Why this comparison matters
Siyuan and Atomic are both open source, and both keep your notes on hardware you control. Beyond that they have very little in common. SiYuan is a block editor for people who want to structure their knowledge by hand, with block references, databases, flashcards and a graph view. Atomic is for people who would rather not organise anything. You drop notes in, it tags and embeds them, and then you search the pile by meaning or chat with it.
The real choice is who does the filing. In SiYuan you do it yourself. In Atomic a language model does it for you. The rest of this comparison follows from that difference.
The facts below come from each vendor's site and GitHub repository as of September 2026. SiYuan has about 46.5k GitHub stars and uses the AGPL-3.0 licence. Atomic uses the MIT licence and has about 1.6k stars. That gap reflects how long each has been around: SiYuan has shipped for years, and Atomic is still early software.
Feature comparison
| Feature | SiYuan | Atomic |
|---|---|---|
| Licence | AGPL-3.0 | MIT |
| Desktop | Windows, macOS, Linux | Windows, macOS, Linux (GitHub releases) |
| Mobile | Android, iOS, HarmonyOS | iOS, plus a browser client |
| Storage format | .sy files (JSON) in a local workspace | Markdown "atoms" in SQLite with sqlite-vec embeddings |
| Editor | Block editor with 20+ block types | Markdown notes, chunked automatically |
| Linking | Manual block-level bidirectional links, backlink panel, graph | Automatic semantic connections, spatial canvas |
| Organisation | You build it: notebooks, tags, references, databases | Auto-tagging of everything you add |
| Search | Full-text and SQL-style queries | Semantic search by meaning |
| AI features | Writing help, translation, summaries, Q&A chat | Agentic chat with citations, wiki synthesis from tagged notes |
| AI providers | Any OpenAI-compatible API (bring your own key) | OpenRouter, Ollama, any OpenAI-compatible API; managed AI on Atomic Cloud |
| Flashcards | Yes, FSRS spaced repetition | No |
| Databases | Yes, with relations, rollups and templates | No |
| MCP server | No native server (agent support via plugins) | Yes, works with Claude and Cursor |
| Sync | End-to-end encrypted, via official cloud or your own S3/WebDAV | One server acts as source of truth for all clients |
| Self-hosting | Official Docker image (b3log/siyuan), Kubernetes | Docker Compose, Fly.io, or a standalone Rust binary |
| Extras | Plugin and theme marketplace | RSS ingestion, browser extension, API |
| Our rating | 7.5 / 10 | 7.2 / 10 |
Where SiYuan is stronger
SiYuan's biggest advantage is linking at the block level. You can reference, embed or query one paragraph from another document, and the backlink panel shows every place that paragraph is used. Most PKM tools link whole pages, so this finer control matters once your notes run into the thousands. It also handles very large documents, into the millions of characters, by loading them on demand. It includes FSRS flashcards and Notion-style database blocks, so you may be able to drop two other apps.
The costs are real, though. The documentation and community are mostly Chinese-first, and English guides lag behind. English speakers will find it harder to learn than Obsidian. The mobile apps work but feel less finished than the desktop app.
Where Atomic is stronger
Atomic is built around AI in a way SiYuan is not. In SiYuan, AI is a writing assistant attached to the editor. In Atomic, AI is how you find and connect things. Every note is embedded and tagged when you add it. You can then ask a question and get an answer with citations, or have it write a wiki page from everything under one tag. The MCP server lets Claude or Cursor read your knowledge base directly, which is the most useful feature here if you already work inside those tools.
It also supports Ollama. You can run the whole system, embeddings and chat included, without anything leaving your machine. SiYuan can reach a local model through an OpenAI-compatible endpoint, but its AI features are much less central to the app.
Atomic's weak points come with being young. It has a short track record, few integrations beyond MCP and RSS, and self-hosting means running a three-service Docker Compose stack rather than opening a single app.
Pricing comparison
SiYuan
- Free, $0: every local feature, including the editor, block links, databases, flashcards, AI with your own key and Docker self-hosting. Commercial use is allowed.
- PRO features, $64 one-time (list price $96): adds sync through your own S3 or WebDAV storage. You pay the storage provider separately.
- Subscription with 8GB official cloud: the pricing page currently shows $148 lifetime (list price $296), marked "seats are limited". It includes all PRO features plus official sync and backup, cloud asset hosting and Cloud Inbox.
No SiYuan plan includes AI credits. You pay your model provider directly, or run a local model.
Atomic
- Self-hosted, free: the full MIT-licensed server and apps. Your only cost is the AI provider, or nothing if you use Ollama.
- Atomic Cloud Free, $0/month: 250 atoms, 1 knowledge base, 100 MB storage and $0.50 a month of included AI. You get search, wikis, chat, canvas, the extension, iOS, API and MCP access.
- Atomic Cloud Pro, $12/month: unlimited atoms and knowledge bases, 10 GB storage, $10 a month of included AI and access to frontier models for chat and reports. New accounts get a 14-day Pro trial with no card required.
The iOS app connects to either your own server or Atomic Cloud. The site shows no separate price for it. Atomic also says data is never deleted for non-payment, and you can export to Markdown or move to a self-hosted server with migrate push.
What it costs over three years
If you self-host, both are free apart from the server and any AI usage. For the hosted options, SiYuan's most expensive plan is $148 once, while Atomic Cloud Pro comes to $432 over 36 months. That comparison flatters SiYuan a little, because Atomic's $12 includes $10 of AI usage each month and SiYuan includes none. If you would spend $10 a month on API calls anyway, the gap mostly disappears.
Use case scenarios
Pick SiYuan if you write and structure long-form work
Thesis research, technical documentation and course notes all benefit from block references and embedded queries. If you already think in outlines and linked paragraphs, SiYuan gives you more control than anything else in this category.
Pick SiYuan if you study with spaced repetition
The built-in FSRS flashcards work on any block in your notes. Atomic has nothing equivalent, so you would need Anki alongside it.
Pick SiYuan if you need Android or fully offline mobile
SiYuan runs natively on Android and iOS with local data. Atomic currently offers iOS and a browser client, and both depend on a reachable server.
Pick Atomic if you capture a lot and never file it
If your notes are a mess of clipped articles, meeting scraps, RSS items and half-finished ideas, Atomic's auto-tagging and semantic search will get more out of them than a manual system you never keep up.
Pick Atomic if you work inside Claude or Cursor
With the MCP server, your coding assistant or chat model can search and cite your notes without copying and pasting. For developers this matters more than any editor feature.
Pick Atomic if you want to build on it
The MIT licence lets you embed or fork Atomic in a product with few conditions. SiYuan's AGPL-3.0 requires you to release your source code if you offer a modified version over a network.
Verdict
As a primary notes app, Siyuan wins, with a 7.5 against Atomic's 7.2. It is more mature and offers more editing power. Its encrypted sync is private by design, and a one-time price makes it cheaper over time than almost any rival. Expect a steeper learning curve if you don't read Chinese.
For a large pile of unsorted notes, Atomic wins. Semantic search, cited chat and MCP access do things SiYuan's writing assistant was never built for. It is early software, so keep Markdown exports and don't rely on it as your only copy yet.
If you can't decide, ask whether you enjoy organising your notes. If you do, choose SiYuan. If you know you won't, choose Atomic.
