Heptabase

Most note-taking apps optimise for the moment of capture: get the thought out of your head and into the system with as little friction as possible. Heptabase optimises for what happens after — the long stretch where you are trying to understand a field, not just record it. That difference shows up everywhere in the product. The core is a whiteboard-and-cards model. You clip a card from a PDF, a YouTube transcript, a journal entry, or the web clipper, and then arrange it spatially. Because the cards support bi-directional links, a card you write on Wednesday can point back to a highlight you took from a paper three months ago. For literature review work this is genuinely better than a linear document, because you can lay 50 papers out by argument, see the clusters, and spot the gap. The AI Tutor is the part worth evaluating carefully. It is not a chat box bolted onto the sidebar. Given your sources, it produces a course syllabus with topics and sub-lessons — the homepage demo shows a Western philosophy course structured from the Pre-Socratics through Aristotle — then it delivers each lesson as a page, tracks progress ("Lesson progress 0/4 completed"), and responds to you mid-lesson. If you have ever tried to self-teach a dense subject from a stack of PDFs, you know the hardest part is sequencing: what do I read first, and what am I supposed to have understood before moving on. That is the problem this feature solves, and it solves it with your material rather than a generic curriculum. The AI agent handles a different job: asking questions across all your sources at once, with answers that cite back to the source page. Combined with Zotero for reference management, that makes Heptabase a credible tool for a graduate student assembling a literature review. The CLI is the sleeper feature. Heptabase CLI works with coding agents such as Claude Code and Codex, which means your knowledge base becomes addressable from a terminal session. For developers who keep research documentation next to their code, that is a real workflow rather than a novelty. Where it strains. The credit system is the main friction. Pro gives you 100 AI credits/month, Premium gives 1,800, and Premium+ gives 8,100 with a 33% discount on additional credits. Heavy AI use will push you up a tier, sometimes faster than you expect, and the model menu is tied to your plan: Pro is Gemini models only, while GPT and Claude appear only at Premium and above. Collaboration is deliberately thin — the pricing page lists "Unlimited collaborator invites" in Pro, so you can invite people, but this is not a shared project-management workspace and does not try to be one. The honest read: Heptabase is priced like a serious tool for one serious person. If you are self-teaching philosophy, history, or a technical subject from primary sources, the AI Tutor at $17.99/mo billed yearly is cheaper than a tutor and more structured than a chat window. If you want frictionless capture, shared team