Every Reflect vs Capacities vs Tana comparison I’d read was written from a two-week trial and a feature spreadsheet. So I did the thing nobody bothers to do: exported 500 real notes — four years of project work, meeting notes, and book highlights — and migrated them into all three.
What I found wasn’t a winner. It was three completely different machines pretending to be the same product. The hard part wasn’t picking the best one. It was figuring out which one matched how my brain actually works.
The Quick Verdict (For People Who Hate Long Intros)
Reflect is for thinkers who write to remember. AI-first daily notes, chronological flow, your knowledge graph held together by backlinks. Best if your notes are a stream and you want AI as a writing partner.
Capacities is for researchers. Object-typed PKM where a Book is a Book and a Person is a Person, and the structure scales without choking. Best if your knowledge has shape — and you want it to still have shape in three years.
Tana is for systems builders. A node-graph layered with supertags that turn notes into databases. Best if you’d rather design a schema once than capture notes every day.
That’s the tidy answer. The migration is where it falls apart.
What Actually Survived the 500-Note Migration
Reflect: Markdown import worked cleanly. Backlinks survived. Dates mapped to daily notes automatically. The only loss was nested structure deeper than two levels — it flattens. Migration score: 9/10. I was writing in it within an hour.
Capacities: Import worked, but every note came in as an untyped “Note” object. Re-typing 500 notes into Person, Book, Project, and Concept took me three evenings on the couch. Backlinks survived. Tags became collections. After the work, it was beautiful. Before the work, it was 500 nameless cards. Score: 7/10, generous.
Tana: The painful one. Notes imported as nodes, but supertags don’t apply themselves — you have to retroactively schema 500 nodes or accept that half your library is invisible to structured queries. I gave up at note 180. Score: 4/10, before I’d really started using it.
If you’ve moved between Notion AI, Obsidian, and Mem, this story will sound familiar — except those tools share enough mental model that migration is mostly mechanical. These three don’t. The shape of your data changes between them.
So Reflect won the migration. But the migration isn’t the product. The product is what the AI does once your notes are inside.
The Connection Test: Which AI Actually Finds Links You Didn’t Write
I picked 10 notes from my library and asked each app’s AI the same question: “What else in my notes is related to this?” I wasn’t looking for matches I’d already linked. I was looking for the connections I’d forgotten existed.
Reflect found 7/10. Its AI reads across the graph and surfaces semantic links between notes from different years — similar to NotebookLM’s grounded retrieval from your own sources, but across your personal graph instead of uploaded documents. Twice it pulled up notes I’d genuinely forgotten I’d written. The output was a flat list, but the list was right.
Capacities found 5/10 — but with reasoning. When it surfaced a connection, it told me why: shared object type, shared property, semantic similarity. Fewer hits, more explainable hits. For research work where I needed to defend a citation, that mattered more than raw count.
Tana found 8/10 — the highest score — but only on the 180 nodes I’d retroactively schema’d. The other 320 were invisible to its structured queries. The AI was incredible on the part of the library I’d taught it to read. Useless on the rest.
The pattern is uncomfortable: Reflect rewards laziness, Capacities rewards typing discipline, Tana rewards schema work. AI quality is roughly proportional to the work you put in upfront. There is no free lunch — only different bills.
Which brings up the bill.
The Real Cost: What You Pay Once AI Is Included
Reflect: ~$10/month, AI included. No add-ons. Two weeks of heavy use and I never saw a cap warning.
Capacities: Free tier covers casual use; Pro is ~$10/month with AI features included. Cheapest of the three for real PKM work.
Tana: Plus is ~$10/month, but the AI features — meeting agent, AI search, the things that make Tana feel like the future — are a separate add-on that meters by usage. My two-week test billed at roughly $22/month equivalent once everything I cared about was switched on.
The real tradeoff isn’t dollars. It’s predictability. Reflect and Capacities give you one bill. Tana gives you a base bill plus a usage bill that scales with how much you actually use the thing you bought it for. If you’re already careful about LLM costs the way you’d be with token accounting on direct API use, this won’t surprise you. If you’re not, it will.
Money sorted, schema effort sorted. The only question left is the one I couldn’t answer on day one.
The Two-Week Verdict: Which One I Actually Reached For
Reflect won my mornings. Daily notes, capture, AI summaries of yesterday — it became the front door to my thinking within four days. I stopped thinking about it. That’s the highest compliment a tool gets from me.
Capacities won my project work. Anything with structure — a client engagement, a book I was reading, a person I was tracking — lived there because objects made it findable six months later. The typing tax paid me back the first time I searched “everything I know about [specific client]” and got a clean answer — and paired with the research tools I actually use daily, it’s become the backbone of how I prepare for client work.
Tana won nothing for me. But I’d recommend it to one specific person: someone whose work is the schema. Consultants building reusable frameworks. Researchers with citation graphs. People who’d rather design a system once than capture notes every day. If that’s you, Tana isn’t expensive — it’s the only thing that does what you need.
Honest admission: I’m running Reflect and Capacities side by side now. The dream of one tool didn’t survive contact with real data.
Who should NOT use each: skip Reflect if you need typed structure or Android. Skip Capacities if you want AI as a writing partner more than as a librarian. Skip Tana if you’d rather take notes than design schemas.
The Bottom Line
These three aren’t really competitors. They’re three answers to a question most note-app reviews never ask: how do you want the AI to know your notes?
Reflect makes the AI a fast reader of whatever you write. Capacities makes it a careful librarian of whatever you’ve typed. Tana makes it a query engine on top of a database you build yourself. Pick the one that matches the work you’d already do.
If you’re not sure which that is, start with Reflect for two weeks — it has the lowest migration cost if you change your mind. And take my advice: migrate 50 of your own notes, not 500. The right tool announces itself by day eight.