Most "AI" in trading apps is a ChatGPT box bolted onto the dashboard. Journali's AI Coach is different: it reads your full trade history through Claude (Anthropic), finds the specific patterns you keep repeating, and tells you where your edge actually lives — and where it's quietly leaking.
"Your entries before 10:15 AM EST are net +$3,840 across 62 trades. Everything logged after 2:30 PM is net −$2,918 across 41 trades — you're losing 68% of them and holding them 2.3× longer than winners. You're a morning-session trader who keeps showing up for the afternoon."
Open any trading journal that slaps "AI" on the landing page and you'll find the same thing: a chat window that forwards your question to GPT-4 with no access to your actual data. It gives you the same generic advice it would give anyone — "Always respect your stop loss. Journal consistently. Manage your risk." It doesn't know you. It has never seen a single trade you made.
If the AI feature in your trading journal can't cite a specific trade, a specific time window, or a specific setup you traded — it has no idea who you are. It's a generic LLM answering generic trading questions. You could get the same answer from a free ChatGPT tab.
Journali's AI Coach is wired directly into your trade table. Every analysis is grounded in your symbols, your timestamps, your sizes, your P&L. When it says you're losing money after 2:30 PM, it's because it counted them.
Your trading edge is hidden inside your own history. You just can't see it because you traded it. The AI Coach reads your trades like a senior mentor would — at scale, without fatigue — and tells you the specific, falsifiable patterns that are either making you money or quietly costing it.
Scans your trades for setup-level, time-of-day, day-of-week, and symbol-level patterns. Returns ranked findings with sample size and P&L impact.
Pick any trade — winner or loser — and get a written postmortem. What the setup was, where it worked, where it failed, what a cleaner version would have looked like.
Every Sunday, an AI-written review lands in your inbox: your best setup of the week, your most expensive mistake, and one specific adjustment to test next week.
Flags revenge-trading signatures — size doubling after a loss, trading outside normal session hours, back-to-back losses on the same setup — before the day gets worse.
Treats each of your tagged setups separately. Your Breakout win rate, your VWAP Reject avg R, your Opening Drive worst-case drawdown — analyzed in isolation.
Premier includes 50 AI queries per day — enough for a full debrief, several reconstructions, and ad-hoc questions. Resets midnight UTC.
Morning: "Show me my five worst days by P&L. What do they have in common?" The AI surfaces the setup, the time window, the market regime, and the behavioral signature (size creep, outside normal hours, revenge entry).
Mid-session: "I just took two losses on NQ. Should I keep trading?" The AI checks your history for the exact signature of two-losses-in-a-row: did you recover that day historically? What was your size behavior? What was the P&L outcome in the remaining session?
Sunday review: A weekly debrief email summarizing the week — best setup, worst setup, one concrete adjustment, and three questions to hold yourself accountable to next week.
| Capability | Journali AI | TradeZella "AI" | Generic GPT tab |
|---|---|---|---|
| Reads your actual trade history | Yes | Partial | No |
| Cites specific trades in answers | Yes | No | No |
| Weekly written debrief | Yes | No | No |
| Tilt detection | Yes | No | No |
| Trade reconstruction | Yes | No | No |
| Model used | Claude | GPT-3.5 | GPT-4o |
| Daily limit | 50 calls | Throttled | Rate-limited |
AI Coach is a Premier ($30/mo) feature. It's built for traders who already have enough history to learn from but not enough time to dig through it manually.
Claude, made by Anthropic — specifically the Sonnet and Opus family depending on the query type. We picked Claude over GPT because it's better at long-context analytical reasoning, which is exactly what reading hundreds of trades requires. The AI sees your trade table directly; it's not a chatbot with a search tool duct-taped on.
No. Anthropic's API terms prohibit training on customer data by default, and Journali never ships your trades to any third party for model training. Your data stays in Supabase, goes to Claude only for the specific query you ran, and is not retained by Anthropic beyond the inference window.
Each AI query runs a Claude inference, which costs us money. 50/day is the ceiling we can sustainably offer at the Premier price point — and in practice it's more than most traders need. A weekly debrief is 1 call. A reconstruction is 1 call. Ad-hoc questions are 1 call each. The limit resets midnight UTC.
The AI will still answer questions, but pattern detection needs sample size to be meaningful. Under ~50 trades, most findings will come with the caveat that the sample is too small. This is a feature, not a bug — we'd rather the AI tell you it doesn't have enough data than hallucinate a pattern that isn't real.
No. It's a retrospective analyst, not a signal service. It reads what you did and tells you what it noticed. It will never tell you "buy NQ at 18,420." That's not what it's for — and anyone promising AI trade signals is selling you something that doesn't work.
A senior mentor who reads every trade you've ever taken — without fatigue, without judgement, and at the speed of Claude. That's what Premier unlocks.
Start Journali Free → Premier · $30/mo · 50 AI calls/day · Upgrade anytime