An AI lab that keeps researching your trading strategies.

The whole research chain is free. Research Agent runs on your computer, searches for strategies, writes them, and simulates them. The studio that comes after it is free too: fine-tuning a strategy the Agent found, testing it again, and reading the results.

Automations is free too, while OpenTradeStudio is in Public Beta. There is no payment and no credit balance. The free period is how we test this infrastructure with real use. An account can create up to 5 automations. Each one sends buy and sell signals to the channels you configure. Future post-beta value: $49/month. Current price: $0.

No. Most people do not have time to write Java, and the product is not built around that. Research Agent is the start of the chain: install it, and it searches for strategies, writes the code, and tests each one. You come back to the studio only for a strategy it already found — to fine-tune it, to test it again, or to set the channels that will receive events when the website runs it.

Startup templates are available if you want to try a rule by hand. Example: open a moving-average template and edit it yourself. That works. It is not the main idea. The main idea is that Research Agent does the search, and the studio is the place for the few results worth a second look.

You do not have to sit and watch. Install Research Agent, sign in, set up Local AI, and it starts searching. It works one strategy at a time, about an hour each by default, and only when you want it to. Research is switched on or off from Configuration, so the Agent can run for an evening and stay off the rest of the week.

Example: turn it on after dinner and off before sleep. In those hours it has searched, simulated, and left the better ideas in the studio. When you have a minute, open one there to fine-tune it or test it on another range. A startup template can be run by hand in the editor. That is the manual path, and it is not the one this product is built on.

What makes the search efficient is a faster, smarter Local AI. Research Agent calls that model again and again, so the hardware decides both how smart the model can be and how fast each request comes back. Better hardware runs a smarter model and finishes those requests sooner. An embedded model needs a graphics card with at least 8 GB of VRAM, or the Agent can use llama.cpp or Ollama already running on your network. Keep the context size inside what that computer can hold.

A cloud AI is faster and smarter, and it is not advised. Research Agent keeps calling the model for hours. In about 8 hours that can use up every token in a monthly subscription. Local AI is the setup this product is built on.

A smaller job finishes sooner. Example: BTCUSDC over the last two years, for two hours, goes further than every crypto pair and ten years of history in the same evening. Notes, a document, a link, or a YouTube transcript tell the Agent what to look for. On the Research backlog, star what should run next and skip the rest.

You do not stay at the screen. The Agent works in the stretches you choose. The studio is where you come back to fine-tune what it found, not where the search starts.

Research Agent searches on crypto from Binance, US stocks and ETFs from Alpaca, and forex, metals, and oil from OANDA, down to 1-minute bars. The free Alpaca stock feed is delayed by 15 minutes. Symbols and intervals are set once, in Configuration, and the Agent uses them while it searches. The buy question and the sell question can use different intervals.

Example: leave the Agent on EURUSD, a buy check every 4 hours and a sell check every 15 minutes. It brings the strategies back to the studio. There you can point one of them at AAPL or at gold and test it again. That second test is fine-tuning. The search itself belongs to Research Agent.

Scalping means taking many very small moves and getting out in seconds or a few minutes. Traders who scalp usually watch a 1-minute or 5-minute chart and may place dozens of trades in a session. Each win is tiny, so the spread, the fee, and a late fill can erase it. On a chart that fast, a lot of what looks like a signal is noise: the price ticks up and down without a move that lasts.

OpenTradeStudio can ask a strategy every minute, because the history goes down to 1-minute bars. That does not make scalping a good job for Research Agent. The Agent would spend its hour testing rules that fire on noise, then bring back a pile of trades that look busy and mean little. Scalping is not advised. Set the interval longer, and let the Agent search for a move that is still there after the noise has passed.

Day trading means opening and closing in the same session. Nothing is held overnight. A day trader might buy in the morning and sell 45 minutes later, often working from a 5-minute or 15-minute chart, sometimes a 1-hour chart for the direction of the day. It sits between scalping and a multi-day hold: fewer trades than a scalper, and no overnight gap.

Maybe. The intervals can be set that way — 15 minutes, 30 minutes, an hour — and Research Agent will search with whatever you configure. Day trading itself has not been tested here, so there is no result to stand on. If you try it, treat what comes back as an experiment: fine-tune it in the studio and run it again before you let the website trade it live. The style this product has actually measured is the slower one, in the next answer.

Swing trading holds for a few days or a few weeks, long enough to catch a move and short enough that you are not waiting months. Published practice uses hourly, 4-hour, and daily charts for that. A 4-hour candle already smooths out a lot of the noise that fills a 1-minute chart, so a buy or a sell is a decision about the swing, not about the last tick.

Yes. This is the style to give Research Agent. On OpenTradeStudio the intervals that have returned the best results are 3 hours and 4 hours — 180 minutes and 240 minutes in Configuration. Example: set the buy check to 4 hours and leave the Agent on that symbol for the evening. It searches for rules that survive that candle, not rules that flicker inside it. Bring the better ones into the studio, test them again, and only then decide whether the website should run one live.

No. Research Agent only researches. It never holds money and it never places an order. A strategy it found does not go live on the Agent’s computer. The chain splits there: the studio is where you fine-tune the strategy and configure the channels, and the website — not the Agent — runs that strategy on live data and sends the buy or sell event.

Example: Research Agent finds a rule. You fine-tune it in the studio and set Telegram. You turn the strategy on on OpenTradeStudio. The site posts “buy”. A script you run reads that message and sends the order to your broker. If that script is off, nothing is traded.

A result from Research Agent shows how those rules behaved on past candles. It is one step in the chain, not a promise about the next year. Fees, slippage, and a delayed stock feed are not the same as a live fill.

Example: the Agent can bring back a rule that bought every dip through 2023. That same rule can lose in a year that does not dip the same way. Before the website runs it, open it in the studio and test it again on another range. Use the profit and the number of trades to compare versions. What the Agent found is research. The live run, if you choose one, is a later step on the website.