Glossary

The words used in pubmed-stream's tutorials and output, in plain language.

Abstract. The summary at the start of a paper. Paper notes include it for downloaded papers.

AI assistant. Here, an AI that can run commands in a terminal, such as Claude Code, Codex or Gemini CLI. It can drive pubmed-stream for you; see tutorial 7.

API key. A free personal code from NCBI that raises its speed limit from 3 to 10 requests per second. See getting a key.

Bases. An Obsidian core plugin that shows notes as a table, filtered and sorted by their properties. Every knowledge graph has one, Papers.base; see the Obsidian guide.

Boolean operators. AND, OR and NOT, written in capitals, combine search terms. A AND B needs both, A OR B either, A NOT B excludes B. Brackets group them: (A OR B) AND C.

Bridge paper. A paper that links parts of a knowledge graph that are otherwise far apart, such as two lines of research. Measured by betweenness centrality. Bridge papers are good early reads.

Cell. One box of a notebook. Text cells explain; code cells run and show their results underneath.

Chunk. A passage of a few hundred words cut from a paper, the unit that embedding models and RAG systems work on. Tutorial 6 makes them.

Collection. In a knowledge graph, a note for one search or one evidence neighborhood, listing the papers it brought in.

Colab. Google Colaboratory, a free service that runs notebooks in the web browser. See option 1.

Concept note. In a knowledge graph, a note for one MeSH heading, listing every paper indexed with it.

Corpus. The set of papers a study works with, such as all the downloads of a literature review.

DOI. Digital Object Identifier, a permanent ID for a paper, such as 10.3390/nu11071633. https://doi.org/ followed by the DOI opens the paper at its publisher.

Embedding. A list of numbers that represents the meaning of a passage, so a computer can find passages with similar meaning. Used for semantic search and RAG.

E-utilities. NCBI's public programming interface to PubMed and PMC. pubmed-stream uses it to search, download and follow links.

Evidence neighborhood. The papers linked to one key paper: the ones PubMed rates as similar, the ones that cite it (cited by) and the ones it cites (references). Tutorial 4 explores one.

Field tag. A label in square brackets that limits a search term to part of a record: [Title], [Title/Abstract], [mh] for MeSH, [pdat] for publication date. frailty[Title] finds papers with frailty in the title.

Full text. The whole paper, not only its abstract. pubmed-stream downloads full text from PMC when the publisher allows it.

Graph view. Obsidian's map of a vault: every note a dot, every link a line. See the Obsidian guide.

Instructions file. A text file, such as AGENTS.md or CLAUDE.md, that an AI assistant reads when it starts in a folder. pubmed-stream's tells it how to search and to cite only papers it has read. The plugin brings the same rules as a skill, literature-search.

JATS XML. The standard format in which PMC stores papers, with every section, figure caption and reference marked up. pubmed-stream can keep it for text mining (fmt="both").

JSON. A plain-text file format for structured data. pubmed-stream saves each paper as one JSON file, PMC<number>.json, readable by any programming language.

Key paper, or seed. A paper whose evidence neighborhood you add to a study, usually a landmark review or discovery. Tutorial 0 finds one automatically: the paper your downloads cite most.

Knowledge graph. A network of linked facts. Here, an Obsidian vault in which papers, MeSH concepts and searches are notes, linked by citation, similarity and topic. pubmed-stream names it kg_<topic>.

Local model. An AI model that runs on your own computer, so nothing you give it leaves the machine. Ollama runs them; small ones summarize text but cannot run commands reliably. See tutorial 7.

Major topic. A MeSH heading that NLM's indexers mark as a main point of a paper, as opposed to one that is only discussed. Search for it with [majr].

MCP. The Model Context Protocol, an open standard by which AI assistants such as Claude Code, Codex and Gemini CLI call tools directly. pubmed-stream's MCP server is part of pubmed-stream: the assistant starts it with the command pubmed-stream mcp, as the plugin or pubmed-stream init tells it to. See tutorial 7.

MeSH. Medical Subject Headings, NLM's controlled vocabulary of about 30,000 biomedical concepts. Indexers tag each paper in MEDLINE, PubMed's curated core of journals, with MeSH headings, so a MeSH search finds papers about a concept whatever words the authors used. Look up headings in the MeSH browser; search with [mh]. New papers take weeks to months to be indexed, and papers in journals outside MEDLINE never are.

Model. The AI that an assistant runs on. Each assistant offers several, from fast to most capable; see choosing a model.

Metadata. Information about a paper rather than its text: title, authors, journal, year, identifiers, keywords and references.

NCBI. The National Center for Biotechnology Information, part of the US National Library of Medicine (NLM). It runs PubMed and PMC.

Notebook. A document that mixes explanations, code and results, such as the tutorials. Jupyter and Colab run them.

Obsidian. A free note-taking app that shows linked notes as a graph. See the Obsidian guide.

Ollama. A free program that runs AI models on your own computer. pubmed-stream review uses it to summarize abstracts locally.

Open access. Papers whose license allows free reading and reuse. In PMC, "open access"[filter] keeps only papers that pubmed-stream can always download.

PMC. PubMed Central, NCBI's free archive of full-text papers. pubmed-stream searches PMC and downloads full text from it.

PMCID. A paper's ID in PMC, such as PMC6683074.

Plugin. An add-on that gives Claude Code or Codex new tools and instructions in every folder. pubmed-stream's plugin connects the assistant to pubmed-stream's MCP server and brings its rules as a skill; it uses uv to fetch pubmed-stream. See tutorial 7.

PMID. A paper's ID in PubMed, such as 31319564. Knowledge-graph notes are named after it: PMID31319564.

Precision. The share of a search's results that are really about the question. Tutorial 2 measures it.

Property. A labelled detail at the top of an Obsidian note, such as year: 2019. Obsidian can search and sort by properties.

PubMed. NCBI's database of more than 35 million biomedical paper records: titles, abstracts, MeSH headings and links, but mostly not full text.

Publication type. PubMed's label for what kind of paper a record is: Journal Article, Review, Systematic Review, Meta-Analysis, Randomized Controlled Trial and more.

Query. A search, written in PubMed's search language, such as "gut microbiota"[Title] AND frailty[Title/Abstract].

RAG. Retrieval-augmented generation: an AI system that answers questions by first finding relevant passages, such as chunks of papers, and then writing an answer from them, so the answer can cite its sources.

Rate limit. The maximum number of requests NCBI accepts per second. pubmed-stream stays within it automatically; an occasional 429 Too Many Requests message means it waited and retried.

Recall. The share of all relevant papers that a search finds. Stricter searches gain precision but can lose recall; combining word and MeSH searches with OR recovers it.

Reference manager. Software that stores and formats citations, such as Zotero, EndNote or Mendeley. Tutorials 0 and 3 export files they can import.

Retraction. The withdrawal of a published paper, for error or misconduct. PubMed marks retracted papers; pubmed-stream flags them, and Obsidian's graph view colours them red.

Skill. A set of instructions that Claude Code or Codex loads when a task calls for it. pubmed-stream's plugin has one, literature-search, with the same rules as AGENTS.md.

Similar articles. PubMed's list of papers with similar words and MeSH headings, scored by its own algorithm. Scores compare within one paper's list, not across lists.

Stub. A paper note without full text, for a paper in a key paper's evidence neighborhood that was not downloaded. It has PubMed's details and links, but no abstract.

Terminal. A window for typing commands: Terminal or Command Prompt on Windows, Terminal on a Mac.

uv. A small, free program that downloads and runs Python programs. pubmed-stream's plugin uses it to fetch and start pubmed-stream, so the assistant needs no pip install. Install it from docs.astral.sh/uv.

Vault. Obsidian's name for a folder of notes. A pubmed-stream knowledge graph is a vault.

Wikilink. A link between Obsidian notes, written [[note name]].