Roam Research

Roam Research is the application that introduced networked note-taking to a wide audience, and its ideas are now so thoroughly copied that evaluating it in the present requires separating what it invented from what it currently offers. Launched publicly around 2019 after a period of invitation-only beta, Roam popularised a set of concepts that are now standard elsewhere: daily notes as the default capture surface, bidirectional links that show where a page is referenced, block-level referencing and embedding, an outliner as the primary writing structure, and a graph view of connections. Its early adopters were disproportionately researchers, writers and software engineers, and the community around it produced a substantial body of method writing about how to build a personal knowledge system.

The core model is an outliner of blocks. Everything you write is a bullet, bullets nest, and each bullet is an addressable object with a stable identifier. Pages hold blocks, and pages are created implicitly: typing a name inside double brackets links to it and brings it into existence if it did not exist. There is no folder hierarchy and no requirement to decide where a note belongs before writing it, which is the essential departure from conventional note-taking. The assumption is that structure should emerge from use rather than be imposed in advance, and that the cost of filing is high enough to suppress capture.

Backlinks are the mechanism that makes emergence work. Every page displays the pages that link to it, and also unlinked mentions of its title elsewhere in the graph, which can be converted to links. Over time a page about a concept accumulates every context in which that concept appeared, without anyone having built an index. Block references let you pull a single bullet into another location, and block embeds render a block or its children inline, so the same material can live in a project page, a daily note and a summary simultaneously with one source of truth. Page embeds do the same for whole pages. This is genuinely powerful for synthesis work, and it is the feature that users of other tools most often miss.

Queries provide retrieval. Filter blocks combine conditions such as page, tag, text content, date, checkbox state and logical operators into saved views, which can act as dynamic dashboards: every open task tagged with a project, every block mentioning a client, everything written this month containing a keyword. More advanced querying is possible through community-developed tools. Task management is supported through todo checkboxes, date markers for due and scheduled, and recurrence.

The graph overview renders pages as nodes and links as edges, which is visually compelling and, in honest assessment, of limited practical use for navigation once a graph passes a few hundred pages. Its real value is as a diagnostic: clusters reveal where your thinking has concentrated and sparse regions reveal what you have neglected.

Extensibility comes through roam/js, where users write JavaScript that runs in their own graph, and through a community ecosystem of extensions, themes and tools. This produced a rich culture of customisation, but it is also a signal: a great deal of what people wanted required writing code or installing an extension, and the platform itself evolved more slowly than its community wished.

Roam is a hosted web application with iOS and Android clients. Data lives on the vendor's servers. Export is available in JSON and Markdown, which is a real and important safeguard, and an offline mode caches the graph locally, but this is not a local-first tool in the sense that Logseq or Obsidian are. Your notes are accessible only through Roam, and if the service stops, your working environment stops with it, leaving only exports.

Against competitors, the field has changed substantially since Roam's early dominance. Obsidian offers a local-first, Markdown-based vault with bidirectional links, a very large plugin ecosystem, block references through core features and plugins, strong performance and free personal use with paid sync and publish services; for most people evaluating today, it is the more capable and more durable choice, though its model is page-centric rather than natively block-centric. Logseq is the closest structural equivalent: open source, local files, outliner-first, block references and embeds native to the model, daily notes, queries, free to use. Tana takes structured data further with typed nodes, fields and views, targeting users who want a database and an outliner in one. Notion dominates the mainstream with documents, databases, collaboration and sharing, but has no genuine bidirectional linking model and is not built for graph-style thinking. Reflect, Mem and Capacities offer AI-assisted networked note-taking in hosted form. Athens Research, an open-source Roam clone, ceased active development, which is a cautionary example of how dependent this category is on sustained maintenance. Roam's position now is that of the originator with a loyal community, a distinctive block model, and less momentum and less polish than its main alternatives.

Pricing is paid only. A subscription is billed per user per month or at a discounted annual rate, and there is a long-term multi-year Believer plan aimed at early supporters that reduces the effective monthly cost substantially in exchange for prepayment. There is a trial period but no permanent free tier, which is a significant competitive disadvantage against Obsidian and Logseq, both of which are free for personal use and can be synced at no cost. The subscription funds a small company, and buyers should weigh the continuity risk that comes with a paid, proprietary, hosted product whose development pace has varied. Current figures are published on the vendor pricing page.

The limitations should be stated plainly. Cost with no free tier puts it behind free local-first alternatives on pure value. Performance on large graphs has been a long-standing complaint, with slowdown in search, in page rendering and in the graph view as block counts grow, and this affects precisely the users who get the most from the tool. The interface is dense and unpolished, with a learning curve that is steeper than almost anything in the category and little onboarding. Mobile apps have historically lagged desktop in capability and stability. Data lives on vendor servers with export as the escape hatch. Collaboration is weak: Roam is fundamentally a single-user thinking tool, and shared graphs are awkward rather than designed for teamwork. The extension ecosystem depends on community volunteers. And feature development has been slower than users hoped, which matters when competitors ship continuously.

Roam Research fits writers and researchers whose work is synthesis across many sources and who think in fragments rather than documents, academics building a literature-linked knowledge base, people already invested in a Roam graph with years of notes, users who specifically want a hosted tool and do not want to manage local files and sync, and those drawn to the community and its accumulated method writing. It does not fit budget-conscious individuals who can get the same model free from Logseq or Obsidian, teams needing collaboration, people who want a polished and fast interface, users with strict data-ownership requirements, or anyone unwilling to invest weeks in learning an unfamiliar structure before the tool pays off.

In the closing comparison, choose Roam when you value the original block-based networked model, want a hosted service with no file management, and accept the subscription and the interface. Choose Logseq when you want the same structural model for free, with open source and local files, and accept a rougher application and self-managed sync. Choose Obsidian when you want the best all-round local-first knowledge tool with performance, polish and an enormous plugin ecosystem, and you can live with a page-centric rather than block-centric core. Choose Tana when you need structured typed data as well as networked notes. Choose Notion when collaboration and databases matter more than graph-style linking. The honest summary is that Roam created a category and its descendants now often serve it better; the reason to choose Roam today is preference for its specific model and community rather than any objective capability advantage.