Heptabase thinks in space, AI notes think in chapters

Heptabase is a visual research tool: you put cards, PDFs and highlights on a whiteboard, move them around until the structure makes sense, and keep a daily journal on the side. SumizAI has no canvas. It takes the answers from your conversations with an AI model and turns them into AI notes placed in a numbered table of contents.

Both use AI, in very different ways, and they cost very different amounts. This comparison explains both, says plainly when Heptabase is the better tool, and lists what survives a move. Heptabase details come from heptabase.com, its pricing page and its help centre, read on 23 October 2026.

A chat with an AI model in the app, with the answer about to be saved as a note
The starting point is a conversation, not a canvas: the answer you keep becomes the note.

Heptabase next to the chat-first vault

Heptabase plans and features on 23 October 2026
SumizAIHeptabase
Unit of workAn AI note made from one chat answerA card on one or more whiteboards
Who writesThe model; you correct and mergeYou, with help from the built-in AI agent
Price7 days free, then 1 USD a month; model costs paid to your providerPro 8.99, Premium 17.99, Premium+ 53.99 USD a month; yearly billing 25% cheaper; 7-day trial
AI7 providers on your own key (Anthropic, OpenAI, Gemini, Groq, OpenRouter, Ollama, custom)AI credits per plan: 100, 1,800 or 8,100 a month; Pro uses Gemini models only
PlatformsmacOS first, coming to the Mac App Store; Windows, iOS, Android in preparationmacOS (Intel and Apple silicon), Windows, Linux, iOS, Android, web
StructureTable of contents up to 6 levelsWhiteboards, mind maps, journal with calendar
PDFsImported as source text for a new noteAnnotate and highlight; OCR and unlimited uploads on Premium
CollaborationNoneReal-time collaboration, invite others for free
Storage.md files on your disk, no serverAccount with sync; notes available offline
ExportZIP of Markdown, PDF in two levels, PowerPoint with 15 themesAll notes as Markdown plus All-Data.json

Two ways to handle a pile of papers

Say you are reading twelve papers on attention in transformer models. Heptabase lets you open each PDF, highlight the key passages, turn highlights into cards and arrange the cards on a whiteboard called "Attention": one cluster for sparse attention, one for linear variants, arrows between them. The map is the thinking, and you build it with your hands.

The chat-first route starts elsewhere. You import the twelve PDFs, one by one, and each becomes an AI note with the extracted text kept as its source. Then you ask questions: "Which of these papers measure memory use, and how?" The model answers from the table of contents and the full text of up to five notes (24,000 characters in total), names the notes it read, and the answer you keep becomes another note in chapter 4.2.

No map appears, ever. What you get is a growing, numbered set of AI notes, each one tied word for word to the question that produced it.

Heptabase is the better tool if

  • You think spatially. Seeing cards side by side on a canvas is the point, and nothing in the vault replaces that.
  • You annotate PDFs heavily and want the highlights next to your notes.
  • You work with a supervisor or a team: Heptabase offers real-time collaboration.
  • You need iPad, Android, Linux or a browser now.
  • You prefer AI inside one subscription, with credits included, over managing your own API key.
  • You keep a daily journal with tasks; Heptabase's journal has a calendar and a to-do view.

Choose the chat-first vault if

Your best ideas already come from long conversations with a model, and what you lack is a place where those answers end up sorted as AI notes. Here you never drag a card. The model picks the chapter by its number, or asks to open a new subchapter, and a duplicate check runs before every save.

Cost is the other reason. Heptabase's cheapest plan is 8.99 USD a month, and access to GPT and Claude starts at Premium, 17.99 USD. Here the app costs 1 USD and you pay your provider only for what you use, or nothing with a local Ollama model.

And files. A vault exports as plain Markdown with relative links, which is also why people shopping for Obsidian alternatives end up here: the notes open in Obsidian as they are. The comparison with Obsidian, written for anyone after an Obsidian alternative, goes deeper, and the AI notes app page shows the basics.

What moves, and how

Heptabase to here: Settings → Backup → Export now writes all notes as Markdown files together with All-Data.json. Point the importer at the folder; it reads .md, .markdown, .txt and .pdf and processes one file at a time. The JSON file with whiteboard positions has no use on this side, so the spatial layout is lost.

Here to Heptabase: the vault ZIP gives you base.md plus a notes/ folder. Heptabase's help centre describes the export, not a Markdown import, so I would not count on one.

Heptabase and AI notes, briefly

Is there a free Heptabase plan?

The pricing page lists three paid plans and no free tier. The home page offers a 7-day trial.

Does Heptabase have AI?

Yes: an AI agent and monthly AI credits on every plan, plus an AI tutor on Premium.

Can SumizAI show AI notes on a whiteboard?

No. There is no canvas, no graph and no mind map. Notes sit in chapters.

Which keeps data closer to me?

The vault is a folder of .md files on your Mac plus a local index, and no server keeps a copy. Heptabase works with an account and sync, and you can export everything as Markdown.

Can I use both?

Yes. Research maps in Heptabase, answers from long AI chats in the vault. Export the vault and drop single notes onto a whiteboard when you need them.

Heptabase: heptabase.com (platforms, trial, features), heptabase.com/pricing (plans, AI credits) and the help centre article on export, read on 23 October 2026. The left column follows the app's current build.

Other comparisons

A canvas for maps, a vault for answers

Try the vault free for seven days. Keep Heptabase for the whiteboards.