March 31, 2026 · 11 min read
Heptabase is genuinely well-designed. The whiteboard interface is clean, the card system is thoughtful, and for researchers who love manually arranging ideas like puzzle pieces, it works beautifully.
But a growing number of knowledge workers are quietly walking away from it. Not because it broke, but because it stayed still while their needs evolved.
The shift isn't about features. It's about a fundamental question: should your research tool be a place where you store thinking, or a place where you do thinking? For researchers who spend hours synthesizing papers, exploring competing ideas, and connecting dots across domains, that distinction matters.
This article breaks down what's driving people away from Heptabase, what they're looking for in an alternative, and why AI-first tools are increasingly the answer.
Heptabase earned its reputation. The spatial canvas model was a meaningful step forward from linear note-taking apps. Being able to see your cards arranged visually — clustering related ideas, building concept maps, linking journal entries to whiteboards — gave researchers real intellectual ownership over their material.
The core insight behind Heptabase is solid: thinking is non-linear, so your tools shouldn't force linearity on you.
The problem is that Heptabase stopped there.
Every connection in Heptabase is one you have to make yourself. Every card, every tag, every whiteboard arrangement — it's all manual. That's fine when you have ten cards. It becomes a job when you have three hundred.
Researchers who use Heptabase heavily describe a familiar frustration: the tool becomes a second project to manage. You spend time maintaining your system instead of using it. The overhead of keeping your knowledge base organized starts to rival the overhead of the actual research.
This is the classic PKM trap. The system becomes the work.
Heptabase is a visual container. It holds your thinking, but it doesn't participate in it. There's no AI layer that can help you surface connections you missed, challenge an assumption, synthesize a cluster of cards into a summary, or help you explore a concept you're stuck on.
For researchers who are increasingly using AI as a thinking partner — not just a writing assistant — this is a significant gap. You end up context-switching constantly: research in Heptabase, AI conversation in ChatGPT or Claude, then manually carrying insights back into your cards. The workflow is fragmented by design.
Cards and whiteboards are a metaphor borrowed from physical space. They work well for organizing what you already know. They work less well for exploring what you don't know yet.
Research is inherently exploratory. You follow a thread, hit a dead end, backtrack, branch in three directions at once. A card-based system asks you to commit to structure before you've earned it. That's backwards for anyone in the early or middle stages of a research project.
When people search for a Heptabase alternative, they're usually not looking for a slightly different card interface. They want something that solves the underlying frustrations.
Here's what that looks like in practice:
AI that's embedded in the workflow, not bolted on Not a separate tab. Not copy-paste. The AI should be part of how you explore and synthesize, not something you interrupt your work to go use.
Flexibility without overhead The tool should adapt to how you think, not demand that you adapt to it. Spatial organization is great — but it shouldn't require constant maintenance to stay useful.
The ability to explore, not just organize Research involves a lot of "what if I look at this from a different angle?" moments. A good alternative supports that kind of exploratory, branching thinking natively.
Context that travels with you When you're deep in a research thread, you shouldn't have to re-explain your context every time you open a new conversation or switch tools. Your sources, your prior thinking, your files — they should all be available where you need them.
Model flexibility Different AI models have different strengths. GPT-4o is strong on reasoning. Claude is often better for nuanced writing and long documents. Gemini handles certain technical domains well. Researchers who've been using AI long enough know this — and they want to choose the right tool for the right question without leaving their workspace.
The category of "AI-first knowledge tools" is still young, but it's attracting serious researchers for a clear reason: they're built around a different assumption.
Traditional PKM tools — Heptabase, Roam, Obsidian, Notion — assume that the human does the thinking and the tool stores the output. AI-first tools assume that thinking is a collaborative process between the human and the AI, and they're designed to support that loop from the ground up.
That's not a small philosophical difference. It changes everything about how the tool is structured.
RabbitHoles AI is built on an infinite canvas — familiar territory for Heptabase users — but the similarity ends there. Where Heptabase gives you cards to fill, RabbitHoles gives you a thinking environment that actively participates in your research.
Here's what makes it a meaningful alternative:
The core unit in RabbitHoles isn't a card — it's a chat node. You can start a conversation with an AI model, then branch it in multiple directions from any point. Each branch becomes its own thread, but they all live on the same canvas, spatially connected to their origin.
This maps directly onto how research actually works. You're exploring a question about attention mechanisms in transformer models. One branch goes deep on the mathematical formalism. Another explores practical implications for long-context tasks. A third compares different architectural approaches. All of it is visible, navigable, and connected — without you having to manually build that structure.
The branching happens naturally as you explore, not as a filing task you do after the fact.
This is a feature that sounds like a nice-to-have until you've used it. Being able to switch between AI models within the same research session — without losing context, without starting over — is genuinely powerful.
You might start a thread with Claude because you're working through a dense academic paper and want careful, nuanced interpretation. Then switch to GPT-4o when you need to reason through a methodological question. Then pull in a different model for a specific technical domain. The canvas holds all of it together.
For researchers who've developed strong intuitions about which models are better at which tasks, this flexibility is a significant unlock.
You can add PDFs, documents, and websites directly as context sources within RabbitHoles. This means your research materials travel with your thinking. When you're working through a paper, you're not copying excerpts into a chat window — the document is part of the conversation.
This solves one of the most tedious parts of AI-assisted research: the constant re-contextualization. Every time you open a new ChatGPT window, you start from zero. In RabbitHoles, your sources persist across your canvas, available to any node that needs them.
The infinite canvas in RabbitHoles isn't just decorative. Because your conversations branch and grow spatially, the layout of your canvas actually reflects the structure of your thinking. You can see at a glance which threads you've explored deeply, where you branched, which areas are still underdeveloped.
This is spatial organization that emerges from your work, rather than spatial organization that you impose on your work before you've done it. That's a subtle but important distinction for anyone who's felt the friction of Heptabase's manual arrangement model.
| Feature | Heptabase | RabbitHoles AI |
|---|---|---|
| Visual canvas | ✓ | ✓ |
| Card/node-based structure | Cards (manual) | Chat nodes (dynamic) |
| Native AI integration | Limited | Core feature |
| Branching conversations | ✗ | ✓ |
| Multi-model support | ✗ | ✓ |
| File/website context | Limited | ✓ |
| Spatial organization | Manual arrangement | Emerges from exploration |
| Best for | Organizing existing knowledge | Exploring and synthesizing new knowledge |
The honest summary: Heptabase is better if your primary workflow is organizing and reviewing knowledge you've already processed. RabbitHoles is better if your primary workflow is active research, exploration, and synthesis — especially if AI is already part of how you work.
The researchers moving from Heptabase to AI-first tools tend to share a few characteristics:
They're heavy AI users. They've already integrated ChatGPT, Claude, or other models into their daily workflow. They're not skeptical of AI — they're frustrated by how poorly it integrates with their knowledge tools.
They work with complex, multi-threaded topics. Literature reviews, competitive analysis, technical deep dives, cross-domain synthesis — work that involves holding many threads simultaneously and following them in non-linear ways.
They've hit the maintenance ceiling. Their Heptabase workspace has grown to the point where maintaining it is a real time cost. They want a tool that stays useful without constant gardening.
They think in conversations, not documents. For many researchers, the most productive thinking happens in dialogue — whether with a colleague or an AI. A tool built around conversational nodes feels more natural than one built around static cards.
RabbitHoles isn't the only option. Depending on your specific needs, these alternatives are worth considering:
Obsidian — The gold standard for local-first, markdown-based PKM. Highly extensible with plugins, including AI plugins. Better for people who want full control and don't mind configuration overhead. Less visual than Heptabase.
Notion AI — Familiar interface with AI features layered in. Good for teams and structured projects. Less suited for exploratory, non-linear research.
Mem — AI-native note-taking that auto-organizes your notes. Strong on capture and retrieval. Less visual, less suited for spatial thinkers.
Roam Research — The original graph-based PKM. Still has a loyal following. Steep learning curve, no native AI, but powerful for networked thinking.
Napkin AI — Focused on visual thinking and idea organization with AI assistance. More of a presentation/communication tool than a research tool.
For researchers who want the spatial canvas they're used to from Heptabase, combined with genuine AI integration and branching exploration, RabbitHoles is the most direct match.
If you're coming from Heptabase, the mental model shift is worth preparing for.
In Heptabase, you build structure first, then fill it. In RabbitHoles, you explore first, and structure emerges. That's not worse — but it's different, and it can feel disorienting for the first few sessions if you're used to the card-based approach.
A few things that help:
What's driving the move away from tools like Heptabase isn't dissatisfaction with any particular feature. It's a broader shift in how researchers think about their tools.
For most of the PKM era, the ideal was a perfect system — a well-organized, comprehensive second brain that captured everything and made retrieval effortless. Tools were evaluated on how well they supported that vision.
AI changes the calculus. When you have a capable AI available, perfect organization matters less. You don't need to find the exact card — you can ask a question and get a synthesized answer. The value of a knowledge tool shifts from storage and retrieval to exploration and synthesis.
Tools built for the old model — even good ones like Heptabase — are increasingly mismatched with what researchers actually need. Tools built for the new model are winning users not because they're flashier, but because they fit the actual workflow better.
Heptabase solved a real problem. Spatial, visual knowledge organization was a genuine improvement over linear notes, and it still has a place for certain workflows.
But if you're a researcher who uses AI regularly, works with complex multi-threaded topics, and has started to feel like your PKM tool is working against you instead of with you — the frustration is legitimate, and the alternatives have caught up.
RabbitHoles AI is built for exactly this moment: the point where AI becomes a genuine thinking partner, not just a text generator, and where your research tool needs to support that kind of collaborative, exploratory, branching work.
If that sounds like where your work is heading, it's worth seeing what a canvas built around conversations can do.
Learn more at rabbitholes.ai