Why I moved my notes from Notion to Obsidian

Aug 21, 2026 · 6 mins read · #obsidian #notion #knowledge-management #privacy #rag #self-hosting #cost #ai

I ran my whole knowledge base in Notion for years. I moved it all to Obsidian — 4,500+ pages, databases, links, attachments — and I'm not going back. Here's the honest why. 📝

The short version

I moved from Notion to Obsidian because four things matter to me more than Notion's convenience: privacy, a RAG that actually searches my notes, automation and AI interaction, and cost. Notion is a great product. It's just not the right home for my notes anymore.

Privacy: the notes are mine, and only mine

This is the one that tipped it. With Notion, my notes live on Notion's servers. Everything I write — work context, client details, personal notes, health data — sits on a third party's computer, and any AI feature I use sends content to their models.

Obsidian is the opposite: every note is a plain Markdown file on my own machine. There's no server, no account, no "trust us" — the data simply never leaves my hardware. When I want it on another device, I sync it myself (P2P, encrypted) instead of handing it to a cloud.

The strongest privacy guarantee isn't a policy. It's the absence of a third party.

For someone who works with client and lab data, that distinction stopped being theoretical. It became the default.

The RAG: my notes became actually searchable

This is the part that surprised me most. Notion has search, but it's keyword search — it finds the word you typed, not the idea you meant. And its AI search over "your whole workspace" is a black box that sends your content to their models.

In Obsidian, because my notes are plain text files, I could build a RAG (retrieval-augmented generation) on top of them — the same targeted RAG approach I use for application notes, applied to my whole vault.

The index is local and free: embeddings run on my own machine, so the vectors never leave the box. It's hybrid — keyword and semantic search, in French and English — so I can ask "what did I decide about X?" and get the right note even when I don't use the exact words. And it's always fresh: a cron re-indexes incrementally every few minutes, so a note I just wrote is findable immediately.

The result: my notes stopped being a place I stored things and became a place I find things. That changed how I use them.

Automation and AI interaction: it's mine to wire up

Notion's AI is a feature you rent. Obsidian is a platform I own, so the AI layer is something I build and control.

Chat over my vault — a plugin gives me RAG chat with cited sources [[...]], using whatever model I choose.
Meeting notes — I record a call locally, it's transcribed and turned into a structured note automatically. No vendor lock-in to a specific meeting platform.
My own agent — my self-hosted assistant (why I self-host) reads the vault directly and combines it with web search. I can ask it to think across my notes and the web, and drop the result back into the vault.
Scripts and pipelines — because everything is files, I can script anything: digest books into the vault, index PDFs, generate summaries. Notion's API can't touch that level of control.

The pattern is the same everywhere: I decide how the automation works, what model runs, and where the data goes. Notion decides those things for you.

Cost: less than Notion Business, with more

Notion Business runs $20 per user per month (billed annually). For a single user that's $240 a year — and the AI features are volume-limited on top of that.

My Obsidian stack costs me about $20 a month total, and that's the whole AI budget, not a per-seat fee:

Item	Cost
Obsidian app	$0 (free)
RAG embeddings (local)	$0
Meeting transcription	$0 (free tier)
AI model access (OpenCode Go ×2)	$20/mo
Total	≈ $20/mo

The two OpenCode Go subscriptions are the whole AI budget. For my usage, a single subscription is a little tight — with two, I can use the AI without any real limits. Everything else in the stack is free.

The key difference is what that $20 buys. Notion Business gives you a fixed per-seat price with limited AI volume. My $20 gives me a monthly allowance of model usage that I can't realistically saturate — roughly 150,000 RAG questions, or thousands of meeting summaries, or hundreds of full document reviews. And the models I use are the efficient ones, not a proprietary black box.

Same price as Notion Business, roughly six times the AI volume, and the data stays on my hardware.

Choosing the models

Because the AI layer is decoupled from the notes, I can pick the right model for each job instead of being stuck with one vendor's AI:

Efficient general work (chat, RAG, summaries) — a cost-efficient model with a huge context window, so even a full document fits in one pass.
Vision (scanned pages, images) — a separate vision model.
Sensitive work — a local model on my own GPU, so nothing leaves the machine at all.

The rule is the one I use everywhere: pick the smallest model that does the job well, keep sensitive data local, and let the rest go where it's cheapest and best. Notion doesn't give you that choice — you get their AI, on their terms, at their price.

What I gave up

Honest trade-offs, because it's not all one-way:

Notion's polish. Its databases and views are slicker out of the box. Obsidian's Bases plugin is close, but it took setup.
The migration itself. Moving 4,500+ pages, databases, and links took real work — the API import, rebuilding views, verifying nothing was lost.
A little maintenance. Self-hosting the sync and the AI layer means occasional upkeep. That's the price of control.

None of it was a dealbreaker. The payoff — my notes on my hardware, searchable by meaning, wired to the AI I choose, for less than Notion Business — was worth every hour of the move.

In short

I moved from Notion to Obsidian for privacy (my notes never leave my machine), a RAG that searches my notes by meaning instead of keyword, automation and AI interaction I control end to end, and a cost that matches Notion Business while giving me far more AI volume and better model choice. The trade is a little setup and maintenance — which I consider a fair price for owning the thing I rely on daily.

← Back