SiYuan’s most unusual signal is the gap between its breadth of shipping and its visible contributor count. ToolVitals records 1,442 GitHub commits in the last 30 days, 23 release events in the same period, and 5 active contributors. That is not proof that every change is good, but it does support a sharper thesis: SiYuan is being pushed toward an agent-aware, self-hosted knowledge workspace at a pace that demands disciplined evaluation, not casual note-app testing.
SiYuan is built around blocks, not pages
The official SiYuan product page presents the application as a privacy-first personal knowledge management system built around Markdown formatting, block-level references, and two-way links. The GitHub repository uses a broader description: an open-source, privacy-first, self-hosted knowledge workspace where humans and AI agents work together.
Those descriptions point to the same product decision. SiYuan treats the content block as the primary unit of knowledge. A document is not just a long page. Its paragraphs, headings, lists, tables, and other elements can be focused, rearranged, referenced, embedded, and connected to other blocks.
The official site says SiYuan includes more than 20 types of block elements and more than 10 types of inline elements. It also highlights block zoom, breadcrumb navigation, large-document editing, linked databases, backlinks, relationship graphs, list outlines, and folding. The page describes moving, splitting, and reorganizing blocks without affecting existing links, which is a meaningful distinction for anyone building a long-lived research archive.
That model targets a problem that conventional note apps often leave to the user. A collection of pages can store information, but it does not automatically preserve the relationships between claims, references, tasks, and derived ideas. Block-level references and backlinks give those relationships a first-class place in the interface. The value is not the visual graph by itself. The value is being able to move a piece of knowledge without losing the references that make it useful.
SiYuan also puts control of the data near the center of its product story. The official page says data is stored on devices controlled by the user and remains usable offline. It describes end-to-end encrypted synchronization for multiple devices. It also presents mobile server use and Docker deployment as practical ways to reach a workspace across devices or run it on a server.
The repository adds operational detail. Its Docker instructions use the serve subcommand explicitly, and the README says that this became required for Docker usage beginning with v3.7.0. The documented container options include a workspace path, an access authentication code, and configurable user and group IDs for mounted directories. These details do not make deployment automatic, but they show that self-hosting is part of the supported operating model rather than a community workaround.
The repository’s AI positioning is also specific enough to matter. It lists AI writing and question-and-answer chat through the OpenAI API, and its current description puts humans and AI agents in the same workspace. That does not establish that model inference runs locally. It establishes that SiYuan is building an interface and extension surface for AI-assisted work while retaining its self-hosted knowledge base as the underlying context.
The activity signal is concentrated shipping
ToolVitals gives SiYuan a hot score of 275.8, a health score of 98, a shipping score of 99, and an overall ToolVitals score of 98. The measured repository has 45,945 GitHub stars, 1,442 commits in 30 days, 30 GitHub releases in 90 days, 23 release events in 30 days, and 5 active contributors. ToolVitals reports data confidence of 100 for this snapshot.
The notable pattern is not simply that SiYuan is active. It is that the activity is dense and concentrated. Twenty-three release events in 30 days is a substantial public delivery cadence. The 1,442 commits provide a second signal that work is happening below the release surface. Five active contributors make the pattern more interesting because the visible output is not paired with a large active contributor count.
There are several possible explanations for that concentration. A small core team may be carrying a large amount of work. Some of the commit volume may come from tightly managed internal workflows. The public metric may also capture changes that vary greatly in size and importance. ToolVitals does not identify which explanation is correct, so the responsible conclusion is narrower: SiYuan has high measured activity, and that activity appears to be concentrated among a small active contributor group.
The scores are useful for triage, not for procurement approval. A health score of 98 tells a reader that the public project signals measured by ToolVitals are strong. A shipping score of 99 says the recent delivery pattern is strong. Neither score verifies test coverage, upgrade safety, documentation quality, or the behavior of a particular feature in a real workspace.
The star count needs the same treatment. Forty-five thousand nine hundred forty-five stars indicate substantial public visibility and interest, but stars are cumulative. They do not show how many people use SiYuan every week, how many depend on it for critical records, or how many have stopped using it. The data confidence score refers to confidence in the collected project evidence, not confidence that the software will meet every team’s requirements.
For an engineering lead, this is the practical reading. SiYuan has earned a serious technical evaluation because the repository is active, releases are frequent, and the product has a defined architecture. The same cadence raises the need to test upgrades and deployment procedures rather than assuming that a high score removes operational risk.
Releases show where the product is going
The SiYuan v3.8.1 release, published on August 18, frames its work in three areas: more flexible AI integrations, more efficient editing and database workflows, and improved cross-platform reliability. That framing is more revealing than any individual feature. The project is not treating AI as a separate experiment. It is adding AI controls to the editor, the agent panel, shared skills, APIs, publishing, and synchronization.
The AI changes are concrete. Version 3.8.1 supports syncing custom AI editor actions across devices. It adds Agent panel shortcuts and user-level skills from a shared agents directory. It supports nested rerank API request and response formats, along with the OpenAI Responses API across AI features. These changes expand the ways an agent can retrieve context, operate within the editor, and fit into a user’s existing workflow.
The release also works on the less glamorous parts of knowledge management. It improves database grouping performance and virtualizes grouped table rendering. It adds table-column sizing controls, editable lists and formulas when pasting from office applications, better document loading, and more flexible slash-menu configuration. Those changes matter because a knowledge system fails in practice when everyday editing becomes slow or unpredictable, regardless of how interesting its agent features look.
Cross-platform work is another large part of the release. The notes mention HEIF and HEIC previews, mobile behavior, PDF handling, export reliability, marketplace changes, and fixes involving security, MCP, publishing services, and formulas. This is a broad maintenance surface. Desktop, mobile, browser, export, plugins, and remote serving all create separate failure modes.
The follow-up v3.8.2-alpha.1 release and v3.8.2-alpha.2 release make the direction even clearer. The alpha notes include independent execution across Agent sessions, improved Agent response tags, fixes for malformed structured arguments in Agent question and todo tools, backlink filtering, database interaction improvements on mobile, and better detection of data synchronization conflicts involving prefetched and cached files.
The same alpha sequence continues work on graph rendering, large-list virtualization, mobile navigation, custom editor toolbars, asset upload handling, Word image pasting, duplicate asset filenames, and plugin event hooks. This is not one isolated AI feature being polished. It is a broad attempt to make agents, structured data, editing, synchronization, and extensions operate inside the same block-oriented system.
The alpha labels matter. The recent release evidence includes stable, beta, and alpha versions, so the 23 release events in 30 days should be read as release activity, not as 23 stable production versions. That cadence can be useful for users who want fixes quickly and are willing to test. It can be a liability for teams that need a slow upgrade channel, a long support window, or predictable behavior across many clients.
The release notes support a product-direction claim, not a roadmap guarantee. SiYuan is moving toward a workspace in which structured personal knowledge is available to both a human editor and software agents. The evidence does not say how far that design will go, or whether the resulting workflows will outperform simpler tools for a given team.
What ToolVitals cannot tell you
ToolVitals sees repository activity, contributors, releases, license signals, stars, and other public project evidence. It does not see code quality, user satisfaction, revenue, support responsiveness, internal planning, or whether the product actually works well for a particular workload. Those are separate questions that require source inspection, hands-on testing, and conversations with users or maintainers.
The privacy story needs the same distinction. SiYuan’s official page describes local data control and end-to-end encrypted synchronization. Those are important architectural claims, but the supplied evidence does not include an independent security audit, a threat model, or a review of every plugin and integration. Self-hosting changes who controls the server, but it also makes backup, access control, patching, network exposure, and recovery part of the operator’s job.
ToolVitals classifies SiYuan as OSI-approved open-source software and lists its license as AGPL-3.0. That is the correct license language for this project. Teams that modify, embed, redistribute, or connect to SiYuan in a larger product should review the license and their deployment model with qualified counsel instead of treating the open-source label as a substitute for legal analysis.
A skeptical buyer should turn the metrics into an acceptance test. Create a representative workspace with large documents, linked blocks, databases, attachments, and exports. Run it through the intended Docker or desktop deployment, test backup and restore, exercise synchronization conflicts, verify mobile behavior, inspect plugin permissions, and document what data leaves the local environment when AI features are enabled. A score of 98 cannot answer those questions.
The public evidence also does not establish the size or quality of the support operation. Five active contributors can produce excellent software, but it can also mean that institutional knowledge is concentrated. The only defensible response is to test the workflows that would be difficult to replace, and to keep an export and recovery plan before making SiYuan a system of record.
SiYuan against note-taking peers
The related_tools data places AFFiNE and Anarlog in the same note-taking category. AFFiNE has 71,813 GitHub stars, a shipping score of 100, a hot score of 219.9, and 16 release events in the last 30 days. SiYuan has fewer stars at 45,945, but its 23 release events and shipping score of 99 indicate a more active recent release stream in this snapshot.
Anarlog has 9,139 GitHub stars, a shipping score of 98, a hot score of 219.8, and 14 release events in 30 days. SiYuan’s measured hot score is higher at 275.8, while its star count is also much larger. Those numbers make SiYuan look like the more active project among these listed signals, but they do not make it the better product for every use case.
The comparison also shows why one metric is insufficient. AFFiNE’s higher star count suggests broader accumulated visibility, while SiYuan’s recent release count points to a different strength, sustained delivery. Neither comparison reveals which application handles a team’s data model, migrations, offline requirements, or collaboration habits best.
Recommendation for evaluators and maintainers
Evaluate SiYuan if your team wants a self-hosted knowledge workspace where block-level references, two-way links, structured databases, local data control, and AI-assisted editing belong in one system. The fit is strongest for individuals, research groups, and small technical teams that are willing to own deployment and recovery rather than outsource those responsibilities.
Use the v3.8.1 release as the evaluation baseline and treat the v3.8.2 alpha releases as a test channel. The release notes show meaningful work in agents, databases, mobile, synchronization, and extensions, but alpha tags are not a reason to put new behavior directly into a critical production workspace. Pin versions, record the workspace format and backup process, and repeat the same tests after upgrades.
Do not choose SiYuan only because ToolVitals reports a 98 overall score, 99 shipping score, or 45,945 stars. Choose it if the block model solves a real information problem and if your team can accept the operational obligations of self-hosting. The measurable activity is a reason to investigate. It is not a replacement for an integration test.
Maintainers should read the five active contributors alongside the 1,442 commits and 23 release events. The project is demonstrating strong output, but high output from a concentrated group creates a communication burden for users and potential contributors. Clear stable, beta, and alpha boundaries would help users interpret the release stream.
The release notes already show attention to malformed Agent arguments, synchronization conflicts, mobile failures, export issues, and plugin behavior. Turning those fixes into visible regression tests, migration guidance, compatibility notes, and explicit AI data-flow documentation would make the activity easier for engineering teams to trust. For SiYuan, the next credibility gain will not come from shipping more features alone. It will come from making a fast-moving project easier to operate safely.