Gemini Trading Bot in Python

This guide shows how to build a Gemini trading bot in Python with LumiBot, for both AI agent trading and algorithmic trading. LumiBot connects to Gemini through CCXT, the open source library that talks to more than 100 crypto exchanges. You will create a Gemini account and API key, connect it, and run two examples: a simple algorithmic trading bot and an agentic trading bot where an AI agent makes the calls.

Gemini at a glance

CCXT exchange id

gemini

Markets

Spot, Perpetual swaps

Testnet (sandbox)

Yes

Credentials

API key, API secret

Live data streams

order book, trades, candles, your orders

CCXT request spacing

about 100 ms between requests

Official site

https://gemini.com/

API docs

https://docs.gemini.com/rest-api

Fees

https://gemini.com/api-fee-schedule

What is Gemini?

Gemini is a crypto exchange that offers spot crypto trading, perpetuals, margin and staking, with REST, WebSocket and FIX APIs.

Sign up for Gemini

  1. Sign up for a Gemini account at https://exchange.gemini.com/.

  2. Verify your identity by uploading a clear photo ID. Gemini only lets fully verified accounts create API keys.

  3. For practice, also register a separate account on the Gemini demo environment at https://exchange.sandbox.gemini.com. Gemini fills new demo accounts with test balances, and demo keys only work in the demo environment.

Open a Gemini account

Create a Gemini API key

  1. Log in and open the API settings page (https://exchange.gemini.com/settings/api).

  2. Click Create a new API Key.

  3. Choose the scope: the single account the bot should trade in (most people have one account called Primary). Avoid the Master scope, which reaches every account you have.

  4. Choose the Trading role (see below).

  5. Under IP Allow List pick Trusted IPs Only and add the IP address of the machine that runs your bot. Gemini recommends this for every key.

  6. Copy the API key and API secret and keep the secret private, like a password.

Which permissions to turn on: Use the Trading role (Gemini also calls it Trader). It can check balances, place and cancel orders and view orders and trades. Do not add Fund Management, which can withdraw crypto, and leave the Master scope and Administrator role alone. A bot never needs to move money out of your account.

Warning

Never turn on withdrawals for a Gemini trading bot key. Bind the key to your server’s IP address when Gemini lets you, keep the secret out of your code and out of Git, and delete the key if it ever leaks.

Connect Gemini to LumiBot

The simplest way: put these lines in a .env file next to your strategy. LumiBot loads the file, sees TRADING_BROKER=gemini and connects to Gemini for you, so your strategy needs no broker code.

TRADING_BROKER=gemini
GEMINI_API_KEY=your_apiKey
GEMINI_API_SECRET=your_secret
GEMINI_SANDBOX=true   # testnet; remove this line for your live account

LumiBot reads exactly the fields CCXT needs for Gemini (apiKey, secret). If one is missing, LumiBot stops at start-up and names the variable you still need.

You can also build the broker in code. This is handy when one script talks to more than one exchange, or when you need extra CCXT options. The keys in the config are CCXT’s own credential names for Gemini:

import os

GEMINI_CONFIG = {
    "exchange_id": "gemini",
    "apiKey": os.environ.get("GEMINI_API_KEY"),
    "secret": os.environ.get("GEMINI_API_SECRET"),
    "sandbox": True,  # testnet first; set False for your live account
}

from lumibot.brokers import Ccxt

broker = Ccxt(GEMINI_CONFIG)
# then: MyStrategy(broker=broker, ...)

Either way, LumiBot turns on CCXT’s built-in rate limiting for you.

Practice on the Gemini testnet

Gemini runs a full demo environment (sandbox) at https://exchange.sandbox.gemini.com with test funds and simulated order book activity. Make a separate account and API key there. Set GEMINI_SANDBOX=true (or "sandbox": True in a code config) to use it.

CCXT sends sandbox requests to: https://api.sandbox.gemini.com, https://docs.gemini.com. Testnet and live accounts use different keys.

Algorithmic trading example: a simple Gemini bot

This bot buys about 50 USD of BTC on Gemini once and then holds it. It is a starting point to check that your keys, balances and orders work. If the exchange’s minimum order is larger, raise the amount. Gemini accepts only limit orders through CCXT, so the example places a limit order just above the last price.

from lumibot.entities import Asset
from lumibot.strategies import Strategy


class GeminiBuyOnce(Strategy):
    """Buy a small amount of BTC once, then hold it."""

    def initialize(self):
        self.sleeptime = "1H"
        self.set_market("24/7")  # crypto trades all day, every day

    def on_trading_iteration(self):
        coin = Asset("BTC", asset_type=Asset.AssetType.CRYPTO)
        if self.get_position(coin) is not None:
            return  # already holding BTC
        price = self.get_last_price(coin, quote=self.quote_asset)
        if price:
            quantity = round(50 / price, 6)  # about 50 USD worth
            # This exchange takes only limit orders through CCXT.
            order = self.create_order(coin, quantity, "buy", limit_price=price * 1.002, quote=self.quote_asset)
            self.submit_order(order)


if __name__ == "__main__":
    # LumiBot reads TRADING_BROKER and the keys from your .env file.
    strategy = GeminiBuyOnce(quote_asset=Asset("USD", asset_type=Asset.AssetType.CRYPTO))
    strategy.run_live()

Agentic trading example: an AI agent that trades on Gemini

Agentic trading means an AI agent reads the market and decides what to do. In this example one LumiBot agent looks at prices and your Gemini account, then places the orders itself. There is no Python logic in between: you set the rules in plain English in the prompt, such as what it may trade and how big each order can be. This follows the same pattern as LumiBot’s AI trading team examples.

from lumibot.entities import Asset
from lumibot.strategies import Strategy


class GeminiAITrader(Strategy):
    """One AI agent researches the market and places the trades itself."""

    def initialize(self):
        self.sleeptime = "4H"
        self.set_market("24/7")
        self.agents.create(
            name="trader",
            model="openai/gpt-6-luna",
            reasoning_effort="high",
            allow_trading=True,
            system_prompt=(
                "You manage a small crypto account on Gemini. "
                "You may trade BTC and ETH against USD (asset_type 'crypto', quote_symbol 'USD'), or hold cash. "
                "Before every order, call account_portfolio, account_positions and "
                "market_last_price for that exact symbol in the same run. "
                "Keep each order under 10% of the account value. "
                "Explain each decision in one short paragraph."
            ),
        )

    def on_trading_iteration(self):
        self.agents["trader"].run(
            task_prompt="Review recent BTC/USD and ETH/USD prices and your current holdings, then decide whether to buy, sell or hold.",
            context={"exchange": "Gemini", "quote": "USD", "date": self.get_datetime().isoformat()},
        )


if __name__ == "__main__":
    # LumiBot reads TRADING_BROKER and the keys from your .env file.
    strategy = GeminiAITrader(quote_asset=Asset("USD", asset_type=Asset.AssetType.CRYPTO))
    strategy.run_live()

The agent needs a model key such as OPENAI_API_KEY in your environment. See the AI agents quickstart for models, tools and memory.

What you can trade on Gemini through CCXT

Market

Listed by CCXT

Spot

Yes

Margin

Not listed by CCXT

Perpetual swaps

Yes

Dated futures

No

Options

No

CCXT lists these order features for Gemini: limit orders, editing open orders, canceling orders.

For perpetual contracts, use the full CCXT symbol so the settlement currency is clear, for example BTC/USDT:USDT. See the CCXT guide for contract quantities, position modes and leverage.

Gemini tips

  • A Gemini API key that is not used for 180 days expires and stops working. If a bot sits idle for months, plan to make a new key.

  • Give each bot its own key. Gemini says each application should have its own key so permissions stay separate.

  • Spot is plain BTC/USD. Gemini perpetuals trade in a separate derivatives account and appear in CCXT as BTC/GUSD:GUSD; you move funds there from your exchange account.

  • When you create a key, Gemini lets you pick a nonce mode. Time-based nonces need your computer clock within 30 seconds of Gemini’s server time, so keep the clock synced.

Frequently asked questions

Can I build a Gemini trading bot in Python?

Yes. Install LumiBot (pip install lumibot), set your Gemini keys and use the gemini CCXT exchange id as shown above.

Does Gemini have a testnet?

Yes. Set GEMINI_SANDBOX=true and use keys from the Gemini testnet.

Which API permissions does a Gemini bot need?

Use the Trading role (Gemini also calls it Trader). It can check balances, place and cancel orders and view orders and trades. Do not add Fund Management, which can withdraw crypto, and leave the Master scope and Administrator role alone. A bot never needs to move money out of your account.

Can an AI agent trade on Gemini by itself?

Yes. The agentic example above gives one agent permission to trade. You control what it may trade and how much through its prompt.

Can I trade Gemini perpetual futures with LumiBot?

CCXT lists perpetual swaps for this exchange. Use the full contract symbol, such as BTC/USDT:USDT.

Is LumiBot free?

Yes. LumiBot is open source on GitHub at https://github.com/Lumiwealth/lumibot. You pay only your exchange’s trading fees and, for AI agents, your model provider.

Run it in the cloud with BotSpot

Do not want to keep a computer running for your Gemini bot? BotSpot is the hosted home for LumiBot strategies and AI trading agents: build with AI, backtest, connect a broker, and watch every trade with logs and alerts.

Try BotSpot

Sources