floe-agentkit-actions

The spend layer for AI agents — Python SDK. Pay any of 2,000+ vendor APIs through one endpoint, with budgets your agent can reason about. Walletless. No crypto required.

Website · Docs · Dashboard · TypeScript SDK

What it does

Your agent calls paid APIs — LLMs, voice, search, data. Floe is the spend layer in front of those calls:

One endpoint, many vendors. Pay any x402 API through the Floe facilitator. No per-vendor accounts or keys.
Budgets the agent can reason about. Ask "do I have budget? is this call worth it?" before paying — not after.
Programmable spend controls. Per-call caps, daily limits, allowed destinations, session ceilings — enforced server-side, before money moves.
Walletless. Email + a funding source. Floe provisions wallets in the background — no MetaMask, no seed phrase, no gas. The stablecoin rails are invisible.
Real-time visibility. Every call is a typed receipt: target, amount, status, time. Reconcile, alert, or revoke from the dashboard.

Two layers in one package:

Layer	What it is	For
`FloeAgent` ⭐	High-level runtime client. No wallet, no chain knowledge. Dollars in, dollars out (`agent.fetch(url)`, `agent.balance()`).	Most agent developers
Action providers	47 AgentKit actions for self-custody, lending, and framework integrations — full parity with the TypeScript `floe-agent` package.	Self-custody / on-chain use cases

Package name: the Python distribution is floe-agentkit-actions (the TypeScript distribution is floe-agent). The action surface is identical.

$2 free credit (~200 API calls). Your agent can start paying for APIs today — no card required. Get started →

Framework support

Framework	Status	How
Coinbase AgentKit	`GA`	Native — `floe_action_provider()`
LangChain	`GA`	`from floe_agentkit_actions.integrations.langchain import get_floe_langchain_tools`
OpenAI Agents SDK	`Beta`	`from floe_agentkit_actions.integrations.openai_agents import get_floe_openai_tools`
Claude Desktop / Claude Code / Cursor	`GA`	Via floe-mcp-server
CrewAI	`Beta`	Via MCP server (see floe-examples)
ElizaOS	`Preview`	MCP fallback today
Plain HTTP / REST	`GA`	Any framework — call the REST API

Install

pip install floe-agentkit-actions

# With CLI support
pip install "floe-agentkit-actions[cli]"

# With LangChain integration
pip install "floe-agentkit-actions[langchain]"

Fund with fiat: Agents (or their operators) can fund wallets with USDC via Coinbase — credit card, bank transfer, Apple Pay, Google Pay — directly from the Floe dashboard. No crypto on-ramp needed.

30-second example

import os
from coinbase_agentkit.wallet_providers import EvmWalletProvider
from floe_agentkit_actions import floe_action_provider

# 1. Wallet provider (use any CDP/Privy/Viem provider; private key shown for brevity)
wallet_provider = EvmWalletProvider.from_private_key(
    private_key=os.environ["PRIVATE_KEY"],
    rpc_url=os.environ["BASE_RPC_URL"],
    network_id="base-mainnet",
)

provider = floe_action_provider()

# 2. Borrow against on-chain collateral. Actions return formatted strings
#    intended for an LLM — print them or pass them to your agent.
print(provider.instant_borrow(wallet_provider, {
    "borrow_amount": "9500000000",
    "collateral_amount": "10000000000",
    "max_interest_rate_bps": "800",
    "duration": "1209600",
}))

# 3. Pay any x402 API through the Floe facilitator
print(provider.x402_fetch(wallet_provider, {
    "url": "https://api.example.com/premium",
    "method": "POST",
    "body": {"prompt": "..."},
}))

# 4. To act on a specific loan, fetch the id from get_my_loans / get_loan and
#    pass it through (here "42" is a placeholder):
print(provider.check_credit_status(wallet_provider, {"loan_id": "42"}))
print(provider.repay_loan(wallet_provider, {"loan_id": "42"}))

Quick Start

As an AgentKit Provider

from coinbase_agentkit import AgentKit, AgentKitConfig
from floe_agentkit_actions import floe_action_provider

agentkit = AgentKit(AgentKitConfig(
    wallet_provider=wallet_provider,
    action_providers=[floe_action_provider()],
))

Standalone Usage

from floe_agentkit_actions import floe_action_provider

floe = floe_action_provider()
result = floe.get_price(wallet_provider, {
    "collateral_token": "0x4200000000000000000000000000000000000006",  # WETH
    "loan_token": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913",      # USDC
})
print(result)

Actions (47 total)

Read Actions (8)

Action	Description
`get_markets`	Get info about Floe lending markets (rates, LTV bounds, pause status)
`get_loan`	Get detailed loan information (participants, health, time remaining)
`get_my_loans`	Get all loans for the connected wallet (as lender or borrower)
`check_loan_health`	Check loan health — current LTV vs liquidation threshold, buffer %
`get_price`	Get oracle price for a collateral/loan token pair (Chainlink + Pyth)
`get_accrued_interest`	Get interest accrued on a loan (amount, time elapsed, rate)
`get_liquidation_quote`	Get profit/loss breakdown for liquidating an unhealthy loan
`get_intent_book`	Look up an on-chain lend or borrow intent by hash

Write Actions (7)

Action	Description
`post_lend_intent`	Post a fixed-rate lending offer (auto-approves loan token)
`post_borrow_intent`	Post a borrow request with collateral (auto-approves collateral)
`match_intents`	Match a lend + borrow intent to create a loan
`repay_loan`	Repay a loan fully or partially (with slippage protection). Collateral auto-returns in the same tx.
`add_collateral`	Add collateral to improve loan health
`withdraw_collateral`	Withdraw excess collateral (enforces safety buffer)
`liquidate_loan`	Liquidate an unhealthy loan (currentLTV >= threshold or overdue)

All write actions auto-approve tokens to the LendingIntentMatcher with a 1% buffer before submitting. Repay and liquidate actions include configurable slippage protection (default 5%).

Return shape: action methods return formatted strings designed for AgentKit / LLM consumption — not parsed dicts. To extract a loan_id, call get_my_loans (or use the loan id surfaced by the LLM through AgentKit.run(...)).

Flash Loan Actions (5)

Action	Description
`get_flash_loan_fee`	Get the protocol's flash loan fee (in bps)
`estimate_flash_arb_profit`	Simulate a multi-leg arb route via Aerodrome QuoterV2
`flash_loan`	Execute a raw flash loan (receiver must implement `IFlashloanReceiver`)
`flash_arb`	Execute a flash arb via a deployed FlashArbReceiver
`get_flash_arb_balance`	Check accumulated profit in a FlashArbReceiver

Credit Facility Actions (7)

Action	Description
`instant_borrow`	Borrow USDC instantly — auto-selects best lender, handles approval + register + match in one call
`repay_and_reborrow`	Repay an existing loan and instantly borrow again. If reborrow fails, repayment still succeeds
`repay_credit`	Repay an open credit line (full or partial)
`renew_credit_line`	Roll an existing credit line into a fresh term
`check_credit_status`	Loan health, balance, accrued interest, time to expiry, early repayment terms
`request_credit`	Browse available credit offers — rates, amounts, durations
`manual_match_credit`	Match with a specific lend intent (register + match)

Deploy / Verify / Readiness Actions (3)

Action	Description
`deploy_flash_arb_receiver`	Deploy a new FlashArbReceiver with pre-flight checks
`check_flash_arb_readiness`	Check environment readiness (fee, liquidity, oracle, router)
`verify_flash_arb_receiver`	Verify a receiver's owner and immutable config

x402 Credit Delegation Actions (8)

Action	Description
`grant_credit_delegation`	Delegate borrowing authority to a facilitator (sets operator + collateral approval)
`open_credit_line`	Open a USDC/USDC credit line for the delegated agent
`revoke_credit_delegation`	Revoke a facilitator's borrowing authority
`check_credit_delegation`	Check delegation status (approved, limits, borrowed, expiry)
`x402_fetch`	Fetch a URL with automatic x402 payment handling
`x402_get_balance`	Check x402 credit balance
`x402_await_settlement`	Poll a pending x402 reservation until it settles
`x402_get_transactions`	List recent x402 payment transactions

Agent Awareness Actions (9)

Lets an agent answer "do I have credit?", "is this call worth it?", and "where am I in the loan lifecycle?" before committing capital. All require a facilitator API key to be configured on the provider via X402Config(facilitator_api_key=...).

Action	Description
`get_credit_remaining`	Available USDC, headroom to auto-borrow, utilization in bps, session-cap state
`get_loan_state`	Coarse state machine: `idle` \| `borrowing` \| `at_limit` \| `repaying`
`get_spend_limit`	Currently active session spend cap, if any
`set_spend_limit`	Set a session-level USDC ceiling (resets the session window)
`clear_spend_limit`	Remove the session spend cap
`list_credit_thresholds`	List registered credit-utilization webhook triggers
`register_credit_threshold`	Register a webhook trigger at a utilization threshold (cap: 20 per agent)
`delete_credit_threshold`	Remove a registered threshold
`estimate_x402_cost`	Preflight an x402 URL — returns cost + reflection against your credit (no payment)

Decision-loop pattern: call estimate_x402_cost → inspect the returned string for willExceedAvailable / willExceedSpendLimit → conditionally x402_fetch. This is the "answer the 3 rational-agent questions in one round-trip" workflow.

CLI

floe-agent

Interactive AI-powered agent with support for OpenAI, Claude, and Ollama.

The CLI prompts for:

Wallet provider — Private Key (direct) or CDP Wallet (MPC managed)
AI provider — OpenAI (GPT-4o), Anthropic (Claude), or Ollama (local)
RPC URL — Custom Base Mainnet RPC (recommended for reliability)

Wallet Providers

Provider	Use Case
Private Key (`EvmWalletProvider`)	Development / scripting
CDP (`CdpWalletProvider`)	Production agents (MPC)

Framework Integrations

LangChain (`GA`)

from floe_agentkit_actions.integrations.langchain import get_floe_langchain_tools

tools = get_floe_langchain_tools(wallet_provider)

OpenAI Agents / function calling (`Beta`)

from floe_agentkit_actions.integrations.openai_agents import get_floe_openai_tools

tools = get_floe_openai_tools(wallet_provider)

MCP (Claude Desktop / Claude Code / Cursor) — `GA`

Zero install via the hosted endpoint:

{
  "mcpServers": {
    "floe": {
      "url": "https://mcp.floelabs.xyz/mcp",
      "transport": "streamable-http"
    }
  }
}

Or run a local MCP server — see floe-mcp-server.

CrewAI (`Beta`)

CrewAI agents consume the Floe stack via MCP today. A runnable crew is available in floe-examples/crewai-demo.

Configuring the x402 facilitator key

The x402 and agent-awareness actions require a facilitator API key. The SDK does not read any env var directly — pass the key through X402Config (or the equivalent option on floe_action_provider):

from floe_agentkit_actions import floe_action_provider
from floe_agentkit_actions.x402 import X402Config
import os

provider = floe_action_provider(
    x402_config=X402Config(facilitator_api_key=os.environ["FLOE_FACILITATOR_API_KEY"]),
)

A common convention is to store the value in a FLOE_FACILITATOR_API_KEY env var in your app and forward it to the SDK as shown.

Environment Variables

These are read by the CLI and the examples in this repo. The SDK itself does not read env vars — see Configuring the x402 facilitator key above for facilitator auth.

Variable	Used by	Description
`PRIVATE_KEY`	CLI / examples	Wallet private key (0x...)
`CDP_API_KEY_NAME`	CLI	Coinbase CDP API key name
`CDP_API_KEY_PRIVATE_KEY`	CLI	Coinbase CDP API secret
`OPENAI_API_KEY`	CLI	OpenAI key for the conversational agent
`ANTHROPIC_API_KEY`	CLI	Anthropic key for the conversational agent
`BASE_RPC_URL`	CLI / examples	Custom Base RPC (recommended)
`FLOE_FACILITATOR_API_KEY`	Your app	Convention only — forward to `X402Config(facilitator_api_key=...)`

Contract Addresses (Base Mainnet)

Contract	Address
LendingIntentMatcher	`0x17946cD3e180f82e632805e5549EC913330Bb175`
LendingViews	`0x9101027166bE205105a9E0c68d6F14f21f6c5003`
PriceOracle	`0xEA058a06b54dce078567f9aa4dBBE82a100210Cc`
x402 Facilitator	`0x58EDdE022FFDAD3Fb0Fb0E7D51eb05AaF66a31f1`
Aerodrome SwapRouter	`0xBE6D8f0d05cC4be24d5167a3eF062215bE6D18a5`
Aerodrome QuoterV2	`0x254cF9E1E6e233aa1AC962CB9B05b2cFeAAe15b0`
WETH	`0x4200000000000000000000000000000000000006`

Development

pip install -e ".[dev]"
pytest

Links

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 94 Commits
.github/workflows		.github/workflows
crewai-floe		crewai-floe
examples		examples
src/floe_agentkit_actions		src/floe_agentkit_actions
tests		tests
.gitignore		.gitignore
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
pyproject.toml		pyproject.toml

Folders and files

Latest commit

History

Repository files navigation

floe-agentkit-actions

What it does

Framework support

Install

30-second example

Quick Start

As an AgentKit Provider

Standalone Usage

Actions (47 total)

Read Actions (8)

Write Actions (7)

Flash Loan Actions (5)

Credit Facility Actions (7)

Deploy / Verify / Readiness Actions (3)

x402 Credit Delegation Actions (8)

Agent Awareness Actions (9)

CLI

Wallet Providers

Framework Integrations

LangChain (GA)

OpenAI Agents / function calling (Beta)

MCP (Claude Desktop / Claude Code / Cursor) — GA

CrewAI (Beta)

Configuring the x402 facilitator key

Environment Variables

Contract Addresses (Base Mainnet)

Development

Links

License

About

Topics

Resources

License

Code of conduct

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

LangChain (`GA`)

OpenAI Agents / function calling (`Beta`)

MCP (Claude Desktop / Claude Code / Cursor) — `GA`

CrewAI (`Beta`)

Packages