MyATMM ships with pre-curated screenshot parsers for the brokers that show up most often in our user base — ThinkOrSwim and Tastytrade today, Schwab and Tradier on the near horizon. But the AI import path was never meant to be locked down to those four. The whole point of leaning on a vision model to read screenshots is that it should, in theory, work on any broker whose transaction page contains the right pieces of information.
The question is: does it actually?
In this entry of the AI Import series we set up a clean portfolio in MyATMM and threw two non-curated brokers at the screenshot importer with zero pre-tuning: Fidelity and Robinhood. One worked. The other didn't — but the way it failed turned out to be more instructive than another success would have been.
To keep the experiment clean, we created a brand-new portfolio called AIOther001 inside MyATMM. No existing tickers, no prior transactions, nothing that could muddy the results of what the parser was extracting. From the main app, the path is straightforward: Tools → Import / Export → AI Import tab.
On that screen you can paste an image directly from the clipboard. There's no upload dialog, no file picker required — copy a screenshot, right-click the drop zone, paste, and the AI gets to work. That's the entire interaction model. The convenience is the point: snap a screenshot of your broker's transaction history, switch tabs, paste, click parse.
For this test we didn't tell the parser which broker the screenshots came from. We left it on auto-detect. That's the harder test — auto-detect has to figure out the broker layout from the screenshot pixels alone, with no hints.
We grabbed a screenshot from a Fidelity account that had two transactions in it: a cash transfer of $100 and a one-share stock purchase. Tiny sample, deliberately. The goal was to see if the AI could parse Fidelity's layout at all, not to stress-test it against thousands of rows.
Right-click on the AI Import drop zone, paste, click Parse. Within a couple of seconds the review grid populated:
Cross-checking against the source screenshot, every value matched. From there it's the same flow as a curated broker: tick the checkboxes for the rows you want to commit, click Confirm, Record, and Select. Because the portfolio was brand new, MyATMM also asked whether to create the ticker. Confirm that, and within seconds the transactions were live on the Cost Basis V2 page with last-price data already populated.
Fidelity's win wasn't really about Fidelity. It was about the layout of Fidelity's transaction page. The AI didn't need to know it was looking at a Fidelity screen. It just needed columns containing ticker, side, quantity, and price, separated cleanly enough that the model could distinguish row from row.
That's the rough rule of thumb for any non-curated broker: if a human can look at the screenshot and instantly tell what stock or option each row belongs to, the AI will probably be able to as well. If a human has to mentally translate "Bitcoin" into "BTC" or guess at quantities based on the dollar amount, the AI is going to have the same problem — only without the human context that lets you smooth over the gaps.
That's the lens we used going into the Robinhood test, and it predicted the result almost exactly.
We pulled up a Robinhood crypto transaction history screen and grabbed a screenshot. The data on it: a handful of crypto buys including Bitcoin, XRP, Stellar Lumens, and a DeFi token. The screen looks tidy enough at first glance, but it has a fundamental problem for any import tool: it shows the full names of the assets, not the ticker symbols. "Bitcoin" instead of BTC. "Stellar Lumens" instead of XLM. "XRP" is on screen as itself, but the rest are name-only.
Paste, parse, and the AI did exactly what you'd expect: it grabbed the dollar amounts cleanly, but it couldn't decide what ticker each transaction belonged to. So it imported every row as a ledger entry — meaning the cash side of the transaction recorded correctly, but the position itself didn't get attached to a ticker.
That's not really a "failure" in the sense of the AI giving up. It's the AI being honest: it knows it doesn't have enough information to commit a stock or option transaction, so it falls back to the most truthful thing it can record — the cash movement.
The vision model has no way to know that "Bitcoin" means BTC and "Stellar Lumens" means XLM. There are thousands of ticker symbols across crypto and equities; a name-to-ticker translation table is not the kind of thing you want a parser to guess at, because the cost of being wrong is silently filing a transaction under the wrong asset. Refusing to guess is the correct behavior.
This is the part that turned out to be more interesting than another clean import. Even when the AI can't fully parse a screenshot, the review grid is still useful — because you can edit any row before you commit. So a "failed" Robinhood import isn't actually a dead end. It's a partial import where you fix the missing pieces inline.
Walking through one row to show what that looks like:
BTC.The moment you save, the row gets pulled out of the generic ledger bucket and grouped under the correct ticker. The ledger entry count drops, the BTC position appears, and the import is essentially complete for that row. Repeat for XLM, repeat for the rest, and you've turned a "failed" parse into a working import.
It's slower than the Fidelity flow. But it's still faster than typing every transaction in by hand, and it preserves all the dollar amounts the AI got right.
Two specific tests don't generalize perfectly, but they do suggest a reasonable mental model for what to expect when you point the AI importer at a broker that doesn't have a pre-curated parser yet.
The deciding factor is what's on the screen, not which broker logo is in the corner. If the broker's transaction page surfaces ticker, side, quantity, and price as cleanly readable columns, auto-detect will probably work the same way it did on Fidelity. If the broker's UI is friendly to humans but light on machine-readable detail — full asset names instead of tickers, fractional quantities that don't map to whole shares, fees hidden behind a settlement workflow — expect a partial parse and plan for the manual recovery step.
For brokers in that second category, the CSV export path is almost always the better answer. We'll be testing the CSV side of this in upcoming entries in the series for Schwab, Tastytrade, Robinhood, and Tradier specifically.
Both the curated parsers and the auto-detect path improve when we see real-world examples of broker screens that don't work. If you try the AI importer with your broker and the result is a partial parse, a totally failed parse, or anything that doesn't look quite right, the most useful thing you can do is email support@myatmm.com with a screenshot of the original page (with personal info redacted) and a note about what you expected to see.
Curated parsers can typically be added within a week from a handful of sample screenshots. The more samples we receive for a given broker, the higher the floor on parse accuracy for everyone using that broker.
If you're new to MyATMM, the screenshot importer is the on-ramp — the way you get existing trades into the system without keying them in by hand. From that point on, MyATMM is a purpose-built tracker for option sellers running covered calls, cash-secured puts, and the wheel strategy. Cost basis updates automatically as you add transactions. Premium collected gets factored in. Assignments are tracked separately so you always know your true cost basis, not just the share-purchase number.
That's a meaningful upgrade over the alternative most option sellers default to, which is running the wheel strategy on a spreadsheet. Spreadsheets work fine for one ticker. They start to fail the moment you have rolls, assignments, dividends, and multiple positions across a portfolio — which is exactly when accurate tracking starts to matter most.
For a fuller picture of what kinds of tracking tools work well for this style of trading, the options trading journal software guide is a good starting point.
Options trading involves risk and is not suitable for all investors. Past performance does not guarantee future results. MyATMM is a tracking tool, not a financial advisor. The information in this post is for educational purposes only and should not be considered investment advice. Always consult with a qualified financial advisor before making investment decisions.
MyATMM provides purpose-built cost basis tracking for option sellers, with the flexibility to track covered calls, cash-secured puts, and wheel strategy positions.
Track up to 3 tickers completely free forever. No credit card required.
Create Your Free Account TodayBuilt for covered calls, cash-secured puts, and the wheel strategy