Header-only C++ market data and execution for 31 venue ids, on a Bloomberg-shaped API.
Self-hosted crypto trading and market data tools
Where the process runs, not what the licence says — and why those are two different lists that happen to overlap.
Last updated
Where the process runs, and nothing else
This page answers one question: can you install it and run it yourself? Not what it costs, not what licence it carries, not whether anyone else can see what it does. Just where the process lives.
That is a narrower claim than it sounds, and it is worth separating from the one next door. Open source is about the licence on the code. This list is about the deployment. The two overlap heavily — nearly everything open source here is also runnable on your own machine — but they are not the same list, and the places they come apart are the most useful things on this page.
The twelve where the code is yours to run and not to read
These are the cards that make this a different list rather than a second door onto the same one. Each ships something you install on your own hardware under terms that are not open source.
Proprietary libraries you compile into your product. TradingView Advanced Charts is free to use and explicitly not open source — the repository is granted on application, the attribution has to stay visible, and the terms bar putting any part of it in a public repository. Highcharts Stock and ChartIQ are commercial licences, bought per developer seat and per year, that you then host yourself. DXcharts is the paid half of an open core — its free Lite edition is genuinely open, and the product with the indicators and the drawing tools is proprietary and quoted from 499 dollars a month. In all four cases the code sits inside your build and outside your control.
Source-available, which reads as open until you check. vectorbt is Apache-2.0 with a Commons Clause — you may read it, run it and modify it, and you may not sell it. That one added condition takes it out of the open-source definition and off the other page, while changing nothing about your ability to install it.
Its paid edition, vectorbt PRO, goes a step further: the licence is proprietary and the source is available to members, which means you install it from a private repository you keep access to by continuing to pay. Run it wherever you like; read it for as long as the subscription lasts.
A licence that can be withdrawn. dYdX v4 Clients carries a bespoke grant whose text reads like MIT but is conditioned on complying with the vendor's terms of use, with rights terminating automatically if you do not. The repository's own badge says AGPL and one package manifest says AGPL-3.0; only the Rust crate labels it honestly. Install it wherever you like — just do not call it open source on the strength of the badge.
Paid applications rather than libraries. Gunbot is the clearest case: a signed binary you install on your own machine. New buyers have paid for it by the month since October 2026, 29 euros for Gunbot Unlimited; the older lifetime licences, now sold only as upgrades to people who hold one, are registered on a blockchain against an ERC-20 wallet. You own the process, the server and the key file, and the source is not published. ProfitTrailer is the same shape: a closed Java application you run on your own Windows, macOS or Linux machine, or a Raspberry Pi. HaasOnline sells it as one edition of two — TradeServer is a licence you install on a machine you control, one active instance per licence, beside a cloud edition the vendor hosts. Gainium publishes its self-hosted edition's code, but under the Fair Core License, which bars a competing bot service, so it belongs with vectorbt above rather than on the open-source page. And tread.fi offers its execution engine in your own cloud with Docker, by request and at no published price. Everything else in this section is something you compile into your own product; these are the thing itself.
The one going the other way
L2BEAT is open source and is not on this list: the code is public and the product is a site somebody else runs. That is the shape most of the on-chain and market-data half of the catalogue takes, and it is why this page is dominated by engines rather than by data.
What self-hosting actually buys
Nobody else holds your keys. For anything that connects to an exchange account with trade permission, this is the whole argument. The key sits in a file on a machine you administer instead of in a vendor's database. That moves the risk rather than removing it — the security of that machine is now part of your trading system — but it is your risk to manage.
Nobody can change the terms under you. A hosted product's free tier can be retired, its price can move and its API can start requiring a key; six pages in what to use instead exist because exactly that happened. An installed binary keeps working on the terms you installed it under. What it cannot keep working without is its upstream data.
The feed is still somebody else's. This is the honest limit of the whole page. You can run the backtester, the bot and the chart yourself, and every price they draw still arrives over the network from a venue or an aggregator, on that vendor's terms and subject to that vendor's outage. Self-hosting buys independence from the application layer and none at all from the data layer.
What you take on
The machine, its uptime and its patches; the key file; and, for anything that trades, the fact that it has to keep running while you are asleep, which is a server rather than a laptop. The hosted versions of several products on this page exist precisely because that is real work, and paying somebody to do it is a legitimate answer.
All 32 of them
Showing 32 of 32
Sort by
Crypto Trading Bots14
One MIT client for 104 crypto exchanges, 76 of them over websocket.
Four language clients for one perp DEX, under a licence that is not the one on the badge.
Write the strategy in Python, backtest it, then run the same class live.
DCA, grid and combo bots priced in bot credits, with a free tier and its source published.
A Go trading engine for 22 exchanges that has never cut a release or a tag.
A closed-source bot you run on your own machine, sold to new buyers by the month.
A Lua-scripted bot engine sold as a hosted cloud or as a server you run yourself.
Apache-2.0 Python framework for running market-making bots on CEXs and DEXs.
The official Python client for signing and sending orders on one perp DEX.
GPL-3.0 Python bot with a web interface for grid, DCA, basket and TradingView strategies.
A contrarian perp market maker you configure in JSON and leave running for months.
A closed-source Java bot you run yourself — trailing buys, DCA and 40-plus indicators.
A visual, node-based bot workshop you install from GitHub. Last stable release 2024.
Backtesting & Research Libraries9
Two thousand lines, one OHLC series, and no idea what a funding payment is.
Rust trading engine and market-data streams — live execution is a trait you implement.
Tick-by-tick backtesting that models order queue position and feed and order latency.
Python backtesting and research framework whose live-trading half is a paid plugin.
The engine behind QuantConnect, Apache-2.0 and runnable on your own machine.
One Python strategy that backtests and then trades live, across eight CCXT exchanges.
Event-driven Rust engine that settles perpetual funding at the venue boundary.
Vectorised Python backtesting that has never heard of funding or liquidation.
The paid VectorBT — leverage and limit orders, and still no funding leg.
Charting Libraries for Crypto Products7
Heavyweight broker charting SDK, domain-locked, priced by an account manager.
The paid half of the open core — 100+ indicators, 40+ drawing tools, from $499 a month.
The MPL-2.0 renderer under a broker charting product — candles and axes, no indicators.
Per-seat commercial charting whose free licence excludes anything you sell.
Zero-dependency candle engine with no market-hours rules to switch off for 24/7.
The chart engine inside half this market — free with visible attribution, not open source.
Apache-2.0 canvas renderer for candles and lines. No indicators, no drawing tools.
Crypto Charting Platforms & Screeners1
Live multi-exchange tape and liquidations in one chart — GPL-3.0, and yours to host.
Crypto Trading Terminals1
Execution algorithms and a market-making bot over your own CEX keys and perp-DEX wallets.
FAQ
Is self-hosted the same as open source?
No, and this page exists because the difference is easy to miss. Open source is a fact about the licence on the code; self-hosted is a fact about where the process runs. Most of this list is both, but twelve products ship a binary or a source-available licence you install on your own machine under commercial terms — you run it, you do not get to read or fork it. One card in the catalogue is the opposite — open code that only exists as somebody else's hosted service.
What do I actually take on by self-hosting?
The machine, the uptime, the patches, and the exchange API keys in a file on disk. For anything that trades, that machine has to stay up without you, which usually means a server rather than a laptop. In exchange nobody else holds your keys and nobody can change the terms under you. It is a trade, not an upgrade.
Does self-hosted mean free?
Often, but not on this page by definition. Several cards here are paid licences you install yourself — priced per developer seat, per year or perpetually — and one is free to install while charging for the data it runs on. Where the money is, and what it buys, is on each card.
Can I self-host a data provider?
Almost never, and that is the shape of this list. What you can run yourself is the engine — a backtester, a bot, a charting library, a client. The prices, order books and on-chain tables come over the network from somebody who is not you, on their terms. A self-hosted stack is only as independent as its upstream feed.