What can I actually do on a free model, and where will it fall short?

Free is enough for tagging, drafting and summarising — the high-volume, low-judgement work. It falls short on chart reads, on multi-step Coach questions, and anywhere a wrong answer is worse than no answer.

Sorted by how well it goes.

Works well on free:

• Sentiment and emotion tagging. Classifying "I chased this" as frustration is not a hard problem, and it is the highest-volume AI task in the app. Even the paid tier downgrades to its cheapest model for it.
• Debrief drafts and daily summaries. Turning your numbers into prose.
• Weekly pattern reports, if your provider's free model handles tools — on OpenRouter it does.

Works, with caveats:

• Coach questions, for anything answerable from a page of context. Deep questions that need the Coach to go and fetch several things first are where free models stumble, and where economy mode's two-phase trick helps if you also hold a paid key.
• Chart reads, on the vision-capable free models only. Expect a thinner read: fewer levels named, more nulls. A thin read is the app behaving correctly — the vision prompt is written to return nothing rather than guess.

Do not rely on free for:

• Setup health checks and rule generation. These are judgement calls on your actual trading and a plausible-but-wrong answer is expensive.
• Anything you will act on without checking.

The honest recommendation: start on Groq or OpenRouter, which costs nothing and tells you whether the AI features earn a place in your routine at all. If they do, add a paid key and keep the free one — the next question explains why that combination is better than either alone.