SiYuan’s most revealing signal is the gap between its shipping volume and its contributor count. ToolVitals records 791 GitHub commits in the last 30 days, 14 release events in that period, and 4 active contributors. That is a concentrated development model, not a broad community swarm. The recent output also has a clear direction: SiYuan is moving from a private, block-based notebook toward a self-hosted workspace where humans and AI agents work on structured knowledge together.
SiYuan’s product model is structured and local-first
The official SiYuan page centers the product on content blocks. In its description, a block is the basic unit used to organize, reference, move, and connect knowledge. The page lists more than 20 block types and more than 10 inline element types, then extends the model with block references, backlinks, graph views, outlines, databases, and plugins.
That choice matters. SiYuan is not presenting itself as a folder of Markdown files with a search box. Its block model is designed to preserve relationships while users split, reorder, recombine, and embed content. The official page says blocks can be moved or reorganized without breaking existing links. For a long-lived knowledge base, that is a more consequential design decision than a collection of formatting options.
The site also claims support for very large documents through dynamic loading, including million-word documents on ordinary computers. That is a first-party product claim, not an independent performance result, but it explains the target user. SiYuan is built for knowledge bases that grow past the size of a simple daily-notes folder.
The GitHub repository describes SiYuan as an open-source, privacy-first, self-hosted knowledge workspace where humans and AI agents work together. Its repository description highlights fine-grained block-level references, Markdown WYSIWYG editing, custom attributes, SQL queries, embeds, exports, multi-tab workflows, and AI writing and question-answering through the OpenAI API.
The privacy model is equally central. The official site presents data as stored on devices controlled by the user, with local use available when the network or a cloud service is unavailable. It also describes end-to-end encrypted synchronization across devices. Those statements establish SiYuan’s intended architecture. They do not, by themselves, establish that every deployment or synchronization configuration has been independently audited.
Self-hosting is not limited to a slogan in the repository. The GitHub documentation recommends Docker for server deployment and documents a serve subcommand with a mounted workspace directory. Since version 3.7.0, the documentation says that serve must be passed explicitly in Docker commands. It also documents PUID and PGID settings for permissions on host-mounted directories. Those details are useful because self-hosted software succeeds or fails on operational edges such as filesystem ownership, upgrade behavior, and workspace recovery.
ToolVitals classifies SiYuan as OSI-approved open source and lists AGPL-3.0. That is a different signal from fair-code, source-available, or open-core software. For teams that need to inspect, host, and adapt the application under an open-source license, the classification is relevant. It does not remove the need to review the license obligations for a particular distribution or hosted service.
The metrics show fast shipping and concentrated ownership
ToolVitals gives SiYuan a hot score of 256.5, a health score of 95, a shipping score of 98, and an overall ToolVitals score of 96. The project has 46,183 GitHub stars, 791 commits in the last 30 days, 30 GitHub releases in the last 90 days, 14 release events in the last 30 days, and 4 active contributors. ToolVitals assigns the data a confidence score of 90.
The combination is more informative than any single number. The commit and release figures show a project receiving sustained engineering attention. The shipping score of 98 is consistent with the visible release sequence, which includes a full v3.8.1 release followed by beta and alpha builds. The health score of 95 says that ToolVitals sees strong public project signals, while the overall score of 96 summarizes those signals into a high project rating.
The contributor count changes the interpretation. Four active contributors producing 791 commits is concentrated output. That can indicate a focused core team with strong product ownership. It can also indicate a narrow maintenance base, where knowledge about release engineering, synchronization, mobile behavior, and data migration sits with very few people. ToolVitals can measure the contributor count. It cannot determine which explanation is correct.
The release counts need similar care. A release event is not the same thing as a mature production version. The supplied events include v3.8.2-alpha.1 and v3.8.2-alpha.2, so the 30 releases in 90 days and 14 events in 30 days include development cadence as well as generally available release activity. A fast sequence can mean rapid correction and experimentation. It can also increase upgrade and regression risk for users who track every build.
Stars are another imperfect signal. SiYuan’s 46,183 stars show substantial public interest, but they do not reveal how many people use it weekly, how many operate it on a server, or how satisfied those users are. The ToolVitals scores also do not function as a code-quality audit. They summarize observable project activity and public evidence.
The comparison data makes that limitation clear. Among the related note-taking tools in the payload, Anarlog has a hot score of 204.4, 9,252 GitHub stars, a shipping score of 96, and 6 release events in 30 days. AFFiNE has a hot score of 203.0, 72,212 stars, a shipping score of 95, and 6 release events in 30 days. SiYuan has the higher hot and shipping scores in this comparison, but AFFiNE has more stars. Interest, current motion, and measured shipping are different dimensions.
Releases point toward an agent-aware editor
The v3.8.1 release on August 18 frames its work around three areas: more flexible AI integrations, more efficient editing and database workflows, and improved cross-platform reliability. The release notes list custom AI editor actions that sync across devices, shortcuts for the Agent panel, user-level skills from a shared agents directory, nested rerank API formats, and support for the OpenAI Responses API across AI features.
That is a more specific direction than adding a chatbot beside a notes editor. AI actions are being connected to the editor, synchronization, skills, retrieval, and agent sessions. The release notes also mention plugin availability in the publishing service. Taken together, the changes suggest that SiYuan wants agents to operate within the same structured workspace as documents, databases, references, and extensions.
The same release is not only about AI. It improves database grouping and virtualized grouped-table rendering, document loading, graph cluster layout, table sizing, list behavior, slash-menu configuration, and tab interactions. It adds support for editable formulas and lists pasted from Microsoft Office applications, HEIF and HEIC previews, and changes to PDF and export handling.
That mix matters because AI features do not compensate for a weak underlying editor. An agent can only be useful in a knowledge workspace if the workspace can load, organize, reference, export, and synchronize the material reliably. SiYuan’s recent release notes show engineering work across those foundations rather than a single-feature AI push.
The alpha releases continue the same pattern. The v3.8.2-alpha.1 notes include independent execution across Agent sessions, improved backlink filtering, database interaction changes on mobile, large-list rendering virtualization, and better data-sync conflict detection. The v3.8.2-alpha.2 notes add further work around Agent question and todo tools, Agent response tags, mobile database interactions, graph rendering, sync conflicts, and plugin event hooks.
The alpha.2 release also exposes the product’s extension direction. It lists support for intercepting asset uploads through the plugin event bus, adding custom buttons to the editor breadcrumb, and intercepting hyperlink opening. Those are integration points for teams that need to adapt the editor around an internal workflow rather than accept a fixed interface.
The supplied material does not include a formal roadmap. The direction above is an interpretation of the release sequence, not a claim about an unpublished plan. The evidence supports a product that is investing in agents, structured editing, databases, mobile clients, and extensibility at the same time.
What the public data cannot prove
ToolVitals can see repository activity, GitHub stars, contributors, releases, license classification, release notes, and other public project evidence. It cannot see code quality, user satisfaction, revenue, support responsiveness, or whether SiYuan works well for a specific team’s data.
The official site claims offline use, end-to-end encrypted synchronization, large-document performance, and local data control. Those claims are relevant to the product’s positioning, but the supplied excerpts do not include an independent security review, benchmark, threat model, failure analysis, or user study. The release notes show that maintainers are fixing synchronization, mobile, export, database, and agent issues. They do not show the remaining defect rate after those fixes.
The 4 active contributors are neither proof of fragility nor proof of quality. The 791 commits are neither proof of useful progress nor proof of churn. A commit can be a major architectural change, a small correction, or a generated update. ToolVitals’ confidence score of 90 describes confidence in the collected public evidence. It is not a reliability score for the application.
What an engineering lead should test
A skeptical buyer should run SiYuan against a representative workspace instead of treating a 96 ToolVitals score as a deployment decision. Start with the non-alpha v3.8.1 release as a baseline, then test newer alpha builds separately if their Agent or plugin features matter. Use the documented Docker deployment, create a workspace with large documents, nested blocks, databases, embedded assets, backlinks, and exports, then test backup and restore, offline behavior, synchronization conflicts, mobile editing, PDF output, and upgrades.
AI boundaries deserve their own test. Verify which content leaves the self-hosted instance when AI writing, question answering, reranking, or agent features are enabled. Test what happens when a model returns malformed structured arguments, since the alpha notes explicitly mention failures in Agent question and todo tools. Review plugin permissions, authentication settings, workspace access, and AGPL-3.0 obligations before putting sensitive or regulated material into production.
What the metrics suggest to maintainers
For maintainers, the main challenge is not proving that SiYuan ships. The metrics already make that case. The harder task is making a concentrated development model easy for outsiders to trust.
Four active contributors and 791 recent commits can look efficient or fragile depending on the evidence around them. Clear release channels, upgrade guidance, migration tests, backup and restore procedures, synchronization conflict documentation, AI data-flow documentation, and a visible stable-versus-alpha policy would reduce that uncertainty. The recent release notes are detailed, which helps. Publishing more evidence about operational boundaries would help engineering teams make decisions without confusing activity with assurance.
Recommendation
If your team needs a self-hosted knowledge workspace with block-level references, database-style organization, local data control, Docker deployment, and AI-agent integrations, SiYuan deserves a controlled evaluation now. Its 98 shipping score, 95 health score, 46,183 stars, and recent release work justify spending engineering time on a pilot.
Do not choose it from the star count or ToolVitals score alone. Choose it if the block model fits your workflows and the pilot passes backup, synchronization, export, mobile, extension, and AI-boundary tests. If your organization requires a large contributor base, independent security evidence, or a mature team-grade support model, treat those as open evaluation requirements rather than assuming the public activity metrics answer them.