A framework you run, or a desk you connect to
Both quote both sides of a book. The difference is what you are taking on. Hummingbot hands you the engine and the wiring. Vesper hands you a mode that has already been calibrated on a handful of venues. Which is right depends on a question about you, not about the software.
Hummingbotopen framework you run
Vesperhosted desk you connectWhat each one is
Hummingbot describes itself on its own site as “an open source Python framework that helps you run agentic trading strategies on any CEX and DEX” and as “the open source, community-owned framework for crypto market makers.” The code is released under the Apache 2.0 licence, the site lists 318 connectors, and installation runs either on your own machine or on a cloud server you rent. To add a connector or change a strategy you install from source. (Checked on hummingbot.org, 4 September 2026.)
Vesper is not a framework. You connect an exchange account through an API key that can place and cancel orders but cannot withdraw, pick a venue and a mode, and the desk quotes from your account. There is nothing to install, and there is also nothing to modify: the strategy is ours, not yours.
That single sentence contains both the argument for us and the argument against us, and it is worth sitting with before reading any feature table.
- Hummingbot: you own and operate the machine, the code and the outcome.
- Vesper: you own the account and the funds; we operate the quoting.
- Neither takes custody. In both cases the money stays on the exchange, in your account.
Where Hummingbot is the better answer
If you write Python and want to own the strategy, this comparison is already over. An open framework lets you read every line that touches your money, change what you disagree with, and keep running it if the people who wrote it disappear. No hosted product can offer that, and it is not a small thing.
Breadth is the second reason. 318 connectors spanning centralised and decentralised venues is a different order of coverage from ours. If the book you care about is on a large centralised exchange, we very likely do not quote it and are not going to.
The third is cost structure. The framework is free and open; what you pay is a server and your own time. Whether that is cheaper depends entirely on how you value the time.
- You want to read or change the strategy code → use the open framework.
- Your venue is a large CEX → the connector list decides it.
- You are comfortable running and maintaining infrastructure → that is the trade you are making.
Where we are different, and why it is narrow on purpose
We quote a short list of perpetual venues (Ondo, Arcus, RISEx, Lighter, Phoenix, Perpl and HIP-3 builder markets), and the list is short because each one had to be measured before it was added. We record the order book on these venues continuously: over four million snapshots at a 5.5-second cadence and still counting, taken only while the market is open.
That recording is what a calibrated mode is made of. It is also why our market pages can tell you what a $1,000 order really costs against the mid on a given book, and which hour of the day is the cheapest to trade it: numbers that come from our own measurement, not from the venue's API.
A framework can do all of this too. It just will not do it for you on day one; you will measure those venues yourself, which is exactly the work you are choosing to take on.
- Fewer venues, each one measured before it was offered.
- Per-venue calibration derived from our own book recording, not from defaults.
- You cannot read or modify that calibration. If that matters to you, it is a real objection.
The arithmetic that decides more than the software does
Whichever you pick, the same number decides whether market making works on your chosen book: the value of one tick in basis points. A tick is a fixed price increment, so what it is worth depends on the price level.
On a gold perpetual near $4,450 with a one-cent tick, one basis point is 46 ticks, so a single tick costs about 0.02 bp, and stepping inside the touch to buy queue priority is nearly free. On the silver book alongside it, the same one-tick improvement is worth 1.47 bp: sixty-eight times more, and it eats most of the spread you were quoting for.
Same exchange, same code, opposite verdict. No framework and no hosted desk changes that arithmetic. It takes thirty seconds to compute and it should be the first thing you do, before you compare a single feature.
- Tick value in bp = tick size ÷ price × 10,000.
- Below roughly 0.1 bp: queue priority is cheap and the standard toolkit applies.
- Above roughly 1 bp: stepping inside costs most of your spread. Reconsider the market.
| Hummingbot | Vesper | |
|---|---|---|
| What it is | An open-source Python framework you install | A hosted desk you connect an exchange account to |
| Where it runs | Your machine or a server you rent | Our infrastructure; your funds stay on the exchange |
| Licence / source | Apache 2.0, code is public and modifiable | Closed. You cannot read or change the strategy |
| Venue coverage | 318 connectors listed, CEX and DEX | A short list of perpetual venues, each measured first |
| Who holds the funds | Your exchange account | Your exchange account |
| Key permissions | Set by you when you create the key | Trade and cancel only: the key cannot withdraw |
| Setup | Install, configure, maintain | Connect an account, pick a venue and a mode |
| Best for | Someone who wants to own the strategy | Someone who wants a calibrated mode on a venue we measured |
How to decide in two minutes
Answer one question honestly: do you want to own the strategy, or do you want a result on a specific book? There is no wrong answer, and the two products are not really competing for the same person.
If you want to own it, take the framework. You will spend the first weeks on infrastructure and measurement rather than on quoting, and that is the price of control, and a fair one.
If you want a result on one of the venues we quote, the calibration is already done and you can see the numbers behind it on our market pages before you connect anything.
- Want to read and change every line → the open framework.
- Your book is a large centralised exchange → the connector list decides it.
- Your book is one we measure, and you want the calibration done → talk to us.
- Either way: compute the tick value in basis points first.
Which one is for you?
- You want to read and change the strategy: take Hummingbot. Nothing hosted can match that, and we do not try.
- Your book is on a large centralised exchange: the 318 connectors decide it. We almost certainly do not quote it.
- You want a result on a venue we measured, without the first weeks of infrastructure: that is what Vesper is for. Compute the tick value in basis points first, either way.
Questions
Is Vesper open source like Hummingbot?
No. Hummingbot's code is public under the Apache 2.0 licence and you can read and modify it. Vesper's strategy is not public. If being able to audit or change the code is a requirement, that is a genuine reason to choose the open framework.
Does either one take custody of my funds?
Neither. In both cases the money stays in your own account at the exchange. With Vesper the connection uses a key that can place and cancel orders but cannot withdraw; with Hummingbot you set the key permissions yourself when you create it.
Which supports more exchanges?
Hummingbot, by a wide margin. Its site lists 318 connectors across centralised and decentralised venues. We quote a short list of perpetual venues and add one only after measuring it, which means most books are not on our list and will not be.
Do I need to know Python to use Hummingbot?
Its own documentation says beginners can start with the CLI trading client, while adding or customising connectors and extending strategies requires installing from source. So running a packaged strategy does not require it; changing one does.
What is the first thing to check before picking either?
The value of one tick on your chosen market, in basis points: tick size divided by price, times 10,000. Below about 0.1 bp, buying queue priority is cheap. Above about 1 bp it costs most of your spread, and no software fixes that.