Capacities vs Roam Research: pricing, features and which to choose

Capacities is a personal knowledge-management and note-taking app made by Capacities Labs GmbH in Germany. Steffen Bleher and Michael von Hohnhorst started the company in 2020, and the product entered public beta in 2022. It runs on the web, Windows, macOS, Linux, iOS and Android. The company describes itself as bootstrapped with no outside investors, while CB Insights records a different funding history involving Media Lab Bayern. Capacities reports more than 50,000 active knowledge workers and a six-person remote European team. Its defining choice is to treat notes as typed objects, such as people, books, projects and meetings, rather than as unstructured pages.

A typical workflow starts in daily notes, imported material, or captured media. Users turn that material into typed objects, add properties, connect entries with backlinks, then retrieve information through full-text search, smart queries, related content and the graph view. Tasks, whiteboards and calendar integration extend the workspace beyond notes. The in-app assistant can use selected notes or search context, and saved conversations become searchable objects. AI Chat Connectors use MCP, allowing supported clients such as ChatGPT, Claude and Cursor to search, read, create and update objects, append content and work with media. Capacities recommends searching or listing first instead of asking an external client to read an entire workspace.

Capacities is cloud-only, lacks end-to-end encryption, and does not offer team accounts or realtime collaboration. Its object model can take time to learn, while regular AI use requires Plus or additional top-ups. Notion offers broader shared workspaces, pages and databases. Obsidian stores local Markdown files and supports plugins, while Capacities provides hosted sync and structured objects. Anytype takes a local-first, open-source-oriented approach, unlike Capacities' hosted service. Tana focuses on structured outliner workflows and team use, whereas Capacities is more explicitly personal. Pick Capacities if you want an individual knowledge base built around typed objects, linked notes and optional AI.

Roam Research helped make bidirectional links and block-based outlining popular in note-taking, and many later tools, including Logseq and Obsidian's linking features, are often compared with it. It was created by Conor White-Sullivan and is built for researchers, writers, founders and students who think by connecting ideas rather than filing them into folders. Instead of a hierarchy of documents, Roam is a graph of pages and blocks. You write in an outliner, wrap any phrase in double brackets to create or link a page, and every page shows the linked and unlinked references pointing to it. Over months, that reference list becomes a record of how an idea developed.

The block is Roam's core unit. Each bullet has its own identifier, so you can reference or embed it in another page and edits show up everywhere it appears. The daily notes page opens each day as a default place to write, which removes the decision of where a thought belongs. Tags, attributes and queries pull blocks from across the graph into one view, and a graph view shows how pages connect. Roam also supports to-dos, kanban and table views, a sidebar for working on several pages at once, version history for blocks and keyboard-driven navigation. Graphs can be private, encrypted or shared with collaborators for team use.

Roam Research's strengths are block references and embeds that work across the whole graph, a daily notes workflow that lowers friction, powerful queries and an extension system with Roam Depot and roam/js. Its weaknesses are the lack of a free plan, a higher price than rivals, a learning curve for queries and attributes, and AI that depends mostly on extensions and your own API keys. Logseq offers a similar outliner for free. Obsidian stores Markdown files locally. Tana adds structured supertags and built-in AI. Notion suits shared docs and databases. Choose Roam if you want block-level linking and already think in outlines.