An AI trading coach reads every entry in your journal — not just your trades, but your notes, rules, and reflections — and surfaces recurring patterns you'd miss in a Sunday review. The four useful capabilities are per-trade critique, cross-week pattern detection, playbook pressure-testing, and weekly review generation. The three useless claims to ignore are price predictions, "smart entries", and auto-generated trade ideas. Trade Journal AI's AI Coach is built on Anthropic's Claude Sonnet, with three cost models: $3/month fair-use cap (included), $9.95/month Unlimited add-on, or BYOK (bring your own Anthropic key).
- An AI coach reads your full journal — every note, rule, trade, and reflection — not just numeric trade data.
- Four useful capabilities: per-trade critique, cross-week pattern detection, playbook pressure-testing, weekly review generation.
- Three useless claims to ignore: price predictions, smart entries, auto-generated trade ideas. None of these are what a coach is for.
- The model matters less than the grounding. The same Claude Sonnet model answers very differently depending on whether it's reading your journal or guessing from training data.
- Three cost models: $3/mo fair-use cap (included in base), $9.95/mo Unlimited add-on, BYOK with your own Anthropic key.
- The AI coach accelerates your own review — it doesn't replace it. Coaches that promise to "trade for you" are the wrong product.
What is an AI trading coach?
- AI trading coach
- A trading-journal feature that uses a large language model to read across all journal entries — trades, notes, rules, and reflections — and surface recurring patterns the trader would miss in manual review. Distinct from a price-prediction tool or a trade-idea generator.
The useful framing: an AI coach is a research assistant who has read every entry in your journal and can answer questions about it in plain English. It doesn't predict the market. It doesn't tell you what to trade. It reads what already happened in your trading and helps you see it more clearly.
Almost all the value comes from grounding — the AI's answers are based on your actual journal data, not on training-data generalizations. A non-grounded coach answers "what's my best setup?" with generic advice. A grounded coach answers with "your A+ Asia-session BTC long has run 8/12 winners over the last 30 days at +1.8R average; this is your best setup by expectancy." The difference is whether the AI is reading your data or making something up.
The four useful AI-coaching capabilities
Ordered by value-per-question. The first one — per-trade critique — is where most traders start getting value within the first week. The fourth one — weekly review generation — is where the time savings compound.
Per-trade critique
Read a single trade's chart, your pre-entry rationale, your notes, and the outcome. Surface what worked, what failed, and what to do differently next time. The fastest way to find that one mistake you keep making.
Cross-week pattern detection
Read 100+ trades across a month and surface patterns the Sunday review misses. "Your losing trades cluster between 2 and 4 AM UTC." "Every loss in the past two weeks broke the wait-for-15-minute-close rule." Patterns that exist in the data the entire time.
Playbook pressure-testing
Ask specific questions in plain English. "How has my Asia-session ETH-short setup performed in the last 30 days?" Answer in seconds. "Show me trades where I broke my sizing rule" — exact list. The journal becomes queryable.
Weekly review generation
Best/worst setup, rule-adherence rate, time-of-day clustering, one specific suggested adjustment for the coming week — generated from the data rather than the trader's memory. Compresses 30 minutes of manual review into a 5-minute read.
What AI coaches CAN'T do
The claims that show up in marketing copy but don't survive contact with reality:
- Price predictions. "AI will tell you what BTC does tomorrow." It can't. No LLM has predictive edge on price action that's not already in the public consensus. If a tool claims to predict prices, it's marketing copy.
- "Smart" entries. "AI generates trade signals from market data." There is no signal in publicly available market data that an LLM extracts better than the trader does. The LLM doesn't have access to anything you don't.
- Auto-generated trade ideas. "AI suggests trades you should take." This is a trade-idea engine, not a journal coach. They're different products. A journal coach reads what you already did. An idea engine generates what you should do — and the second one rarely works as advertised.
The shorthand: a useful AI coach is backward-looking. It reads what already happened in your trading. The moment a tool advertises itself as forward-looking — predicting prices, generating ideas, signaling entries — it's a different and less reliable category of product.
How Trade Journal AI's coach is built
Three specifics matter:
Model: Anthropic Claude Sonnet
The same model runs across every pricing tier — the differences are in usage allowance, not model quality. There's no "better AI tier." Anthropic's Claude Sonnet is the model. The journal sends your relevant journal entries as context with each request, so the AI sees your actual data.
Grounding: reads your full journal, not just trades
Most trading-journal AI features only see trade data — the numeric records. Trade Journal AI's AI Coach sees everything you've written into your journal: pre-trade rationales, post-trade notes, weekly review entries, rule definitions, setup descriptions. The pattern detection runs against this full corpus, not just the numeric subset. That's why a question like "what's the pattern in trades where I felt overconfident?" gets a real answer — the emotional tag is in the notes, not in the price data.
Privacy: your data stays your data
Journal contents are sent to Anthropic only for the duration of the AI request being processed. Anthropic's API terms commit to not training on data sent through the paid API. The journal app doesn't share your entries with other users or use them to train any internal model. When you cancel, your entries are kept in read-only mode for 90 days for export, then purged.
How the AI Coach is priced
Three options. All three access the same Claude Sonnet model — the choice is about usage allowance, not capability.
- BYOK (Bring Your Own Key)
- A pricing model where the user supplies their own AI provider API key (e.g., Anthropic) directly to the journal app. The journal makes AI calls under the user's account, so the user pays the AI provider directly at metered cost rather than paying the journal's fair-use cap.
When to pick which
- Default fair-use cap ($3/mo, included) — fits most active traders. Generous enough for daily per-trade critique + weekly review generation. Start here.
- Unlimited add-on ($9.95/mo) — pick this if you hit the fair-use cap consistently and don't want to think about it. The add-on is cheaper than BYOK only if your usage is moderate-to-heavy but not extreme.
- BYOK — pick this if you already have an Anthropic account, or if you're a heavy AI user who wants unlimited metered cost rather than paying the Unlimited add-on. Also useful for traders who want billing separation between the journal app and the AI provider.
The AI Coach that reads your full journal.
Built on Anthropic's Claude Sonnet. Grounded in your trades, notes, rules, and reflections. $19.95/mo includes the fair-use cap. Unlimited add-on or BYOK available. 14-day money-back guarantee.
Will an AI coach replace my own review?
No, and that's a feature. A trading journal works because the trader actively engages with their own data. The AI coach accelerates the engagement — surfacing patterns faster, answering questions in plain English, generating draft weekly reviews — but the trader still makes the decision about what to change.
A coach that says "I'll trade for you" is the wrong product. A coach that says "here's what your data shows, here's what I'd consider, here's what you decide" is the useful one. The trader still has to read, think, and act. The coach just compresses the time it takes to do each of those things.
The traders who get the most out of an AI coach are the ones who treat it the way they'd treat a thoughtful trading mentor — ask specific questions, push back on the answers, and ultimately make their own calls. Traders who outsource their thinking to it tend to drift toward bad decisions. The journal and the coach are tools; the trader is still the trader.
Frequently asked questions
What does an AI trading coach actually do?
Four useful things, in order of value: per-trade critique (read a single trade's chart, notes, and outcome, then surface what worked and what didn't), cross-week pattern detection (find recurring losing or winning patterns across 50+ trades that a Sunday review would miss), playbook pressure-testing (answer questions like "how has my Asia-session ETH short performed in the last 30 days?" in seconds), and weekly review generation (produce best/worst setup, rule-adherence rate, and one specific adjustment from the data rather than the trader's memory).
What CAN'T an AI trading coach do?
Three things AI-coach marketing copy often claims that don't survive contact with reality: price predictions ("AI will tell you what BTC does tomorrow" — it can't), "smart" entries ("AI generates trade signals from market data" — there is no signal in market data that an LLM extracts better than the trader), and auto-generated trade ideas ("AI suggests trades you should take" — this is a trade-idea engine, not a journal coach). A useful AI coach reads what already happened. It doesn't predict what will happen.
What AI model does Trade Journal AI use?
Anthropic's Claude Sonnet, accessed via the Anthropic API. The same model runs across every pricing tier — the differences are in usage allowance, not model quality. Trade Journal AI's AI Coach is grounded in your full journal entries (notes, rules, trades, reflections) rather than just trade data, so it has more context than tools that only read the trade history.
How is the AI Coach priced?
Three options. Default: a $3/month fair-use cap, included in the $19.95 base subscription, designed to cover normal active-trader use. Unlimited add-on: $9.95/month (or $99/year), for traders who want no usage limit. BYOK (Bring Your Own Key): supply your own Anthropic API key and pay Anthropic directly at metered cost — useful if you want unlimited usage without paying the Unlimited add-on, or if you already have an Anthropic account. All three options access the same Claude Sonnet model.
How does BYOK work?
In Settings → Integrations, paste your Anthropic API key (from console.anthropic.com). All AI Coach requests then run against your Anthropic account at your metered rate rather than the journal's shared key. The journal app never bills you for AI usage when BYOK is on. Typical active-trader usage runs $2–8/month at Anthropic's metered rate, depending on volume.
What does "grounded in my journal" mean?
The AI Coach reads your actual journal entries — every note, every rule you set, every post-trade reflection, every trade tag — as context for each question. A coach that's NOT grounded would answer "what's my best setup" with generic advice. A grounded coach answers with "your A+ Asia-session BTC long has run 8/12 winners over the last 30 days at +1.8R average; this is your best setup by expectancy." The difference is whether the AI is reading your data or guessing from training data.
Will an AI coach replace my own review?
No, and that's a feature. A trading journal works because the trader actively engages with their own data. The AI coach accelerates the engagement — surfacing patterns faster, answering questions in plain English, generating draft weekly reviews — but the trader still makes the decision about what to change. A coach that says "I'll trade for you" is the wrong product. A coach that says "here's what your data shows, here's what I'd consider, here's what you decide" is the useful one.
Is my journal data used to train other AI models?
No. Trade Journal AI sends your journal context to Anthropic only for the duration of the AI request being processed. Anthropic's API terms commit to not training on data sent through the paid API. The journal app doesn't share your entries with other users or use them to train any internal model. When you cancel, your entries are kept in read-only mode for 90 days for export, then purged.