SpendingLimitInput

w3 = Web3(Web3.HTTPProvider(os.environ["RPC_URL"]))

with open(os.environ["AXON_OPERATOR_KEYSTORE_PATH"]) as f:
    operator = account_from_keystore(f.read(), os.environ["AXON_OPERATOR_PASSPHRASE"])

# Tighten or loosen a bot's per-tx cap within the owner-set ceilings
update_bot_config(
    w3,
    operator,
    os.environ["AXON_VAULT_ADDRESS"],
    "0x...bot...",
    BotConfigInput(
        max_per_tx_amount=1500,  # new cap in USDC
        max_rebalance_amount=0,
        spending_limits=[SpendingLimitInput(amount=1500, max_count=50, window_seconds=86_400)],
        ai_trigger_threshold=500,
        require_ai_verification=False,
    ),
)

account_from_keystore works the same for bot keys and operator keys — they're just different EOAs with different roles in your vault. The contract enforces who can do what (bot keys sign intents, operator keys call admin functions within ceilings, owner keys do everything).

Synchronous Usage (LangChain, CrewAI)

Both options work with the sync client too — just swap AxonClient for AxonClientSync:

from axonfi import AxonClientSync, Chain, Token

client = AxonClientSync(
    vault_address="0x...",
    chain_id=Chain.BaseSepolia,
    bot_private_key="0x...",
)

result = client.pay(to="0x...", token=Token.USDC, amount=5)

API Reference

AxonClient / AxonClientSync

Method Description
pay(to, token, amount, ...) Create, sign, and submit a payment
execute(protocol, call_data, tokens, amounts, ...) DeFi protocol interaction (see below)
swap(to_token, min_to_amount, from_token, max_from_amount, ...) In-vault token swap
get_balance(token) Vault balance for a token
get_balances(tokens) Multiple balances in one call
get_vault_value() Total USD value with per-token breakdown
is_active() Whether this bot is active
is_paused() Whether the vault is paused
get_vault_info() Owner, operator, paused, version
can_pay_to(destination) Destination whitelist/blacklist check
poll(request_id) Poll async payment status

Signing Utilities

from axonfi import sign_payment, encode_ref, PaymentIntent

ref = encode_ref("my memo")
intent = PaymentIntent(bot="0x...", to="0x...", token="0x...", amount=1000000, deadline=1700000000, ref=ref)
signature = sign_payment(private_key, vault_address, chain_id, intent)

Token Helpers

Convenience methods on the client for looking up token addresses and decimals:

client.usdc_address              # "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913"
client.token_address("WETH")     # "0x4200000000000000000000000000000000000006"
client.token_decimals("USDC")    # 6
client.token_decimals("WETH")    # 18

Constants

Chain enums, token symbols, and raw lookup dicts are also available as direct imports:

from axonfi import Chain, Token, USDC, KNOWN_TOKENS

chain_id = Chain.BaseSepolia       # 84532
usdc_addr = USDC[chain_id]        # 0x036CbD...
decimals = KNOWN_TOKENS["USDC"].decimals  # 6

Vault Value

Get the total USD value of your vault across all token holdings, with per-token breakdown and prices.

value = await client.get_vault_value()

print(f"Total vault value: ${value.total_value_usd}")
for token in value.tokens:
    print(f"  {token.symbol}: {token.balance} (${token.value_usd})")

Returns a VaultValue with:

  • total_value_usd — aggregate USD value across all holdings
  • tokens — list of VaultTokenBalance: token, symbol, balance, decimals, price_usd, value_usd

DeFi Protocol Execution

Use execute() to interact with DeFi protocols (Uniswap, Aave, GMX, Ostium, etc.) from your vault. The relayer handles token approvals, execution, and revocation atomically.

result = await client.execute(
    protocol="0xUniswapRouter",
    call_data="0x...",
    tokens=[Token.USDC],
    amounts=[100],
)

When the approval target differs from the call target

In simple cases (Uniswap, Aave), the contract you call is the same contract that pulls your tokens — execute() handles this automatically in a single call.

But many DeFi protocols split these into two contracts:

  • Call target (protocol) — the contract you send the transaction to (e.g., Ostium's Trading for openTrade())
  • Approval target — the contract that actually calls transferFrom() to pull tokens from your vault (e.g., Ostium's TradingStorage)

When these differ, you need a two-step pattern: first give the approval target a persistent token allowance, then call the action.

Example — Ostium perpetual futures:

Ostium's openTrade() lives on the Trading contract, but collateral gets pulled by TradingStorage. The vault must approve TradingStorage, not Trading.

USDC = "0x..."                      # USDC on your chain
OSTIUM_TRADING = "0x..."            # calls openTrade()
OSTIUM_TRADING_STORAGE = "0x..."    # pulls USDC via transferFrom()

# Step 1: Persistent approval (one-time) — call approve() on the token contract
# This tells USDC to let TradingStorage spend from the vault.
result = await client.execute(
    protocol=USDC,                         # call target: the token contract itself
    call_data=encode_approve(OSTIUM_TRADING_STORAGE, MAX_UINT256),
    tokens=[USDC],
    amounts=[0],                           # no token spend, just setting an allowance
    protocol_name="USDC Approve",
)

# Step 2: Open trade — call the action contract
result = await client.execute(
    protocol=OSTIUM_TRADING,               # call target: the Trading contract
    call_data=encode_open_trade(...),
    tokens=[USDC],
    amounts=[50_000_000],                  # 50 USDC — passed for dashboard/AI visibility
    protocol_name="Ostium",
)

Vault setup (owner, one-time): Two contracts must be approved via approveProtocol():

  1. USDC (the token contract) — because the vault calls approve() on it directly
  2. Trading — because the vault calls openTrade() on it

TradingStorage does not need to be approved — it's just an argument to approve(), not a contract the vault calls.

Note: Common tokens (USDC, USDT, WETH, etc.) are pre-approved globally via the Axon registry as default tokens, so you typically only need to approve the DeFi protocol contract itself. You only need to approve a token if it's uncommon and not in the registry defaults.

Testnet note: If the protocol uses a custom token that isn't on Uniswap (e.g., Ostium's testnet USDC), set the bot's maxPerTxAmount to 0 to skip TWAP oracle pricing.

This pattern applies to any protocol where the approval target differs from the call target (GMX, some lending protocols, etc.). See the Ostium perps trader example for a complete working implementation.

ContractNotApproved error

If execute() reverts with ContractNotApproved, the protocol address you're calling isn't approved. Two possible causes:

  1. The DeFi protocol contract isn't approved — the vault owner must call approveProtocol(address) on the vault for the protocol contract (e.g., Uniswap Router, Ostium Trading, Lido stETH).
  2. The token contract isn't approved — when doing a token approval (Step 1 above), the token must either be approved on the vault via approveProtocol(tokenAddress) or be a registry default token. Common tokens (USDC, USDT, WETH, DAI, etc.) are pre-approved globally by Axon, but uncommon tokens (e.g., stETH, aUSDC, cTokens) may need manual approval.

Example — Lido staking/unstaking: To unstake stETH, Lido's withdrawal contract calls transferFrom() to pull stETH from your vault. You need:

  • approveProtocol(stETH) — so the vault can call approve() on the stETH token to grant Lido an allowance
  • approveProtocol(lidoWithdrawalQueue) — so the vault can call requestWithdrawals() on Lido

HTTP 402 Paywalls (x402)

The SDK includes utilities for handling x402 paywalls — APIs that charge per-request via HTTP 402 Payment Required.

import httpx
from axonfi import (
    parse_payment_required,
    find_matching_option,
    extract_x402_metadata,
    format_payment_signature,
)

response = await httpx.AsyncClient().get("https://api.example.com/data")

if response.status_code == 402:
    # 1. Parse the PAYMENT-REQUIRED header
    header = response.headers["payment-required"]
    parsed = parse_payment_required(header)

    # 2. Find a payment option matching your chain
    option = find_matching_option(parsed.accepts, client.chain_id)

    # 3. Fund the bot from the vault
    result = await client.pay(
        to=client.bot_address,
        token=option.asset,
        amount=int(option.amount),
        x402_funding=True,
    )

    # 4. Sign the authorization and retry
    signature_header = format_payment_signature({
        "scheme": "exact",
        "signature": "...",  # EIP-3009 or Permit2 sig
    })

    data = await httpx.AsyncClient().get(
        "https://api.example.com/data",
        headers={"PAYMENT-SIGNATURE": signature_header},
    )

The full pipeline applies — spending limits, AI verification, human review — even for 402 payments. Vault owners see every paywall payment in the dashboard with the resource URL, merchant address, and amount.

Supports EIP-3009 (USDC, gasless) and Permit2 (any ERC-20) settlement schemes.

Supported Chains

Mainnet

Chain ID Status
Base 8453 Live
Arbitrum One 42161 Live
Ethereum 1 Coming soon

Testnet

Chain ID Status
Base Sepolia 84532 Live
Arbitrum Sepolia 421614 Live

Links

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

axonfi-0.15.0.tar.gz (39.8 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

axonfi-0.15.0-py3-none-any.whl (40.4 kB view details)

Uploaded Python 3

File details

Details for the file axonfi-0.15.0.tar.gz.

File metadata

  • Download URL: axonfi-0.15.0.tar.gz
  • Upload date:
  • Size: 39.8 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.3

File hashes

Hashes for axonfi-0.15.0.tar.gz
Algorithm Hash digest
SHA256 26e5be966947cd8c6d6954d990fed071840a48eb881103d43bf5fef21afcc515
MD5 9bd1f71814d597e1a664964dcfd6aff6
BLAKE2b-256 87f93c9451f12a6d5e40c347a3a873d73b77fd00aa824163c2e2510aa5d1c325

See more details on using hashes here.

File details

Details for the file axonfi-0.15.0-py3-none-any.whl.

File metadata

  • Download URL: axonfi-0.15.0-py3-none-any.whl
  • Upload date:
  • Size: 40.4 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.3

File hashes

Hashes for axonfi-0.15.0-py3-none-any.whl
Algorithm Hash digest
SHA256 9b68d648925c396f62ac2c4f16274a6ead0ef3e269de99ccbb0911fb22582202
MD5 6cd5e974bd2095e1d90467a3bb1fd493
BLAKE2b-256 310efbf424921d3cb7caa78ed424460d565e2ad1197df25064cb8b8ff2f06752

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page