Heptabase Review 2026: Visual Notes, AI Credits, and Fit

Short answer

**Shortlist Heptabase for researchers and learners who organize complex reading visually and will check ai explanations against source material.** Avoid it for buyers seeking an ongoing free plan or a turnkey systematic-review screening system. Start with your actual bottleneck; an existing notes workflow may already be enough.

This is a **research-based review**. Numerical ratings are withheld: **Not rated**. DiscoverAI has not conducted the buyer test below, measured time saved, or independently validated AI answer quality.

What the product documents

Heptabase centers on cards and whiteboards for reading and understanding, alongside AI learning tools. It describes offline access and collaboration. Its pricing help distinguishes manual PDF highlighting from parsing; parsing does not consume AI credits. Pro includes Gemini-series agent access, while the higher plans include supported OpenAI and Anthropic models. New provider models are not automatically available.

Pricing and plan boundaries

Checked October 5, 2026: standard Pro is $11.99 monthly or $107.88 yearly; Premium $23.99 monthly or $215.88 yearly; Premium+ $71.99 monthly or $647.88 yearly. Early-bird Pro is separately priced. Included monthly AI credits are 100, 1,800 and 8,100 respectively. Taxes may apply. Trial ends after seven days or when trial credits run out, whichever is first.

Do not combine annual monthly equivalents with month-to-month subscriptions as if they were identical commitments. Review the selected plan, renewal, taxes, usage allowances and cancellation terms at checkout. A trial is a bounded evaluation, not a permanent free entitlement.

Fit for a small team or independent worker

Visual organization requires a maintained reading practice; more AI credits do not establish better learning.

Choose one job: recovering a source, organizing a reading question, or drafting from existing notes. Keep a baseline in the software you already use. Pay only when the new workflow solves that job with acceptable correction effort and manageable data handling. For nonprofits, use public materials for a first experiment; donor and beneficiary records require a separate approval process.

Privacy and portability checks

Review the linked privacy documentation and terms before uploading confidential material. Check which content leaves the device, which AI providers receive it, who can view shared material, and what deletion does to backups or downstream clients. We do not infer a product-specific deletion deadline or compliance certification where this review has not established one.

Test export with your own harmless notes and attachments. A visible export button is not proof that links, formatting, source references and permissions survive migration. Preserve a recoverable copy before importing or changing sync settings.

A reproducible buyer test

Use three public papers on one narrow question. Make ten evidence cards with page references, arrange supporting and conflicting findings separately, and ask the AI to explain two disagreements. Check claims against the papers. Close the app and later reconstruct the answer without AI. Record missed caveats, source-recovery time, AI credits, and a backup/export check. Compare with the same task in your current reading workflow.

Write acceptance criteria before starting: answers must have correct supporting material, uncertainty must remain visible, and sensitive content must stay within the chosen boundary. Count wrong answers, unsupported confident statements, broken references, missing caveats, review minutes and total cost. Keep the test inputs and outputs so a later plan or model change can be checked against the same tasks.

Alternatives and decision

[NotebookLM](/tools/notebooklm) is a relevant source-oriented alternative to investigate; [Notion](/tools/notion-ai) may suit readers already keeping work there. These are comparison candidates, not a claim of equivalent features or verified head-to-head superiority. Use the [Research Benchmark Center](/research-insights/benchmarks) for citation and thematic-analysis methods, and [Tool Finder](/tool-finder) to refine the task.

Start with one reversible trial. Keep your current workflow if source retrieval, privacy or usable output does not improve enough to justify maintenance and spend. A larger graph or more generated text is not the outcome: a supported decision you can recover later is.