Who it is for
Doctoral candidates, postdocs and academics working through a literature review or a thesis chapter. Students putting together a serious reading base. Anyone whose writing has to stand on sources rather than on memory.
Academic research
Nodus for Academic Research
Nodus is a free, open source, local-first workspace for academic research. It sits between your reference manager and your manuscript. You bring in the papers, books and documents you already work with, and Nodus turns them into a corpus of ideas, relations and evidence that you can search, question and cite, without any of it leaving your computer.
The shape of the tool
Built for research, not for notes in general.
Most tools that hold research material are either bibliographies or notebooks. A bibliography knows what you have read. A notebook knows what you thought about it. What neither of them knows is which passage of which work supports which claim, and that is the link your writing rests on.
Nodus is built around that link. Every interpretation it stores is one of five kinds, a claim, a finding, a construct, a method or a framework, and each one carries the work it came from and the excerpt that supports it. If the source gives a page number, that travels with it too. Everything else in the application is a way of moving through those anchored units. There is a graph of how they relate to each other, a map of an argument you are putting together, a list of the questions your corpus cannot answer yet, and a report whose citations you can open and check for yourself.
Nodus organises work into vaults, and this page describes the academic one. The same engine also runs the vaults for teaching, study, structured databases, genealogy and more, but the academic vault is the one designed for literature reviews, theses and research projects.
What it costs
Nothing. Nodus is released under the AGPL-3.0-only licence, with no accounts, no subscriptions and no telemetry. The desktop app runs on macOS, Windows and Linux.
What it does not do
It does not replace your reference manager, it does not read for you, and it does not decide whether an argument is sound. What it does is keep every interpretation next to its evidence, so that judging it is quick for you.
01 · Sources
Academic papers, books and documents, wherever you keep them.
Nodus keeps one cross-vault Library that holds bibliographic metadata, files and searchable text. A source lives there once and can be linked into any compatible vault without being copied.
01
Connect the research library you already have
Nodus can mirror a complete Zotero library from the Zotero application running on the same computer, keeping its collection hierarchy, stable item keys, attachments and child notes. It can also import RIS, BibTeX and CSL JSON exported from Mendeley or another manager, or take local documents directly.
The optional Nodus Connector for Chrome captures the academic page or document you are looking at into the Library, detecting embedded bibliographic metadata, DOI or ISBN and any available file.
02
Keep the original and get a readable copy
For each item, Nodus preserves the original attachment untouched and builds a clean Markdown reading copy beside it, along with extracted figures, structured tables and page mappings. PDF, EPUB, image, web, text and office attachments all stay openable in their preserved form.
The reader asks once whether you prefer the clean copy or the preserved original, and that choice can be reset at any time. Page-accurate verification always goes back to the original.
03
Decide what belongs in this project
Linking an item into a vault is what places it in that project's corpus. Removing the link does not destroy the global source, so a paper can inform a thesis vault and a teaching vault at once without being duplicated or falling out of either.
Collections, tags and filtered views keep a large corpus navigable, and duplicate records are worth resolving before analysis so a single source is not counted twice.
04
Read, highlight and annotate in place
Text can be highlighted in both the reflowable and the PDF viewers, and images accept region highlights. Highlights, notes and document chat stay attached to the source, which is what later lets an idea point back at an exact excerpt.
02 · Ideas and evidence
From what you read to what it claims.
Analysis turns the text of a work into structured interpretations. These are not loose summaries. Each one records what kind of interpretation it is, which source it came from, and the excerpt that supports it.
Typed ideas
Every extracted unit is stored as a claim, finding, construct, method or framework. Keeping those apart matters. A method borrowed from one field is a very different kind of thing from a contested empirical finding, and a corpus that lumps them together loses exactly the distinction you need.
Evidence that stays attached
An idea keeps the passage behind it, plus the page when the source supplies one. Nodus also records whether the evidence is quoted directly or paraphrased, so you always know how much of the wording in front of you is the author's.
Typed relations
Ideas are joined by relations that mean something. One idea supports, contradicts, refines, extends or applies to another, shares its method, or is a precondition for it. Each connection also records whether the source said so, or whether the analysis inferred it.
Themes
Related ideas gather into themes, which is what turns a few hundred extracted units into a readable map of a literature rather than a longer list.
The knowledge graph
Themes, ideas and authors appear as an interactive research map with presets and filters, so you can move from an overview of the whole corpus down to one contested question.
The argument map
A separate view arranges claims and evidence into the structure of a paper or chapter, with each piece marked as thesis, support, objection, reply or evidence. You can reorganise it as the argument develops.
Author profiles
An author view only gathers what the active corpus supports. That is narrower than a reputation on purpose, because what it shows you is the documented position of an author inside your material.
Immersion
A guided route through a central research question, with audio if you want it, taking you through evidence stations and commentary in order. It is there to help you get your bearings in unfamiliar material, and it does not stand in for reading the sources.
Extraction is a first pass, not a verdict. Automatically extracted text, metadata and ideas can be wrong, and Nodus is built on that assumption, which is why the excerpt and the preserved original are always one click away and a claim can be checked against the page it came from. For how the AI side of this works and where its limits are, see AI for academic research.
03 · Search
Semantic search across your own corpus.
Nodus searches works, ideas, passages and authors with both exact and semantic retrieval, and results keep the entity type and the source context they came from.
Exact search is the right tool when you remember the wording. Semantic search matters in the common case where you do not: when the corpus says desirable difficulty and you are asking about effortful practice, or when half your sources are in another language. Nodus builds embeddings for that, and you decide which model computes them.
Embeddings can be produced by an external provider such as OpenAI, Google Gemini or OpenRouter. They can also be produced entirely on your machine through Ollama, LM Studio or a small multilingual model that Nodus manages itself. The indexes live on your computer either way.
Searching is also how the rest of the workspace is wired together: a search result opens the idea, the idea opens its passage, and the passage opens the preserved original at the page it came from.
04 · Gaps and debates
Questions a corpus can be asked about itself.
Once a body of literature is stored as anchored ideas and typed relations, you can put some very useful questions to the corpus itself.
Research gaps
A gap records the evidence that exists, what is missing, and candidate directions. Nodus keeps four kinds apart, because they call for different responses. There is future work flagged by the authors, there are acknowledged limitations, there are open questions, and there are unresolved contradictions between sources.
The strength of a gap depends entirely on how well the corpus covers its area, which is why coverage sits next to it rather than in a different menu.
Debates
Debate views group opposed or qualified positions by stance and show the works standing behind each one. The point is not to end up with a pro and con tally. Boundary conditions, sample differences and methodological disagreements usually tell you far more than the headline split does.
Coverage analysis
Coverage measures how well the corpus supports the themes and questions you care about, highlighting both overrepresented areas and thin ones. It is the honest way to state the boundary of a literature review, and it doubles as a reading list for what to add next.
Hypotheses
A hypothesis in Nodus combines a statement, the ideas that support it, the risks against it and a proposed test. Those fields are kept separate on purpose, so that a plausible speculation cannot quietly present itself as an established result.
Reading paths
A reading path puts sources and ideas into a route with a purpose, say foundations first, then a debate, then the open questions. Every stop carries a reason for being there rather than a rank.
Deep Research
Deep Research runs as a queued job over the scoped corpus and produces a long-form, citation-rich report whose citations open the source they refer to. Treat it as a draft you can audit. It is useful because you can check it, not because it is finished.
05 · Writing
Writing with citations you can open.
Research notes lose most of their value if the manuscript forgets where they came from. Nodus keeps projects, notes and drafts in the same workspace as the corpus they draw on.
The Workspace holds projects, drafts and source-linked notes. Early observations start out close to the passage that prompted them, and as they mature they move into a structured manuscript without losing that trail.
Citations
The citation manager uses real CSL styles, including custom .csl files copied over from Zotero, and formats them locally on your machine after installation.
Word and LibreOffice
Companions for Microsoft Word and LibreOffice bring selected Nodus context into a manuscript you are already writing, so the corpus is reachable from the document rather than only from the app.
Other applications, through MCP
Nodus ships an MCP server. It lets an MCP-capable assistant read approved parts of the active vault, such as works, ideas, notes and reports, and work with them under your control.
06 · What stays on your machine
Local-first and open source.
Your vaults, the cross-vault Library and every search index live on your own computer and in the backup folder you choose. There are no accounts, no subscriptions and no telemetry, and model-assisted features stay switched off until you configure a provider or a local model. Nodus is open source under the AGPL-3.0-only licence, and every build links to its corresponding source. What the licence gives you, and what leaves your machine.
100% local storage
Offline AI (Ollama · LM Studio)
Encrypted auto-backups
Audit ledger
Free forever (AGPL-3.0-only)
Keep going