The current AI implementation in MarginNote has a lot of potential, but it still feels underdeveloped. While the underlying AI models are strong and generate great content depending on the prompt, the integration features within MarginNote are heavily restricted.
Here are the core issues I’ve run into:
1. Card Structure & Multi-Topic Limitation
-
The Problem: The AI cannot generate multiple comments for a single “Create Card” action. When a text covers multiple topics, the AI recognizes them but crams everything into one card with a single long comment instead of splitting them into multiple comments within that card.
-
The Workaround: I currently have to use an external AI (like DeepSeek) to generate the cards and manually paste them into MarginNote. This allows me to properly split topics into multiple comments and apply clean formatting (see attached screenshot).
Yellow Card: Created directly with MarginNote AI. It groups everything into just one card with a single, long comment.
Green Card: Created externally with DeepSeek and manually pasted. This allows the AI to split different topics into multiple separate comments within a single card.
2. Formatting & Markdown Hurdles
-
The Problem: Formatting integration is practically non-existent. Markdown is tedious and user-unfriendly for this workflow. The current setup prevents using standard formatting options (like colors, bold, italics) natively.
-
The Goal: Formatting tasks—especially coloring and styling text—should ideally be automated by the AI, rather than requiring manual Markdown workarounds.
3. Redundant Features & Confusion
- AI OCR vs. AI Chat: These two features essentially do the same thing, which creates unnecessary confusion. However, AI OCR requires an excerpt, meaning the text still cannot be formatted normally (e.g., bold and italics break due to emphasis rules).
- Review Card Limitations: The AI Chat lacks a direct option to generate review cards simultaneously when creating a card. While AI OCR has this function, it is limited to only one review card per mind map card, rather than supporting multiple review cards where needed.
My Proposed Solutions & Feature Requests
To fix these structural issues, I would love to see the following changes implemented:
-
Restructure AI OCR & Introduce a Dedicated Card Creator: Keep AI OCR focused on simple tasks like translation, and build a proper, dedicated tool for automated AI card generation. This tool should use custom prompts while fully supporting MarginNote’s core features:
-
Native text formatting with colors.
-
Splitting content across multiple comments per card.
-
Automated generation of multiple review cards per map card.
-
-
Mind Map-Bound AI Chats: Currently, whenever you select text and use the AI button with a custom template, it spins up a brand-new, isolated chat. The only way to maintain a continuous context right now is to manually select text (or use OCR) and copy-paste it into an existing chat window. If you try to use a custom template directly from a selected excerpt or image, it bypasses the main chat and forces a brand-new session every time. It would be much better to bind a single persistent chat to the entire mind map. Every time you use a custom template, it should feed into this main chat history. This way, you can provide ongoing feedback on which cards were good or bad, allowing the AI to learn and improve card quality continuously over time.