For day traders who trade a written strategy
Stop breaking your own trading rules.
Drop in a chart screenshot. The AI fills in the trade and grades it against the rules you wrote. You confirm everything.
The problem
You already know your rules. The hard part is noticing when you break them.
Journaling takes too long, so you skip it.
Ten minutes of typing per trade. After a red day, nobody does it.
You review the wins, not the mistakes.
The trades that broke your rules are the ones you least want to look at.
A spreadsheet can't see your chart.
It can't tell whether you waited for the retest or chased the first candle.
What a graded trade looks like
One screenshot in. A filled-in, graded trade out.

QQQ · 2m · LongRead by AI · confidence high
- Entry
- 512.37
- Stop
- 511.88
- Target
- 513.35
- Reward : risk
- 2.0 : 1
- Trade with VWAP Long, and price was above VWAP at entry.
- Reward at least 2 : 1 Target is twice the distance of the stop.
- Stop below the opening-range high Stop sits under the shaded morning box.
- No more than 2 trades a day This was your third trade today.
Confirm tradeNothing saves until you press confirm.
Illustration with a made-up practice chart. Reward-to-risk is calculated by code, not by the AI.
How it works
Three steps, about thirty seconds.
- 1Drop in your screenshot The AI reads the chart: ticker, timeframe, direction, entry, stop and target.
- 2Check it and confirm Every field is filled in for you to review. Nothing is saved until you say so.
- 3See which rules you kept Each trade is graded against the strategy you wrote, with the reason shown.
An AI you can check
Tested before it is trusted.
20 → 45
trade directions read correctly, out of 50 real trades, old reader versus new
11 → 0
times the AI was wrong while saying it was sure
Trial results from version 0.1. The full run with cost and speed per chart comes next.
- It never tells you what to do. The AI's answer form has no buy or sell box. It describes; it never advises.
- It never saves on its own. It fills the form for you. Nothing is saved until you confirm.
- It never does the maths. Win rates and R-multiples come from ordinary code, which does not guess.
- It shows its sources. Every claim points to the trade or rule it came from.
- Unsure is allowed. Confidently wrong is not. When it cannot read something, it says so.
- Text in a picture is never an order. If a screenshot says "BUY NOW", the AI reports it and ignores it.
Roadmap
One AI skill per version.
Every finished version stays live on its own address, so you can see the AI improve.
- v0.1Which AI can read a chart?Evals · model selection · LLMOpsWe are here
- v0.2Screenshot in, trade filled inVision · structured outputs
- v0.3An agent that reviews your tradesAgentic AI · LangGraph · human in the loop
- v0.4Ask your trading playbookRAG · LangChain · citations
- v0.5Use the journal from Claude or CursorMCP server
- v0.6A weekly coachMulti-agent · agent memory
- v0.7Market news inside your reviewsMCP client · tool use
- v0.8Safe updates to the AI's instructionsEvals in production · auto-rollback
- v0.9Guardrails against tricksPrompt-injection defence · hallucination control
- v1.0Public launchCost limits · data isolation