SiYuan’s strongest current signal is a mismatch. ToolVitals records 1,331 GitHub commits in 30 days and a shipping score of 98, yet it records zero release events in that same 30-day window. This is not a quiet note-taking project. It is a project doing substantial work between packaged milestones, with its product direction moving toward an agent-aware knowledge workspace.
The latest listed release events, SiYuan v3.8.2-alpha.1 on August 20 and v3.8.2-alpha.2 on August 21, point toward active development. ToolVitals also records 30 GitHub releases in 90 days. The evidence supports a clear thesis, but not a simple one: SiYuan is shipping heavily at the code level while its public release cadence has paused inside the most recent measurement window.
The product is a knowledge system built around blocks
The official SiYuan site presents the product as a privacy-first personal knowledge management system built around content blocks. Its central claim is structural, not cosmetic. A document is made from blocks, and blocks can be focused, moved, split, recombined, referenced, and linked without treating the page as an indivisible file.
That model gives SiYuan a different center of gravity from a basic Markdown notebook. The official site describes block-level references, two-way links, backlinks, relationship graphs, list outlines, databases, and block-level zoom. It also lists more than 20 block element types and more than 10 inline element types. The result is a system designed to preserve relationships between pieces of knowledge as the workspace changes.
The site also claims support for very large documents, including million-word editing through dynamic loading. That is a product positioning claim, not an independent benchmark, but it explains why the project puts so much attention on rendering, document loading, large lists, database grouping, and navigation. The block model only helps if the interface remains usable as the workspace grows.
The GitHub repository broadens the positioning. Its current description calls SiYuan an open-source, privacy-first, self-hosted knowledge workspace where humans and AI agents work together. The repository lists Markdown WYSIWYG editing, block-level reference, two-way links, databases, graph views, PDF annotation, exports, multi-tab editing, and AI writing and question answering through the OpenAI API.
Self-hosting is not an optional detail in this description. The official material says data is stored on devices controlled by the user and can remain available without a network connection. It also describes end-to-end encrypted synchronization, a mobile server mode, and Docker deployment. The repository’s deployment guidance includes a serve subcommand, a configurable workspace path, and an access authentication code for server use.
ToolVitals classifies SiYuan as OSI-approved open source and identifies its license as AGPL-3.0. That makes the licensing model part of the deployment decision. Teams evaluating a hosted service, internal fork, or modified distribution should review the AGPL-3.0 obligations alongside their architecture and compliance requirements.
The metrics show activity, but not a finished product
ToolVitals assigns SiYuan a hot score of 222.2, a health score of 97, a shipping score of 98, and an overall ToolVitals score of 97. Those are strong project-health signals within the ToolVitals model. They do not mean the product has no defects, but they do indicate that the public project evidence looks active and maintained.
The raw figures add useful detail. SiYuan has 46,479 GitHub stars, 1,331 commits in the last 30 days, 30 GitHub releases in the last 90 days, and 7 active GitHub contributors. ToolVitals reports data confidence of 90.
The commit count is the most striking number. More than a thousand commits in a 30-day period suggests a busy engineering cycle, especially when the release notes show simultaneous work on the editor, databases, mobile clients, AI integrations, exports, plugins, and synchronization. It does not show whether those commits improve the product for most users. A large commit count can include refactoring, generated changes, fixes for narrow cases, or work that has not reached a stable release.
The release figures need the same care. Thirty releases in 90 days indicates frequent packaging across the broader window, while zero release events in the latest 30-day window indicates that no release event was captured in that narrower period. Those statements can both be true. The latest listed events are alpha builds from August, while the ToolVitals snapshot was generated later. The data does not explain whether the gap reflects a planned stabilization period, release-channel timing, or a measurement boundary.
Seven active contributors is also informative but limited. It shows that current activity is not represented by a single visible contributor, yet it does not reveal the full team, ownership of critical components, review practices, or support capacity. It should not be treated as a direct measure of organizational resilience.
Stars provide reach, not usage. The 46,479-star count says that SiYuan has attracted substantial public attention. It does not establish the number of active workspaces, the size of the self-hosted user base, or the number of teams relying on it for important information.
The ToolVitals scores are best read as signals about public project activity and maintenance. A health score is not an uptime guarantee. A shipping score is not a release-quality audit. The confidence score tells readers how much trust to place in the available measurements, not whether the underlying application works well for a particular workflow.
Recent releases show an agent layer becoming part of the product
The SiYuan v3.8.1 release notes frame the release around three areas: more flexible AI integrations, more efficient editing and database workflows, and improved cross-platform reliability. That framing matters because it places AI beside the core editor and database system rather than treating it as a separate experiment.
The release adds support for syncing custom AI editor actions across devices. It also adds Agent panel shortcuts, user-level skills from a shared agents directory, nested rerank API formats, and support for the OpenAI Responses API across AI features. Publishing services can configure plugin availability, which connects the agent and extension story to deployment controls.
The same release is full of less visible but important work. It improves database grouping performance and virtualizes grouped table rendering. It adds table-column sizing controls, database item templates, editable pasted formulas and lists, HEIF and HEIC previews, document loading improvements, PDF export handling, graph interactions, and mobile behavior. This combination suggests that the maintainers are trying to make agent features operate inside a broader knowledge workflow, not just add a chat panel.
The v3.8.2-alpha.1 notes and v3.8.2-alpha.2 notes extend that direction. They mention independent execution across Agent sessions, better Agent response tags, handling for malformed structured arguments in Agent question and todo tools, and improved database interactions on mobile. They also include data-sync conflict detection, large-list rendering virtualization, and plugin event bus support for asset uploads and link handling.
Those details point to a shift from simple AI assistance toward agent operations inside the workspace. Independent sessions, structured arguments, skills, reranking, plugin events, and configurable editor actions are integration features. They define how agents interact with documents, assets, databases, and the interface.
The alpha label is important. These release notes show where the project is working, not where every feature has reached production maturity. A team that needs dependable knowledge storage should separate the stable v3.8.1 baseline from the v3.8.2 alpha line and test the latter as a development channel.
The release sequence also exposes the cost of this direction. Once AI actions can change or query workspace content, reliability depends on more than model quality. Structured arguments, permissions, session boundaries, sync conflicts, plugin behavior, mobile state, and undo semantics all become part of the product surface. SiYuan’s recent changes show that the maintainers are working on those edges, but the supplied evidence does not show how they perform under real workloads.
What the data does not tell you
ToolVitals sees repository activity, contributors, releases, licensing signals, and public project evidence. It does not see code quality, user satisfaction, revenue, support response times, backup success, security audit results, or whether the product works well for a specific organization.
The official site describes local storage and end-to-end encrypted synchronization, but the supplied context does not independently verify the implementation or its threat model. The repository lists Docker deployment and AI integrations, but that does not establish that upgrades are painless, that sync conflicts are rare, or that agent actions are safe enough for unattended use.
The metrics also cannot tell a buyer whether SiYuan’s block model fits existing information architecture. A team may value block references and databases, or it may find that page-level Markdown files, conventional folders, or established export formats fit better. That question requires a hands-on evaluation with real documents and assets.
A skeptical engineering lead should test four things before adopting SiYuan for important knowledge. First, import representative documents with images, tables, formulas, and attachments. Second, test backup, restore, and synchronization conflict handling across the devices the team actually uses. Third, verify that the database and block-reference model supports the team’s retrieval and maintenance habits. Fourth, review agent permissions and failure behavior before allowing AI features to modify shared information.
Those tests address questions the public metrics cannot answer. A score of 97 cannot tell you whether a migration preserves every link. A star count cannot tell you whether a broken mobile workflow will be fixed on your schedule. A high commit count cannot tell you whether an upgrade will preserve your workspace.
SiYuan is ahead of nearby projects on activity, not on every popularity measure
The related_tools data puts SiYuan’s activity in context. Among the note-taking entries, Anarlog has a hot score of 167.0, 9,381 GitHub stars, and a shipping score of 97. AFFiNE has a hot score of 166.9, 72,900 stars, and a shipping score of 95.
SiYuan’s hot score of 222.2 and shipping score of 98 are higher than both listed peers. Its 46,479 stars are higher than Anarlog’s 9,381 but lower than AFFiNE’s 72,900. The comparison says that SiYuan currently scores strongly on the ToolVitals activity model while sitting between the two peers on public star count.
These are not feature-by-feature comparisons. The related_tools data does not establish that Anarlog or AFFiNE solve the same workflows, use the same architecture, or serve the same users. It does show why a single popularity metric is insufficient. AFFiNE has more stars, while SiYuan has the higher ToolVitals hot and shipping scores in this set.
Recommendation for buyers and maintainers
If your team needs a self-hosted knowledge workspace with block-level references, databases, local data control, Docker deployment, and growing agent integrations, SiYuan deserves a controlled evaluation. Use the non-alpha v3.8.1 release as the baseline. Test the v3.8.2 alpha line separately if independent Agent sessions, custom actions, skills, or OpenAI Responses API support are central to your workflow.
Do not choose SiYuan because it has 46,479 stars or a ToolVitals score of 97. Choose it if its block model matches how your team records and retrieves knowledge, if its deployment model fits your operations, and if its AGPL-3.0 license fits your distribution plans. The current metrics justify a serious pilot. They do not justify skipping migration, backup, synchronization, and agent-safety tests.
For maintainers, the public evidence suggests that the next challenge is not proving that SiYuan is active. The 1,331 commits, 30 releases in 90 days, and recent AI, database, and mobile changes already make that clear. The harder task is turning activity into user confidence.
That means making stable versus alpha behavior easy to identify, documenting migration and sync-conflict expectations, explaining agent permissions and session boundaries, and showing how plugins interact with assets and links. The release notes expose a wide stream of work. Clear compatibility guidance and focused operational documentation would help buyers understand which parts are ready to carry important knowledge.
The concrete recommendation is simple: teams with privacy, self-hosting, and structured knowledge requirements should put SiYuan through a real pilot, not a star-count comparison. The project has enough activity and product breadth to earn that test. Whether it earns production adoption depends on the results of the workflows ToolVitals cannot measure.