1. LandOffer
  2. Blog
  3. All articles
  4. Heptabase Review: Visual Research Boards, Source Notes, and Long-Term Portability

Productivity & Workflows

Heptabase Review: Visual Research Boards, Source Notes, and Long-Term Portability

Heptabase is a candidate for visual source comparison. This review builds a complete fictional research argument and a portable fallback packet, then separates current plan costs, AI credits, exports and recovery evidence.

9 min readLandOffer team

Heptabase is worth evaluating when your research problem is the relationship between sources: which claim supports a decision, which document contradicts it and which question remains open. Its cards and whiteboards offer a visual way to arrange that argument. The buying decision also needs an exit check: readable notes in another application are valuable, but they are not the same as preserving the board, attachments and recoverable Heptabase state.

This LandOffer.ai review checks public documentation on October 27, 2026. The five-source research board and portability packet below are original fictional materials. Local file checks examine an authored fallback structure, not files exported by Heptabase. No account, trial, PDF upload, AI conversation, shared board, subscription or restore was used. The evaluation assumes one researcher considering the desktop app on Windows 11; actual app performance and device behavior remain untested.

A cartographer holds a loose red thread beside five observation sketches connected across a harbor map

Decide whether a board adds a useful argument

The official Fundamental Elements guide explains that cards belong to the Card Library, rather than being owned by a particular whiteboard. A card can appear on more than one board. Sections, connections and sub-whiteboards provide ways to arrange the material. When one source informs several decisions, place it in each relevant research context without treating its repeated placement as a newly authored source.

A useful board lets you reconstruct a conclusion. Put the source's date, scope and status beside its claim, then write what the connection means. Two cards next to each other do not explain whether one supersedes the other or merely addresses a different audience. Attractive grouping can conceal that distinction as easily as a poorly titled folder.

The vendor's organization guide recommends starting with two whiteboard layers and using sections before adding deeper nesting. That is guidance, not a mandatory architecture. For this review's seven-note case, one project board with three sections is enough: requirements, conflicting context and proposed decision. Adding several nested boards would create navigation work without resolving another question.

If your source list is short and linear, a normal document with headings and links may already do the job. Heptabase's specific value to test is whether spatial comparison makes an unresolved relationship easier to inspect and maintain. This review does not establish that it improves comprehension, research speed or team decisions.

Compare subscription cost separately from AI consumption

Current plans and pricing list standard Pro at $11.99 monthly or $107.88 yearly, Premium at $23.99 or $215.88 yearly, and Premium+ at $71.99 or $647.88 yearly, in USD before applicable taxes. The annual amounts are billed yearly. Early-bird Pro is a separate existing-user price, not this new researcher's assumption.

Included monthly AI credits are 100, 1,800 and 8,100 respectively; unused credits do not carry over. Pro's provided AI Agent models are Gemini-series models, while higher tiers support listed providers. Parsing PDFs does not consume AI credits and currently has no cumulative monthly file/page allowance; individual parsing limits remain 350 MB and 3,000 pages per PDF. Parsing extracts text, including OCR, rather than creating Highlight cards. The documented subscription trial is seven days. These are vendor terms, not observed usage results.

For Elio, the fictional researcher, twelve separate Pro monthly payments would total $143.88; the listed annual commitment is $107.88. The $36 difference assumes unchanged pricing and a full twelve-month need. Treat that difference as subscription arithmetic, not evidence that the annual plan is the better purchase. One monthly period is the proposed cost assumption for an authorized evaluation, not a purchase made here.

The free-plan explanation describes invitation-based collaboration, with access limited to shared content and without the personal library, highlights or AI features. It is not an unrestricted free personal research workspace. Elio should not budget a zero-cost permanent personal system from the existence of free collaborators.

Finish a five-source board before judging the interface

Elio is researching a fictional warehouse product's packing-label instruction. The question is precise: should the operator interface and the outbound carrier label use the same weight unit? Every document, person and product detail in this example is invented. The entries below are authored source excerpts, not statements about an actual carrier, measurement standard or Heptabase feature.

Source card and provenance Authored claim and limitation Completed treatment on the board
S1 — product design draft, October 6, section 3 “Use kilograms throughout.” Draft, not approved; does not distinguish operator screen from carrier output Keep as an interface proposal. Connect to C1 with “draft scope narrowed,” not “final requirement.”
S2 — carrier-profile specification, revision 4, October 15, section 2 This fictional profile requires pounds in its outbound weight field Treat as the controlling output requirement for this profile only. It does not specify the operator display.
S3 — scale integration note, October 19, sample-field description Scale input is recorded in kilograms Preserve the input unit as evidence. A conversion boundary is required before the outbound field; no formula or implementation is verified here.
S4 — old operations instruction, revision 1, March 8 Operators should enter ounces manually Mark retired for the proposed workflow. Keep its history because it explains the conflicting training material.
S5 — operator discussion note, October 22, paragraph 2 A participant asks to see both kilograms and pounds Record a usability request, not a safety finding or approved product requirement. Connect it to Q1 for an explicit decision.

The board's completed synthesis note C1 reads: “Keep kilogram input and the proposed kilogram operator display separate from the profile-specific pounds output requirement. Do not reuse the retired ounces instruction. The dual-unit display request remains unresolved. Product owner Mara must decide that interface scope after reviewing the request; carrier integration owner Dev must confirm the approved profile reference before implementation.”

Q1 records the remaining question: “Would a dual-unit display help this operator group, and who accepts the extra interface and training scope?” It names S5 as the request's origin and S1 as the draft it would change. Its proposed next step is a November 3 scope review with Mara; no meeting was booked or product decision accepted. The answer is deliberately absent. That distinction lets a reviewer separate an approved output constraint from a requested interface change without counting the number of supporting cards as votes. A clean board should make an unanswered question visible instead of promoting the nearest convenient note into approval.

There are five source notes, one synthesis note and one open-question note. Six explicit relationships preserve the argument: S1, S2, S3 and S4 each inform C1 with different meanings; S5 raises Q1; Q1 conditions C1's final interface scope. The local fixture checks those seven note IDs, five stored source excerpts and six relationship endpoints. It does not establish that a Heptabase board was created or that its interface displays this layout correctly.

This is also a test of writing effort. If Elio cannot express those six relationships in words, moving the cards around is premature. If the written explanation already makes the decision easy to inspect, the later trial must show a useful visual advantage over that existing document rather than count the new board as an achievement by itself.

Keep AI assistance subordinate to the source question

The long-PDF guide recommends bounded page ranges for long documents and reviewing results before processing more. Its suggested starting range is not a universal context limit. For this case, a source-aware prompt would ask which of S1–S5 controls the operator display, which controls outbound data and where approval is missing. It should request source locations and preserve disagreements.

No answer is generated here. An evaluator would need to inspect whether an actual answer treats S1 as a draft, limits S2 to one carrier profile and leaves Q1 open. A fluent synthesis saying “all weights must be pounds” would fail the authored requirements even if it cited S2 correctly. A citation can identify a passage while the conclusion still extends beyond that passage's scope.

That is why Elio's initial purchase case is visual source comparison, not a promised number of AI answers. The included credit allowance does not translate into a fixed number of completed investigations. Record actual usage and source quality during a permitted evaluation before paying for a higher allowance.

Two hands hold a folding folio with harbor sketches and a connection index beside a rolled map and question tag

Check what remains usable outside the board

The official data-portability explanation describes Markdown note export and a complete export containing Heptabase backup data in All-Data.json. It warns that another application may not preserve layouts, relationships or metadata in the same way. Importing Markdown elsewhere and restoring Heptabase backup data are different operations. Keep the complete package and consult the recovery guidance rather than assuming a folder of readable notes can recreate everything.

Elio's second completed asset is an authored fallback packet for the same seven-note investigation. It is intentionally not labeled a Heptabase export. Its structure is notes/, source-excerpts/, relations.tsv and README.md. Each source note names the fictional document, revision/date and section; C1 and Q1 use relative links to the other notes. The source excerpts are plain-text fixtures, not exported originals or parsed PDFs.

Check on the authored packet Actual local outcome What a real Heptabase evaluation still must establish
Open seven note files and follow their relative note links Seven present; all authored note-link targets exist Whether the actual export preserves relevant content and translates application links usefully in the chosen receiving app
Trace each of five source notes to its stored excerpt Five excerpt files present and referenced Whether original PDFs, images and other required attachments are complete and readable
Read all six relationship records without the board Endpoints resolve; relationship meanings are written explicitly Which layout, connections, properties and metadata survive actual export or require separate documentation
Reconstruct the final status from C1 and Q1 Proposed unit separation and unresolved dual-display choice remain distinguishable Whether a real exported packet retains the correct current versions and enough context for another reader

This fallback makes a limited but useful result concrete: the authored research argument remains readable as ordinary files without relying on a board's coordinates. It says nothing about Heptabase's export fidelity. No All-Data.json is fabricated, and no restore or import is attempted.

The backup guide says automatic cloud backups preserve positions, states and metadata but exclude attachments. It directs users to manual export for attachment files and warns against relying on exports with missing images or broken image links. Synced data, a recovery snapshot and a verified exit packet therefore need separate checks.

Make the purchase depend on a repeatable exit check

Version history is another separate boundary. The history guide distinguishes whiteboard-layout recovery from card-content recovery and says tag/database property changes are not covered by card history. Its retention period is not a guarantee that every snapshot can be directly restored. Do not turn an undo feature into an unlimited recovery commitment.

Elio's provisional choice is to evaluate Pro against this seven-note packet before considering an annual term or higher AI tier. The pass condition is specific: an authorized trial must show a maintainable visual argument, correct source scope, a complete readable exit package and a documented recovery route. The current article supplies the argument and fallback packet; it leaves those product observations open.

Use the LandOffer.ai workplace-tool library after saving the evidence that would decide your own purchase: one conflicted research question, a written synthesis with unresolved points, a labeled price assumption and an export review. Have the permitted evaluator retain the original source repository and record any missing files, links or context before considering a replacement. A useful board should strengthen an explanation you can still understand when the board is unavailable.

Sources and Further Reading