---
title: "Glossary"
type: "glossary"
canonical_url: "https://www.graphcanon.com/glossary"
---

# Glossary

_Plain-language definitions for the AI, graph, and data terms used across GraphCanon._

## AI & LLMs

The model-layer vocabulary you meet across tool pages, comparisons, and stacks - what the words mean before you pick a tool.

- [AI agent](/glossary/ai-and-llms/ai-agent.md): An AI agent is a system that uses a language model to decide and take actions - calling tools, running steps in a loop - toward a goal, rather than just answering once.
- [Context window](/glossary/ai-and-llms/context-window.md): The context window is the maximum amount of text (measured in tokens) a language model can consider at once, including your prompt and its own reply.
- [Embedding](/glossary/ai-and-llms/embedding.md): An embedding is a list of numbers (a vector) that represents the meaning of a piece of text, image, or code so that similar things sit close together.
- [Fine-tuning](/glossary/ai-and-llms/fine-tuning.md): Fine-tuning further trains a pretrained model on your own examples so it adapts to a specific task, tone, or format.
- [Hallucination](/glossary/ai-and-llms/hallucination.md): A hallucination is when a language model produces confident but false or unsupported output - a made-up fact, citation, or API.
- [Large language model (LLM)](/glossary/ai-and-llms/large-language-model.md): A large language model is a neural network trained on huge amounts of text to predict the next token, which lets it generate and reason over language.
- [Model Context Protocol (MCP)](/glossary/ai-and-llms/model-context-protocol.md): MCP is an open protocol that lets AI agents discover and call external tools and data sources over a standard interface, instead of every app inventing its own.
- [Retrieval-augmented generation (RAG)](/glossary/ai-and-llms/retrieval-augmented-generation.md): RAG fetches relevant documents at query time and feeds them to a language model as context, so answers are grounded in your data instead of only the model's training.
- [Semantic search](/glossary/ai-and-llms/semantic-search.md): Semantic search matches on meaning instead of exact keywords, usually by combining embeddings with traditional text ranking.
- [Token](/glossary/ai-and-llms/token.md): A token is the small chunk of text (roughly a word-piece) that language models read and generate; usage and limits are counted in tokens, not characters.
- [Vector search](/glossary/ai-and-llms/vector-search.md): Vector search finds the items whose embeddings are closest to a query embedding, ranking by meaning rather than exact word matches.

## Knowledge graph

How GraphCanon models tools as a connected graph instead of a flat directory - the pieces that make a graph a graph.

- [Canonical](/glossary/knowledge-graph/canonical.md): The canonical version is the single official form of something - one URL, one name, one record - that all variants and duplicates resolve to.
- [Edge (relationship)](/glossary/knowledge-graph/edge.md): An edge is a connection between two nodes - the relationship that makes a graph a graph.
- [Entity](/glossary/knowledge-graph/entity.md): An entity is a real-world actor in the graph - an organization or person - that tools can be linked to, such as the company that maintains a repo.
- [Knowledge graph](/glossary/knowledge-graph/knowledge-graph.md): A knowledge graph stores information as entities connected by typed relationships, so meaning lives in the connections - not just in rows of a table.
- [Node](/glossary/knowledge-graph/node.md): A node is a single thing in a graph - in GraphCanon, usually one tool, category, or entity.
- [Provenance](/glossary/knowledge-graph/provenance.md): Provenance is the recorded origin of a piece of data - where it came from, how it was derived, and when - so a claim can be traced and trusted.
- [Taxonomy](/glossary/knowledge-graph/taxonomy.md): A taxonomy is a controlled, agreed-upon set of categories used to classify things consistently, rather than free-form tags.
- [Typed edge](/glossary/knowledge-graph/typed-edge.md): A typed edge is a relationship with a named kind - such as alternative, integrates_with, depends_on, or successor - so the connection is precise.

## Corpus & data pipeline

The words for how GraphCanon collects, refreshes, and reasons over its data - starting with the one that probably sent you here.

- [Backfill](/glossary/corpus-and-data/backfill.md): A backfill retroactively fills in missing or newly added fields across records that already exist, rather than only handling new ones.
- [Corpus](/glossary/corpus-and-data/corpus.md): A corpus is the full body of data a system works over. In GraphCanon it is every tool, repo, and piece of metadata currently in the graph.
- [Dataset](/glossary/corpus-and-data/dataset.md): A dataset is a structured, published collection of records meant to be consumed as data - downloadable and machine-readable.
- [Enrich](/glossary/corpus-and-data/enrich.md): Enrichment runs a language model over already-stored data to add or improve derived fields - summaries, tags, decision facts - without fetching from GitHub.
- [Freshness](/glossary/corpus-and-data/freshness.md): Freshness is how recently a record was updated. Stale data is the opposite - old enough that it may no longer reflect reality.
- [GitHub sync](/glossary/corpus-and-data/github-sync.md): GitHub sync re-fetches a tool's public GitHub facts (stars, language, last push) without re-running the language model.
- [Ingest](/glossary/corpus-and-data/ingest.md): Ingesting is the first-time intake of a source into the system - fetching it, parsing it, and adding it to the corpus.
- [Metric snapshot](/glossary/corpus-and-data/metric-snapshot.md): A metric snapshot is a single dated record of a tool's numbers (like stars and open issues), appended over time to build a history.

## Trust & signals

What the numbers, labels, and badges on a tool page actually mean - and what they deliberately do not.

- [Adoption](/glossary/trust-and-signals/adoption.md): Adoption is how widely a tool is actually used. Stars approximate attention; real adoption blends multiple signals like dependents, releases, and backing.
- [Maintenance label](/glossary/trust-and-signals/maintenance-label.md): The maintenance label summarizes how recently a repository was worked on - from Very active to Dormant or Archived - based on public GitHub activity.
- [Momentum (star velocity)](/glossary/trust-and-signals/momentum.md): Momentum is the recent rate of change in a tool's popularity - typically its star gain over the last 30 days - a freshness signal, not a quality score.
- [Security scan](/glossary/trust-and-signals/security-scan.md): A security scan checks a repository's declared dependencies against known-vulnerability databases; a clean result is not proof a project is secure.
- [Stars](/glossary/trust-and-signals/stars.md): Stars are GitHub bookmarks users give a repo. They approximate attention and are easy to compare, but they are a lagging, gameable popularity signal.
- [Trust signal](/glossary/trust-and-signals/trust-signal.md): A trust signal is a sourced, dated heuristic that helps you judge a tool - a decision input, not a certification or safety guarantee.
- [Verified publisher](/glossary/trust-and-signals/verified-publisher.md): A verified publisher badge means a signed-in GitHub user with admin access to the repo has claimed the listing - distinct from an automated owner-type guess.

## Discovery for agents

The machine-readable surfaces that let AI agents read GraphCanon as data, not scraped HTML.

- [ai-catalog.json](/glossary/discovery-for-agents/ai-catalog.md): ai-catalog.json is a well-known JSON manifest that advertises a site's agent-facing surfaces - its API, dataset, and MCP endpoint - in one discoverable place.
- [Generative engine optimization (GEO)](/glossary/discovery-for-agents/generative-engine-optimization.md): GEO is optimizing content so AI answer engines cite it accurately - the AI-era companion to SEO, focused on being quotable and verifiable, not just ranked.
- [llms.txt](/glossary/discovery-for-agents/llms-txt.md): llms.txt is a plain-text file at a site's root that gives AI crawlers a curated, compact map of the most useful pages - like robots.txt, but for LLMs.
- [Markdown twin](/glossary/discovery-for-agents/markdown-twin.md): A markdown twin is a clean markdown version of an HTML page, served at the same path plus .md, so agents can read the content without parsing the page.
- [Sitemap](/glossary/discovery-for-agents/sitemap.md): A sitemap is an XML file listing a site's canonical URLs so crawlers can discover every page, including ones that are hard to reach by following links.
- [Structured data (JSON-LD)](/glossary/discovery-for-agents/structured-data.md): Structured data is machine-readable markup (usually JSON-LD following schema.org) embedded in a page so search engines and agents understand what it contains.

---

**Machine-readable endpoints**

- JSON: [`/api/graphcanon`](/api/graphcanon)
- LLM index: [/llms.txt](/llms.txt)
- Full corpus: [/llms-full.txt](/llms-full.txt)

_GraphCanon - The knowledge graph for AI development. https://www.graphcanon.com/_