Self-hosted bots

What self-hosting buys - and what it bills

Bot	Type	Price	Open source	Backtest	Paper	Custom code	Exchanges	US	Founded
Freqtrade	Self-hosted	free	yes	yes	yes	yes	11	US	2017
Hummingbot	Self-hosted	free	yes	yes	yes	yes	50	US	2018
Gunbot	Self-hosted	$29/mo	-	yes	yes	yes	20	US	-
Jesse	Self-hosted	$0-1599/mo	yes	yes	yes	yes	12	US	2018
OctoBot	Self-hosted	$0-29.99/mo	yes	yes	yes	yes	12	US	2018
Superalgos	Self-hosted	free	yes	yes	yes	yes	15	US	2019
Passivbot	Self-hosted	free	yes	yes	yes	yes	8	US	2020

Live values from our database - details and caveats in each review. "-" = not documented.

Self-hosted is the custody ceiling: your keys are generated, stored and used on infrastructure you control[3]. No vendor vault exists to leak - the 3Commas 2022 admission ("DELETE YOUR KEYS NOW")[6] is the incident class this lane deletes by construction. The bill arrives as ops: the figure below is what keeps a bot alive after the install, and none of it is optional.

Two framing numbers before the detail: 7 frameworks currently pass our self-hosted filter, and their median entry price is ~$29/mo/mo - but the median is misleading in both directions, because the strongest frameworks here are free and the closest substitutes are paid. Judge by the ops bill and the feature fit, not the sticker.

The cost comparison against cloud is simple to model and easy to get wrong. Our cost model (subscription + VPS only, trading fees equal across models): exchange-native $0, Freqtrade + $10/mo VPS $120/yr, paid framework + VPS ~$468/yr, cloud subscriptions $180-1,260/yr[5]. The bot cost calculator prices your exact case. What the model cannot price: your hours. One weekend to stand up, then the weekly cadence above - roughly a hobby's worth of attention.

The frameworks, honestly compared

Freqtrade is the default for a reason: GPL-3.0, 55.6k stars, a community that documents everything, backtesting and dry-run as first-class citizens, and a stated minimum of 2GB RAM / 1GB disk / 2 vCPU[1]. Strategy development is Python. Its README carries the most honest sentence in the category: "Do not risk money which you are afraid to lose"[1].

Hummingbot is the market-making specialist: Apache-2.0, 318 exchange connectors, $47B+ reported volume across 111k+ instances[2]. If your strategies are maker/spread-shaped across many venues, it has no peer; for simple grid/DCA it is heavier than you need.

Gunbot is the paid turnkey: $29/mo, 20 exchanges, futures, built-in strategies, desktop app - you are buying the layer Freqtrade makes you assemble. No key ever leaves your machine. The lifetime licenses are gone; new buyers get the subscription.

OctoBot, Passivbot, Superalgos, Jesse round out the lane: a web-UI open-source core, a futures-native parameter engine, a node-based strategy editor, and a research-grade Python framework respectively. Each trades generality for focus; all keep keys local.

Underneath most of them sits the same library layer: CCXT, the unified trading API across 100+ exchanges at 44.4k stars[8]. Its centrality is good news for longevity - exchange API changes get patched in one place, and the frameworks inherit the fixes - and a reminder that self-hosting is an ecosystem, not a product.

Statistics that frame the lane

• 7 bots run self-hosted in our tracking; 6 of them are fully open source - Gunbot is the closed-source exception.
• Median entry price across this lane: ~$29/mo/mo - dominated by Gunbot's subscription, while 6 of the frameworks cost nothing.
• 1 offer a trial window (up to 7 days) - rare in this lane, because software you can already read needs no trial.
• The oldest operator still listed dates to 2017 - self-hosted tooling predates the cloud platforms' marketing era.
• Across the whole tracked universe, 2 platforms have gone under - vendor longevity is a real risk in this market, and the one risk self-hosting largely removes.

Tips that save the first month

• Dry-run for a full week before keys exist. Freqtrade's dry-run wallet[3] and Hummingbot's paper_trade simulate live markets. If the framework cannot outlast a week of your attention in paper mode, real keys will not fix it.
• Buy the static IP. A $5-10 VPS with fixed addressing keeps API sessions stable; laptop sleep and rotating home IPs are the top two "bot stopped" causes. OVH's entry tier (2 vCores, 4GB RAM, backups included, $4.54/mo)[4] exceeds Freqtrade's minimum[1].
• Treat config.json as the balance it controls. Permissions 600, encrypted disk or encrypted backup, never in git. The API key security guide applies to self-hosted keys in full - the custody is yours, so are the mistakes.
• Pin versions and read changelogs. Breaking exchange API changes land without your consent; update on your schedule, in a maintenance window, after a backup.
• Alert on bot death, not just on trades. A dead bot with open positions is worse than a stopped strategy. A 2-line systemd unit with Restart=always plus a heartbeat message to your phone covers the failure mode that actually loses money.
• Size for the exchange's fee schedule, not the software's price. Binance's 0.1% (0.075% with BNB)[7] applies identically to a $29 framework and a $105 cloud plan[5] - venue choice moves more money than hosting choice.

Warnings

• A compromised VPS leaks the keys it holds. Self-hosting removes the vendor from the trust chain, not the endpoint. 2FA on the VPS, firewall on, keys scoped trade-only and IP-bound at the exchange - the API key security guide is the checklist.
• No vendor to sue. Bugs in your config, your strategy or your update window are your losses. That is the deal; the framework licenses say it in so many words[1][2].
• Marketplace strategies are not due diligence. Community-shared configs and strategies backtest beautifully until they do not. Overfitting is the default outcome of copying parameters; validate on regimes you did not download from.
• Uptime is a strategy parameter. Every framework here assumes it can reach the exchange when it needs to. If your infrastructure cannot promise that, the cloud lane - with its vendor ops - is the better fit despite the subscription[5].
• Futures self-hosting multiplies ops stakes. Liquidation risk plus ops risk together: Passivbot's futures presets are powerful and unforgiving. Understand the liquidation math before automating it.

A realistic first month

A workable sequence for the route most people take (Freqtrade, single exchange): day one, stand up the VPS, install the framework, run the sample config in dry-run[3]. Week one, backtest a downloaded strategy across three regimes, then dry-run it live and go on holiday from checking it - the point is to see what it does unattended. Week two, wire alerts (bot heartbeat, fill notifications), pin the version, take a config backup, then connect a trade-only, IP-bound key with the smallest balance you take seriously[6].

The decision point arrives around week three: either the framework disappeared into your routine - in which case the lane fits and scaling size is your only real question - or maintenance is already slipping, in which case the honest move is back to the cloud lane or the exchange-native lane[5], where someone else's subscription buys the uptime you cannot provide. Self-hosting is a fit decision, not a superiority ranking; the custody ceiling only matters if the ops floor is actually covered.

One last framing on hardware: this lane scales down beautifully. A 2GB-RAM VPS[1] runs a serious multi-pair Freqtrade setup; the same stack scales up to dedicated boxes for Hummingbot market-making fleets. The constraint is never CPU - it is your ability to observe what the machine is doing with real funds. Choose hardware you can monitor, not hardware you can brag about. Self-hosting remains a fit decision first and an ideology second; this page exists so that decision gets made with numbers instead of forum anecdotes.