Usage of this software in its current form will likely result in loss of funds.
For all intents and purposes, this software is to be considered an educational example.
See LICENSE for more clarification. And do read this README carefully.
At this point any feedback is really appreciated – do get in touch with me in a way you see fit.
© Stan
This software is capable of fully autonomous trading on popular cryptocurrency exchanges, once set up correctly.
Currently, it utilizes several naive approaches to spot trading ( either price-related crossovers or GPT-powered analytics is an option); however, due to it being highly customizable, one could squeeze better results out of it.
This software is open source under a permissive License, and it's FREE.
Support for older Python versions is not guaranteed, but deemed probable for 3.10 and newer.
Console only or beautiful PlotlyDash GUI.
-
(NEW MODERN STYLES) Optional GUI powered by plotly/dash
-
Test mode to only test the prediction API – no trades would be made (suitable for LLM APIs; Pandas – not so much)
-
LLM connection is through an API (any 'openai' library supported APIs)
-
This script endlessly places buy and sell orders based on predictive modeling (with calculations of price and amount, see function
prepare_orderof classTradingBotin module trading_bot.py):a. Price for 'BUY' orders is
((bid + ask) / 2) * (1 - parametrized premium),b. Price for 'SELL' orders is
((bid + ask) / 2) * (1 + parametrized premium),c. Amount to buy is
parametrized reinvestment_rate x free quote token balance / price buyd. Amount to sell is
parametrized reinvestment_rate x free base token balance -
This software operates under the assumption that one can only use one's own capital
-
This software uses default Python floating point real numbers (
float) for prices & volumes -
This software is shipped with substantial error handling. The script is designed to run indefinitely.
To run this project, one would need to add the following environment variables to the .env file/-s (the currently
programmed logic is in two separate files for predictive and running modules, but it doesn't have to be).
All fields must be filled with valid strings. Fields that end with '_JSON' must be filled with valid json data strings.
.env files must be linked in the starting sequence (e.g., via command line arguments or as Default strings in the code).
It is easier to create a new main.env as per example, that has included explanations for each
variable, as well
EXCHANGE_API_KEY – API to access cryptoexchange
EXCHANGE_SECRET – API to access cryptoexchange
EXCHANGE_PASSPHRASE – API to access cryptoexchange
ALGORITHM_TRUST_PERCENTAGE – Reinvestment rate – how much of one's free balance (per token) is to be used for spot
orders (default is 0.5, can be a real value in range [0.0;1.0])
BASE_SLEEP_TIME – sleep time in seconds between program cycles
CANCEL_ORDER_LIMIT – how many cycles to wait before cancelling all open orders (cancels orders on achieving LIMIT)
RETRIES_BEFORE_SLEEP_LIMIT – how many times to retry without sleeping (only unknown errors)
DATA_VECTOR_LENGTH – Number of past-data points to use in predictive modeling
DEFAULT_EXCHANGE_NAME – ccxt supported exchange name
DEFAULT_EXCHANGE_FEE – price fraction that is collected as order fee by crypto exchange
PREMIUM_OVER_EXCHANGE_FEES – any positive real number (would be kept at 0.0, to only account for exchange fee without
additional premium/discount on bid/ask)
MIN_TRANSACTION_VALUE_IN_BASE – Minimal amount to spot-order base currency for a given trading pair
TIMEFRAME – Depending on cryptoexchange, this value can be set as either of: 1m, 3m, 5m, 10m, 15m, 30m,
4h, 8h, 12h, 1w. See ccxt docs or exchange API for correct info.
TRADING_PAIR – What to spot trade (e.g., XMR/USDT). NOTE: One must have both of these two tokens in any amount in
their portfolio, before an order can be placed (If that isn't the case, the script will show a
ProbablyAIButCouldBeAnything exception with the ticker of the token one doesn't own).
TRADING_BASE, TRADING_QUOTE – if a trading pair doesn't have a / sign, these are necessary (i.e., if TRADING_PAIR=XMRUSDT, then TRADING_BASE=XMR and TRADING_QUOTE=USDT MUST be supplied)
(easier to create a new llm.env or probability.env, or pandas.env as per example 1 or
example 2, or example 3)
DEFAULT_PREDICTION_API – either LLM or PROBABILITY_LLM, or PANDAS is supported
If DEFAULT_PREDICTION_API=PANDAS, then:
PREDICTION_OPERATIONAL_PRICE_TYPE – stockstats.StockDataFrame supported indicator or price type (e.g., middle_2_sma
or close). Numbers cannot exceed DATA_VECTOR_LENGTH.
PREDICTION_INDICATORS_JSON – valid json array of stockstats.StockDataFrame supported indicators (e.g.,
["close_5,15_kama","middle_15_ema"]). Numbers cannot exceed DATA_VECTOR_LENGTH.
PREDICTION_GLOBAL_SIGNAL_LAG – integer value of 1 or greater (cannot exceed DATA_VECTOR_LENGTH).
If DEFAULT_PREDICTION_API=LLM, then:
LLM_BASE_URL – base URL for API of your chosen LLM,
LLM_API_KEY – your LLM api key,
LLM_MODEL – model to use,
Also, if DEFAULT_PREDICTION_API=PROBABILITY_LLM, then additional:
LOWER_PROB – Sell signal, if lower than (default 20)
UPPER_PROB – Buy signal, if higher than (default 80).
Steps 3 and 4 are irrelevant, if Dockerfile is used
Running with a -d or --dashboard flag will result in forwarding program output to a local flask server of a Dash app (only supported for run mode).
This dashboard option will also show a plot of transaction cost for successfully placed orders (upward transaction cost trend coincides with portfolio estimation growth, so it's useful info).
Create .env file/-s:
Parametrization of this software is achieved via the means of environment variables, specifically through the use of
.env files. So the files are essential.
Link these .env files in Python module run.py either as command line arguments, or in the if __name__ == "__main__": section as
DEFAULT_PREDICTION_ENVIRONMENT_FILENAME and DEFAULT_MAIN_ENVIRONMENT_FILENAME constants. They don't have to be separate files, but
at least a filename must be supplied of a file containing the required Environment variables.
Not specifying a <something>.env file would result in scanning the literal file with the path '.env' in the same
directory as the script.
The required packages are installed with:
pip install --root-user-action=ignore --upgrade pip && pip install --root-user-action=ignore -r requirements.txt
Run from inside project directory:
Required:
Specify either run or test command to run the script in main mode or test mode, respectively.
Optional arguments:
-por--predictions– specify.envfile with prediction API needed info-eor--env– specify.envfile with exchange API needed info-dor--dashboard– specify this argument to run in dashboard mode on default 0.0.0.0:8050 (or change in config.py class DashServer)
Example run
python3 run.py run -e main.env -p probability_llm.env
With dashboard
python3 run.py run -e main.env -p probability_llm.env --dashboard
Alternatively run from outside project directory (change <path_to_run.py> to actual path):
sudo python3 <path_to_`run.py`> run -e main.env -p probability_llm.env --dashboard
One might be prompted to enter Administrator password
Run in test mode:
python3 run.py test -p probability_llm.env
Run in test mode (outside project directory):
sudo python3 <path_to_`run.py`> test -p probability_llm.env
Untested. Windows users must be smart enough to figure out the quirks.
Untested! The Dockerfile is shipped with essential commands.
I'm a Full Stack Software Engineer, specializing in all things Python. I code a lot, I don't publish my code a lot. I also happen to be a domain expert in Business, Economics & Finance, holding both M.Sc. & B.Sc. degrees in relevant fields of study.
I might be available for hire at a random time. Reach out with offers via email stevietfg+joboffer@gmail.com.
If you'd like to support my endeavors (or my lifestyle, in general), in accordance with the nature and character of this software, I accept donations in one truly CRYPTO currency – Monero (XMR) at addresses:
83woV72JcSXiPfrddb25znWiPULtkwFmZVXdPGkvNj6DArk3LUxedsG71A7ErK5cRHBTJPpjSorEz6j5sCJs1C1gCjmagaL
Monero is also the default crypto for this Crypto Autotrader to spot-trade.
For support & troubleshooting, email crypto.autotrader@outlook.com (with Subject: [AUTOTRADER SUPPORT]).
Any attention that I offer to the sender, I do at my own discretion; any sender would be respectful of my time.
This repository can be forked and made the needed changes with, in such cases that I don't respond.
For new ideas, issues can be opened, but appreciated are emails at crypto.autotrader@outlook.com (with Subject:
[AUTOTRADER IDEAS])
This software is heavily reliant on the following masterpieces of programming:
- ccxt library with their LICENSE (to date: Feb 20, 2025)
- Pandas library with their LICENSE (to date: Jan 27, 2025)
- NumPy library with their LICENSE (to date: Jan 27, 2025)
- stockstats library with their LICENSE (to date: Jan 27, 2025)
- openai library with their LICENSE (to date: Feb 6, 2025)
- plotly library with their LICENSE (to date: Mar 11, 2025)
- waitress library with their LICENSE (to date: Mar 11, 2025)
- dash library with their LICENSE (to date: Mar 11, 2025)
Currently, the usage of a lagged crossover is hardcoded as per the vision of the developer; however, altering that is very achievable:
updated can be the predict_pandas function in module predict.py – specifically, the way that lists signals and
anti_signals are filled. The part {"_delta" * self.wait_for_n_signal_lags} just needs to be eliminated, as do any
subsequent mentions of self.wait_for_n_signal_lags.
As is known, any trading system that utilizes Technical analysis techniques would be better off in markets with high liquidity. With regard to this, it would be wise to pick trading pairs that are high in trading volume. However, other reasons can be considered; thus making the liquidity-related suggestion moot.
This absolute unit of a Jupyter Notebook is my gift to the noob Technical analysis enthusiasts. It can help them backtest their crossover strategies AND the code happens to be very compatible with this Crypto Autotrader.
See my relevant affiliate links:
The software sends user-appealing info messages to the console.
