Personal knowledge tools — Roam, Obsidian, Logseq
The recovery treated in this chapter is the most direct of the recoveries in Part V. Where local-first is recovering data ownership, ActivityPub is recovering federation, and the static renaissance is recovering the document, the personal knowledge tools — Roam Research, Obsidian, Logseq, Tana, and the surrounding ecosystem — are recovering, almost without modification, the working pattern of NoteCards and KMS from chapter nine. The recovery is recent: Roam launched in 2019, Obsidian in 2020, Logseq the same year. The user bases are substantial — Obsidian alone reports over a million users — and the cultural impact has been significant within the broader “Tools for Thought” community. The recovery is also incomplete in specific ways that are worth being explicit about. This chapter takes the principal tools, the community that uses them, the discourse that surrounds them, and tries to be honest about how much of the original lineage has been recovered and how much has not.
Roam Research
The recovery began with Roam Research, founded by Conor White-Sullivan. Roam launched in private beta in 2017 and reached general availability in 2019. The system was, in its initial form, a web-based note-taking tool with two specific design choices that distinguished it from the existing market. First, it used double-bracket syntax for links: typing [[some concept]] in a note created a link to a page about that concept, with the page being either created on the spot or, if it already existed, linked to. Second, every note was a tree of indented bullet points (what Roam called “blocks”), and every block had a unique address that could be referenced from any other block in the system. The combination produced a graph of interlinked notes in which the unit of linking was the bullet point rather than the page, and in which the user could navigate by following links in any direction.
Roam was the first widely-used tool to ship these patterns in a polished, accessible form. The patterns themselves were not new. The double-bracket syntax came from wiki tradition (Ward Cunningham’s WikiWikiWeb had used CamelCase as a similar shorthand in 1995, and subsequent wikis had moved to brackets). The graph-of-notes model was older than any wiki, descending through NoteCards, KMS, and the academic hypermedia tradition. The block-level addressing was new in its polish but had precedents in various structured-note systems. What Roam did was assemble these patterns into a working consumer tool with a clean interface, intuitive interactions, and a path for non-technical users to get started.
The cultural reception was extraordinary. Through 2019 and into 2020, Roam became the subject of a near-religious enthusiasm in a specific community of academic, technical, and creative professionals who had been frustrated with the existing note-taking tools. Roam users (who called themselves “Roamans” with various degrees of irony) produced extensive blog posts, YouTube videos, courses, and books explaining how to use the tool and what it had changed for them. The discourse around Roam established, for the broader culture, the value proposition of the linked-notes pattern: that thinking with a graph of interconnected notes was qualitatively different from thinking with a folder of separate documents, and that the difference was worth the learning curve of adopting a new tool.
Roam was, however, a closed system. The notes lived on Roam’s servers; the data was accessible only through Roam’s interface; the format was proprietary. The pricing — $15 per month, with no free tier beyond a limited trial — was high by note-taking-tool standards and produced significant pushback. By 2020 and into 2021, a portion of the Roam community had begun looking for alternatives that preserved the linked-notes pattern but addressed Roam’s specific weaknesses. The alternatives that emerged are the subject of the next sections.
Obsidian
Obsidian launched in 2020, the work of Erica Xu and Shida Li (the latter using the pseudonym Silver). Obsidian’s design choice was the opposite of Roam’s: where Roam was a web application with cloud-stored notes in a proprietary format, Obsidian was a desktop application with local notes in Markdown format. The notes were files in a folder; the application was a viewer and editor that overlaid the linked-notes interface on top of an ordinary directory of Markdown files. The result was that an Obsidian user owned their data in the strong local-first sense: the notes existed as plain-text files on the user’s machine, readable by any text editor, modifiable without the application, and persistent regardless of what happened to Obsidian the company.
Obsidian’s data model differs from Roam’s in significant ways. Where Roam treats every block as a first-class addressable unit, Obsidian treats the file as the primary unit, with within-file linking working only through standard Markdown conventions. Block-level addressing exists in Obsidian (through the ^ syntax for block IDs) but is secondary. The file-as-primary model is closer to how most users think about notes — a note is a file, with a name — than Roam’s block-as-primary model, and the consequence is that Obsidian’s learning curve is somewhat lower than Roam’s for new users.
Obsidian’s plugin ecosystem has been one of its principal strengths. The application supports community-developed plugins for almost any extension: graph visualizations, kanban boards, daily-notes templates, integration with various other tools, calendar views, mind-mapping interfaces, AI assistants, and a few thousand other things. The plugin ecosystem has produced, in effect, a personal-knowledge-tool platform on top of which various smaller communities have built their preferred workflows. The flexibility has been a major reason Obsidian has outgrown Roam in user count: where Roam imposes a specific workflow, Obsidian lets the user (or the user’s plugin set) define the workflow.
The business model has worked. Obsidian the application is free for personal use; the company makes money through paid Obsidian Sync (a CRDT-based synchronization service that lets users keep their notes synced across devices) and Obsidian Publish (a way to publish a subset of one’s notes as a static website). The paid features are real services rather than artificial limitations on the application; the free tier is genuinely useful for most users; the model has supported the small team to date, with a reported user base in 2026 of well over a million.
Obsidian has, more than any other tool in this chapter, become the default option for new entrants into the personal-knowledge-tool space. The combination of local-first data, Markdown format, polished interface, extensive plugins, and free-for-personal-use pricing has made it the natural starting point for someone investigating these tools. The choice has its trade-offs (the block-level addressing is less developed than in Roam or Logseq, the user interface for graph operations is less first-class than in Roam, the synchronization story requires either the paid service or external tools), but the trade-offs are usually acceptable for the broader audience the tool has reached.
Logseq
Logseq, also launched in 2020, is the open-source alternative to Roam. Founded by Tienson Qin and a small team, Logseq adopts Roam’s block-as-primary data model — every line is a block, every block has a unique ID, blocks can be referenced from anywhere — and combines it with Obsidian’s local-first storage. Logseq notes live as Markdown (or org-mode) files in a folder on the user’s machine, organized in directories that the application understands. The format is mostly compatible with Obsidian, with some block-reference syntax that is Logseq-specific but readable as plain Markdown.
Logseq’s open-source nature is important. The application’s source code is on GitHub under the AGPL license; the development is community-driven; users can self-host the synchronization service if they prefer not to use the company’s hosted version. The community has been active and the application has improved steadily through the early 2020s, with a notable rewrite around 2023 that improved performance and user interface significantly.
Logseq has, by reputation, the most technically sophisticated user base of the three principal tools. The combination of block-level addressing, journal-based organization (every day has its own page by default), and powerful query capabilities (Logseq’s Datalog-inspired query language is more capable than the equivalents in Roam or Obsidian) has attracted users who want a more structured approach to their notes. The user base is smaller than Obsidian’s but substantial, in the hundreds of thousands.
The three principal tools — Roam, Obsidian, Logseq — together constitute a working ecosystem of mature personal-knowledge tools, with substantial communities, real production use, and continuous development. The differences between them are real and matter to specific users (block-primacy versus file-primacy, cloud versus local, open versus closed source), but the common ground is much larger than the differences. All three implement the linked-notes pattern; all three support block references in some form; all three have rich import-export with each other and with adjacent tools; all three have produced substantial communities of users who think of these tools as core parts of their working life.
The broader ecosystem
Beyond Roam, Obsidian, and Logseq, the personal-knowledge-tool space includes a substantial number of related applications, each occupying a specific niche.
Notion, founded in 2013 by Ivan Zhao and team, is not strictly a personal-knowledge tool in the linked-notes sense — it descends from the document-and-database tradition rather than from the wiki-and-graph tradition — but it has adopted some linked-notes features (page mentions, database relations) and is widely used for personal knowledge work. Notion is cloud-only and proprietary, which puts it on the opposite side from Obsidian on the local-first question. The application has a much larger user base than any of the linked-notes-pure tools, reportedly in the tens of millions, and its design choices have influenced the broader category significantly.
Tana, founded in 2022 by Tobias Vogel and team, is a more recent entrant that combines outliner-style block hierarchy with database-like structured fields. Tana’s distinctive feature is “supertags” — types that can be applied to any block, with the types having structured fields that the user can fill in. The combination produces a tool that sits between Roam-style linked notes and Notion-style structured databases. Tana’s user base is smaller than the older tools but the design has been influential.
Capacities, Reflect, RemNote, Heptabase, Mem, Athens (which sunset in 2023), and several others occupy adjacent positions. The space is, by 2026, dense with options, each with its own design philosophy and target user. The diversity is itself a recovery: a decade ago, there were essentially no serious personal knowledge tools; now there are dozens.
A specific subset of the ecosystem includes tools that integrate machine-learning capabilities, particularly the various AI-assisted notes tools that have emerged since the public availability of large language models in 2022 and 2023. The integration of LLMs into personal-knowledge tools is, in 2026, in an early phase: most of the major tools have some form of AI-assist feature (asking questions about your notes, generating connections, summarizing, drafting), with varying levels of integration. The longer-term consequences of this integration are still unfolding; it has the potential to significantly change what these tools are used for and how, but the current state is preliminary.
The Tools for Thought discourse
The community surrounding personal-knowledge tools has acquired a self-conscious identity over the past several years. The community calls itself, with some self-irony, the “Tools for Thought” community, after the longer tradition of writing about computing as a thinking aid (Howard Rheingold’s 1985 book Tools for Thought, J.C.R. Licklider’s “Man-Computer Symbiosis” essay, the broader cyberculture lineage). The discourse has been substantial: blog posts, podcasts, YouTube channels, online courses, conferences, books, and a steady stream of essays about what makes a good thinking tool, how to use these tools well, and what the historical lineage of the present generation of tools is.
Several specific figures have been prominent. Andy Matuschak, formerly of Khan Academy and Apple, has been writing about “Evergreen Notes” — a specific practice for accumulating durable, well-formed notes — and about the broader problem of how to design tools that support sustained intellectual work. Matuschak’s working notes themselves, published as a public site at notes.andymatuschak.org, have been an unusual model of what a personal knowledge garden can look like when its author treats it as a serious artifact. Matuschak’s collaboration with Michael Nielsen on Quantum Country and related experiments has produced one of the more interesting recent ventures in dynamic-document-meets-personal-knowledge territory.
Tiago Forte’s book Building a Second Brain (2022) was one of the more commercially successful entrants in the popularizing literature about personal knowledge management. Forte’s specific method — Capture, Organize, Distill, Express, or CODE — has been adopted by many users as a working framework. The book has been an entry point for many users into the broader Tools for Thought conversation.
Maggie Appleton’s writing and visual essays on digital gardens, on what notes-as-thinking-tools can be, on the cultural turn toward sustained personal websites — these have been widely shared and have contributed to the broader discourse. Other voices: Cortex Futura, Bianca Pereira, Joel Chan, Robert Haisfield, the various working researchers and educators who have been thinking carefully about these tools.
The cumulative effect of this discourse has been to establish, for a generation of intellectual workers, that personal knowledge tools are a serious category and that learning to use them well is a worthwhile investment. The conversation has analogues with the academic hypermedia community’s discourse from chapter seven, with the difference that the current Tools for Thought community has reach and impact that the academic community of the 1980s and 1990s did not have. The audience is larger; the writing is more accessible; the tools are more polished; the practice is more available.
Connection to the older lineage
The chapter is, on the argument of this book, about a recovery of what the knowledge-base systems of chapter nine had been doing. The recovery is direct enough that it is worth being explicit about the lineage.
NoteCards at PARC had cards with content, links between cards, types on the links, and the FileBox container hierarchy. Roam, Obsidian, and Logseq have notes with content, links between notes (through [[brackets]]), tags as a form of typed annotation, and folder hierarchies. The mapping is direct.
KMS had frames with structured content, fast navigation, and a flat-name addressing scheme. Roam has pages with structured (outlined) content, fast navigation through [[brackets]], and a flat-name addressing scheme. The mapping is direct.
gIBIS had typed nodes (issues, positions, arguments) and typed links between them, with the structure being the artifact. Several recent tools (Tana with its supertags, RemNote with its concept structure) have similar typed-structure systems that recover, in slightly different form, what gIBIS had been doing. The mapping is less direct but recognizable.
The recovery is not, in all cases, conscious. Many users of the current tools have never heard of NoteCards or KMS or gIBIS. The patterns the current tools implement were rediscovered, in many cases, by their designers from first principles or from intermediate sources rather than from the academic lineage. But the patterns are the same patterns. The recovery has, intentionally or not, brought back what the knowledge-base lineage had been doing in 1985, on better hardware, with more polished interfaces, for a much larger audience.
What has been recovered
A summary of what the personal-knowledge-tools wave has recovered:
The linked-notes pattern as a working consumer tool. NoteCards and KMS had been workstation-class research tools. The current personal-knowledge tools work on consumer hardware, on the web or as cross-platform desktop applications, for users with no specific technical training in hypermedia systems. The pattern is now accessible.
Block-level addressing. Roam’s block references, Logseq’s block IDs, and Obsidian’s block anchors all implement, in user-facing form, the precise sub-page addressing that Xanadu had specified and that the web had refused. The addressing is local to each tool, not federated across the web, but within the tool the property is real and useful.
Bidirectional linking. Every linked-notes tool has, by default, automatic backlinks: open a page, and you can see what other pages link to this one. The property the web rejected has been recovered for the personal-knowledge domain, where the small scale makes the bidirectional-link maintenance tractable.
The graph view as an artifact. Most of the major tools have a visualization of the user’s notes as a graph: nodes are pages, edges are links, and the user can navigate by panning around the graph. The visualization is not, for most users, the primary interface, but it provides a visible representation of the linked structure that the user has been building, and the visibility makes the structure feel like an artifact in a way that an invisible link-graph would not.
The author-reader unity that HyperCard had. The user of these tools is, in every meaningful sense, both an author and a reader. There is no separate authoring interface and reading interface; the same tool is used for both, and the user is, throughout, building the artifact they are also reading. The continuum that HyperCard had recovered, and that the web subsequently lost, is back in the personal-knowledge domain.
What remains partial
The federation between tools. Each of the personal-knowledge tools is, in its current form, an island. Roam users have Roam graphs; Obsidian users have Obsidian vaults; Logseq users have Logseq graphs. Cross-tool linking, sharing, and collaboration is possible but limited. The export-import paths exist but are not seamless. The dream of a federated linked-notes web in which any user’s notes can be referenced from any other user’s notes has not been realized. (Obsidian Publish provides a partial path for publishing individual vaults to the web, but the resulting public pages are not interoperable as a network.)
The end-user authorship reach. The personal-knowledge tools are, despite the relative ease of use compared to NoteCards or KMS, still aimed at a specific audience: people who think note-taking is important enough to learn a dedicated tool. The mass-market audience that HyperCard had reached — the teachers, doctors, hobbyists, ordinary computer users making things for their own purposes — has not been reproduced. The user base of the current tools is in the millions, which is substantial but small compared to consumer-software scales.
The long-term durability of the artifacts. Markdown-file-based tools (Obsidian, Logseq) have the local-first property that the files are durable in a strong sense. Web-application-based tools (Roam, Notion) have the property that the user’s artifacts depend on the application’s continued existence. The durability question is being answered differently by different tools, and the answer matters: a user who has invested several years of note-taking in a tool that subsequently dies has lost something significant.
The integration with the broader web. Personal-knowledge tools are, mostly, separate from the web in the sense that the notes live in their own systems and the web is a place the notes refer to. A future generation of tools that better integrates the personal-knowledge graph with the user’s web browsing, web reading, and web publishing is plausible and would be a real advance. The various read-it-later services (Readwise, Matter, Pocket), the various web annotation tools (Hypothesis, Diigo), and the various integration plugins for Obsidian and similar tools are partial moves in this direction. The integration is, in 2026, still developing.
The argument the recovery makes
The personal-knowledge tools wave is, on the argument of this book, the most successful of the recoveries because it has produced the largest user base, the most polished tools, and the most active discourse of any of the recoveries in Part V. The wave’s success has several specific features worth attending to.
The recovery has worked because it operates at the personal scale. The hard problems that broke the federation recovery and the local-first recovery — billions of users, multi-party identity, platform economics — do not apply at the personal scale. A user managing their own notes can adopt a working tool without solving the world’s problems first. The narrowness of the scope is what has made the recovery tractable.
The recovery has worked because the value to the individual user is immediate and tangible. A new Obsidian user can experience, within an hour of starting to use the tool, the difference between linked notes and folders of separate documents. The benefit does not require buy-in from a community or from a platform; the benefit accrues to the user as soon as they start writing.
The recovery has worked because the existing tools are good. Roam, Obsidian, and Logseq are, by reasonable software standards, well-designed and well-implemented applications. They work; they continue to be developed; they are not embarrassing in any obvious way. The contrast with the rough state of the federation tools in chapter twenty-two is real and matters.
The recovery has, in this sense, made a case for what the broader recovery could look like: a generation of tools that take the lessons of the lost alternatives, implement them on contemporary platforms, with contemporary user-experience standards, for the audience that wants them. The personal-knowledge tools are not, individually, a complete recovery — they leave most of the losses this book has chronicled unaddressed — but they demonstrate that some of the losses are recoverable when the conditions are right.
The final chapter assesses what the conditions look like for the other recoveries to follow. Some of the losses, on the argument of the final chapter, are recoverable on similar paths to the personal-knowledge-tool recovery: with patient work, with attention to historical lineage, with the floor lowered enough to bring in a wider audience than the original systems had had. Some are not. Naming which is which is the work of the final chapter, and the closing of the book.