The First Truly No-Code Algotrading Platform Is Here
There have been tons of "no code" algotrading platforms in the past.
The premise: Chain logic together using a drag and drop UI. Run it.
The solution: A proprietary evaluation engine that translates logic into trades.
Now enter LLM's. Platforms generate strategy code on the fly, and run it on hardware.
The compiler-like evaluation engines of the past become obsolete. Flexibility increases ten-fold.
No SaaS subscription? No problem. The free version of ChatGPT will generate the strategy code for you in minutes.
A new generation of algotrading SaaS products pop up: LLM wrappers that abstract away the hassle of deploying to the cloud. Vercel for websites - Tradecel for your trading strategies.
However, we're not done yet. Code is still being generated and run.
Enter the final stage: LLM evaluation. Now LLM's not only generate strategy logic, they also "run" the strategies. Algotrading code ceases to exist. LLM's generate strategies and run them - independently. No code is ever being generated. The only server involved is the server that the LLM runs on.
Time per iteration goes from minutes to seconds. Now, to modify a strategy, all you need to do is change a prompt. No code needs to be generated. No programs need to be re-deployed. As an algotrader, you can now have near-instant feedback.
The cost? The pricing equation changes. Your expenses are no longer:
Server Size * Uptime
But rather:
Evaluation Frequency * Price Per LLM Token * Uptime
This works because the second factor in the equation (Price Per LLM Token) is now exceptionally low. It has decreased by several magnitudes within the last 2 years, and continues to decrease (although no longer at an exponential rate).
With today's pricing you can feed an LLM 1,000 data points - and get a trading decision back - for less than a tenth of a cent.
This allows you to query an LLM thousands of times per day, while still paying less than you would pay for a typical algotrading SaaS subscription.
Good enough for all strategies? No. High frequency strategies require millions of evaluations per day.
Good enough for most strategies run by retail algotraders? Yes. Most strategies run by retail algotraders aren't high frequency enough to require per-second evaluation. Most retail algotraders just want to run a TA-based strategy on higher timeframes such as 30m or 4h. For this, an evaluation once or twice per minute is usually more than adequate.
LLM-based strategy generation enables these algotraders to go from idea to execution in just minutes (the time it takes to write a prompt).
And LLM-based strategy evaluation enables them to iterate in mere seconds (the time it takes to modify the prompt).
Algotrading ceases to be about algorithms, and becomes about prompts instead. In fact, one might now refer to it as prompt trading.
Enter Everstrike. The first platform that combines LLM-based strategy generation with LLM-based strategy evaluation. Unlike previous attempts at AI-based algotrading, there is not a single line of code to be found anywhere. The platform is truly no-code. LLM's generate strategies and run them. The logic is the prompt - not the algorithm. The engine is the LLM - not a Python/C++ executable.
The result? Near-instant strategy iteration at a cost that matches the price of running a 4GB RAM AWS server. Simply modify your prompt, hit "Submit" and watch your changes come into effect within 100 milliseconds.
Cost per evaluation? $0.0002-$0.002, depending on the amount of data you want to include. This translates to a cost of $0.5-$5 per day. Your infrastructure expenses are no longer based on the type of subscription or server that you choose - but rather on the amount of data that you want your LLM to consider.
Coming up next? The "Skynet" of algotrading where funded agents generate strategies and trade without human oversight. For example, you might spin up an agent with a prompt that asks it to try out a ton of different strategies and pick the best performing one.