Heptabase is a visual knowledge base built for students, researchers, and lifelong learners. You think on whiteboards: card notes, sections, mindmaps, annotated PDFs with full OCR parsing, YouTube and podcast transcripts, and an in-app AI agent that creates and edits cards under diff review. Heptabase shipped 1.0 in September 2023 and has released near-weekly since. It runs on macOS, Windows, Linux, iOS, Android, and the web, and its homepage claims search over 10,000 notes in under a second. There is no free plan: tiers run $11.99 to $71.99 a month ($8.99 to $53.99 billed yearly) after a 7-day trial.
Acres is a local-first, infinite-canvas note app for macOS and Windows. Cards are Markdown files with YAML frontmatter, boards are readable JSON, and everything lives in a folder you pick — the app is a view over your files, not their container. The app is a one-time purchase — a 14-day trial, then $59 once — with no account and no feature gates. It is built for people who think spatially, and for people whose agents — Claude Code, Cursor, Codex — do real work in their notes.
Both products put cards on a spatial canvas, and both — still rare in this category — ship official agent access. The disagreement is underneath: Heptabase holds your notes in a synced database and builds doors into it; Acres keeps your notes as files and gets out of the way. Most of the practical differences below follow from that one decision.
Where do your notes actually live?
In Heptabase, notes live in the app’s internal database, cached locally and synced through Heptabase’s cloud on AWS, encrypted in transit and at rest. Offline access is genuine — the homepage promises all notes and files without a connection — but there is no folder of documents another tool can open, and Heptabase’s docs are explicit that the database files are off limits. Getting data out means the official Markdown export or the JSON backup, and the two exits are not equal. A third-party review by The Effortless Academic found the Markdown export writes links as absolute Heptabase URLs rather than wiki-links; whiteboard layout survives only in the proprietary JSON backup. The full-fidelity copy of a Heptabase knowledge base is the JSON backup, not the Markdown export.
In Acres, the file is the note. cards/topic.md opens in any editor; boards/project.board.json is pretty-printed JSON you can read and script against. You can grep your vault, put it in git, back it up with whatever you already use. Nothing to export, because nothing was ever locked in. If plain files are your deciding factor, the nearest neighbor is Obsidian — see Acres vs Obsidian Canvas.
How do agents and other tools reach your notes?
Credit where due: Heptabase took agent access seriously earlier than most of this category. It shipped an official MCP server in December 2025 — read-write, OAuth, nine tools, working with Claude, ChatGPT, and Cursor — and an official CLI in April 2026 that lets coding agents create, edit, and search cards and journals. Two constraints shape both. The MCP server is a cloud API at api.heptabase.com: your agent talks to Heptabase’s servers, not to your machine. The CLI runs locally but requires the desktop app to be running, and the docs instruct agents to touch Heptabase data only through official commands — never the underlying files.
Acres never needed doors; the vault itself is the interface. Every vault self-describes for agents — a generated AGENTS.md front door, a resolved knowledge graph in ai/graph.json, a seeded CLAUDE.md — and the bundled MCP server exposes 29 tools, with one-click setup for Claude Code, Cursor, OpenCode, Codex, and Claude Desktop and no Node install. Agents can edit while the app is open: every save three-way merges, so an agent’s change folds into the canvas around whatever you’re dragging. Each agent edit lands in a ledger, attributed and shown as a word-level diff, with per-change revert. And when the app isn’t running, the vault is still just Markdown and JSON. Anything that reads files can read your notes.
Both apps have official agent access. The dividing line is what has to be running, and who sits in the middle. The full Acres story is at /for-agents/.
How does the built-in AI compare?
AI is now the center of Heptabase, and the shipped surface is broad: Agent Mode creates and edits cards with diff review, AI Skills run from a slash command, semantic search spans the knowledge base, and the AI Tutor runs structured learning sessions. It is metered by monthly credits and gated by tier — Pro ($11.99/month) limits the agent to Gemini models with 100 credits, Premium ($23.99/month) unlocks GPT and Claude with 1,800 credits and the Tutor, Premium+ carries 8,100 credits.
Acres does AI without metering. Bring a key from any of seven providers — Anthropic, OpenAI, Google, OpenRouter, xAI, Mistral, DeepSeek — or run Ollama locally with no key at all; this is not a metered tier — your own models and keys. Keys stay in the system keychain and requests go straight to the provider. The in-app agent has 18 tools; destructive edits arrive as proposals with word-level diffs, and one agent run is one undo step. Agents can also grow a mindmap from an indented outline onto the open board — that pattern is covered at /ai-mind-mapping/. What Acres doesn’t have: a tutor, transcript ingestion, or PDF OCR — Heptabase’s learning stack is deeper today. The Acres bet is narrower and cheaper — your own models and keys, without the monthly credit math.
Can you use it on more than one device?
With Heptabase, yes, thoroughly. Every plan includes sync through Heptabase’s cloud, and the mobile apps are real working surfaces — capture, whiteboards, the AI agent, and since May 2026 full PDF annotation on iOS and Android.
Acres runs on macOS and Windows, and nowhere else — no mobile, no web, no Linux. The notes are plain files on disk, so you can sync the folder however you like, and an optional end-to-end encrypted sync service exists.
What happens if you stop paying?
Heptabase documents this plainly, and deserves credit for the candor: when your subscription ends, your notes become permanently read-only. You can view and export them indefinitely, but you cannot create or edit anything without an active subscription, and there is no free tier to land on. The company backs its side of the bargain with a sustainability page — profitable since year one, and a public pledge to open-source Heptabase if it ever shuts down.
Acres has a read-only state too, and it deserves the same plainness: after the 14-day trial, the app is read-only until you buy — you can still open, search, read and export everything. The difference is what each read-only means. Heptabase’s is a subscription lapsing, over notes that live in Heptabase’s database. Acres’s is the pause before one $59 purchase, over Markdown files on your own disk — editable in a text editor, or by an agent, the whole time. Buy once and there is nothing left to stop paying.
Who should pick Heptabase?
Pick Heptabase if you study or research across devices. If your day involves mobile capture, annotating PDFs on an iPad, transcribing lectures or podcasts, pulling in Zotero or Readwise, or working a shared whiteboard with people who shouldn’t have to pay, Heptabase does all of that now, and does it with years of near-weekly polish behind it. It is the more complete product for a student or researcher who lives in it daily and is comfortable paying $107.88 a year or more for the privilege.
Pick Acres if you want your thinking to accumulate in files you own — bought once, offline, readable by every tool and agent you’ll use over the next decade. The trade is real: fewer integrations, desktop only, no collaboration. If plain files on a canvas is the argument that lands with you, Acres vs Logseq walks the same ground from the outliner side.
Switching
Acres ships a built-in Heptabase importer, and it reads the right file: not the lossy Markdown export, but the full All-Data.json from a Heptabase backup.
- In Heptabase, take a manual full backup — Settings → Backup & Sync → Export Now. The manual part matters. Heptabase’s automatic cloud backups store positions, states, and metadata but exclude attachments; only a manual backup includes your files and PDFs.
- Point Acres at the backup. Every whiteboard becomes an Acres board with the original geometry preserved — cards arrive in their exact places. (The importer’s acceptance gate was a real 15,240 × 10,809-unit Heptabase board with 88 connectors, required to render faithfully.) The whiteboard hierarchy is rebuilt as nested boards with breadcrumbs, Heptabase card colors remap onto Acres tints automatically, and highlights, PDFs, and the journal come along.
- Re-run it whenever. Imported items keep their Heptabase identity as
hb_-prefixed ids, so re-importing a newer backup upserts instead of duplicating — run it twice, zero duplicates. Trashed and empty whiteboards are skipped.
One honest limit: this is a one-time import of a backup, not a live connection. Edits made in Heptabase afterward don’t flow in; re-import a newer backup to catch up. And one relief: because Heptabase lets you view and export your data indefinitely after canceling, there is no deadline. You can move when you’re ready — and land in a vault that looks like the one you left, except every card in it is now a plain Markdown file on your own disk.