Hummingbot Explained

What Hummingbot is

Hummingbot is open source software for building and running trading bots, with a focus on strategies such as market making and arbitrage across exchanges. Open source means the code is public, so anyone can read it, run it and change it. It is a framework: it connects to exchange interfaces, keeps track of orders and balances, and lets you choose or write a strategy.

We have not installed it, run it, or read its current documentation for this page. Which exchanges it connects to, which strategies come with it, how it is installed and how it is licensed are facts to read in its own current documentation. Nothing on this page describes a feature of a particular version.

What market making means

Most orders in a market wait in a list. People who place waiting orders to buy below the price and sell above it are called market makers, because they supply the orders that others trade against. The gap between the best buy and the best sell is the spread. A market maker bot quotes a buy at a bid price and a sell at an ask price around a mid price, and when both fill, it keeps the gap. It must also cancel and replace its quotes as the mid moves, which is most of the work of the bot.

Which side fills first is not up to the bot. If the price drifts down, the buy fills and the sell does not, and the bot is left holding the asset as it falls. This is inventory risk.

The settings a market making bot needs

• Spread. How far the buy and sell sit from the mid. Wider means more per trip and fewer fills.
• Order size. How much each quote offers. Larger means more exposure when only one side fills.
• Refresh time. How long a quote lives before it is cancelled and re-placed. A fast refresh keeps up with the price and sends many requests. See Binance API for the request arithmetic.
• Inventory limits. How much of the asset the bot may hold before it stops quoting one side or skews its prices.
• A stop. What the bot does in a fast one-way move.

Worked example: spread against fees

The numbers are made up. The mid price is 100. The bot buys at the bid and sells at the ask, and pays a fee of 0.1 percent of the order value on each side, so 0.2 percent a round trip. Each row sets a different total spread, measured from the bid to the ask as a share of the mid.

Calculated by us on 31 Oct 2026, exact for the stated inputs. At a spread of 0.1 percent the fees are larger than the gain, so each completed trip loses. The quoted spread must clear the round trip fee before anything is earned. The fee rate here is invented, and real rates differ by venue and account, so use your own.

Worked example: inventory risk

Take the same bot. It buys at the bid. Before its sell fills, the mid price falls 1 percent, and the bot re-quotes its sell around the new mid. It completes the trip at the new ask, which is lower than the old one.

Calculated by us on 31 Oct 2026, exact. Even with a spread of 0.8 percent the trip loses 0.40 percent once the mid has fallen 1 percent. A bot that earns a fraction of a percent per trip can give back many trips of gain in one move. If the bot does not sell at a loss, it keeps holding, which is the same risk with the loss not yet counted.

How a market making bot breaks

• One-way markets. A steady trend fills one side again and again, and the bot builds a large position against the move.
• Stale quotes. If a network or data problem delays a cancel, a quote can sit at an old price while the market has moved, and fill on bad terms.
• Fees larger than the spread. A spread set too tight loses on each completed trip.
• Thin books. Where there are few other orders, the bot is the market, and its fills are large against what the market can absorb.
• Software risk. Open source code can have bugs, and a bot that is left running unwatched can do the wrong thing at speed.

How to check a setup

1. Write down the round trip fee for your account and compare it with the spread you quote.
2. Work out the loss on the largest position the bot may hold if the price falls the width of a normal bad day.
3. Run on a practice environment first, and on paper trading on real prices.
4. Start with a very small size, compare each fill with the quote, and check the log every day.
5. Read the code you run. See open source crypto trading bot.

What this page does not tell you

It does not say how Hummingbot behaves, what it costs, or whether market making earns money. The spreads, fees and the one percent fall are made up. For a related bot type that also rests orders at set prices, see market maker and grid trading bot.