> ## Documentation Index > Fetch the complete documentation index at: https://docs.hyperspell.com/llms.txt > Use this file to discover all available pages before exploring further. # Glossary > The terms you will meet in the Hyperspell dashboard, in your agents, and in these docs. Hyperspell borrows a few words from everyday work and gives them a precise meaning. This page defines them. Terms link to each other, so you can start anywhere. ## Company brain The company brain is the knowledge layer between your workspace tools and the people and AI agents who need what those tools contain. It includes the [memory network](#memory-network) built from your [documents](#documents), and the [pages](#pages) Hyperspell generates from that network. ## Sources Sources are where your data lives today: Slack, Gmail, Notion, Google Drive, Granola, HubSpot, and so on. Hyperspell reads from sources to build the brain. ## Connections A connection is what a user creates when they enable an integration. It stores the credentials Hyperspell uses to read from the source on that user's behalf, or on behalf of the whole organization for a shared integration. A shared integration has one connection. A personal integration can have several connections per user, for example two Gmail accounts. ### Shared connections Shared connections hold data that all or most people in the company already have access to: internal knowledge bases, ticketing systems, CRMs. An admin connects them once, usually with an API key or a service account. Their content is visible to everyone in the organization by default. ### Personal connections Personal connections are connected by each [user](#user) individually, usually through an OAuth sign-in. Email, calendars, and direct messages are typical examples. Their content is visible only to the person who connected them by default. ## Documents A document is a piece of data from a source, converted into Hyperspell's own document format. A document may be an email thread, a Slack channel, a CRM contact with its notes and history, an uploaded PDF, or a web page. Documents are meant to be self-contained. A single Slack message that says "Yes" is useless without the question above it, which is why Hyperspell treats a channel, not a message, as the document. Documents are synced and updated automatically from the connected sources. ## Memories Memories are what Hyperspell extracts from [documents](#documents) during [ingestion](#ingestion). They come in a few kinds: * Facts, which are statements about the world. * User preferences. * Procedures, which tell an agent how to do something. * Learnings, which are answers to earlier questions. When someone asks "What projects did Ben work on in Q2?", Hyperspell resolves who Ben is, interprets Q2 as a date range, finds the relevant projects, and compiles an answer. That answer is stored as a memory, so the next time someone asks a similar question the result may be there right away. ## Memory network The memory network is Hyperspell's version of a knowledge graph. It connects extracted [memories](#memories) with their source [documents](#documents) and with [entities](#entities) such as people or organizations. Most links in the network are unlabeled. They say "this entity is relevant to this document, and so to every memory extracted from it." Their purpose is to help retrieval find relevant information faster by narrowing a query to the entities it is about, rather than to record every relationship between them. ## Entities Entities are the named things your data refers to. Hyperspell tracks three kinds: * People. * Work streams, a broader term than projects. It covers routine processes, team procedures, and long-running tasks that do not fit neatly into a project. A work stream is one or more people working towards an outcome. * Organizations, such as vendors and customers. A named entity is one mentioned in text, as in "Josh said we should ship by the end of the week". A linked entity is one that arrives through structured data: the sender of an email, a customer record in a connected CRM, or an [external account](#external-accounts) on a connected service. ## Content-based rules Content rules decide what spreads through the company brain and what stays put. There are two kinds. Both are set up by an organization admin. ### Content-based filters Filter rules keep content from spreading. They apply to content that would otherwise be shared with the whole company. When a filter rule matches a document, the document is held back from the company brain. Admins write filter rules for the categories that must never spread, such as compensation, HR matters, legal disputes, or credentials. ### Content-based sharing Sharing rules do the opposite. Content from a [personal connection](#personal-connections) is private by default. A sharing rule offers to share personal content with the rest of the organization when it matches, such as a customer call or a vendor conversation. Each person decides for themselves which sharing rules to turn on for their own connections. Only rules a person has opted into are applied to their content. See [Content rules](/advanced/content-rules) for how the two kinds are applied together. ## Pages People and agents can query the [memory network](#memory-network) directly, but it is often more useful to collect everything about one subject in one place. Pages are those pre-compiled summaries, one per project, work stream, person, or other subject. The reasoning behind pages comes from the three axes of [quality](#quality). Retrieval systems are good at surfacing relevant information, but they cannot promise completeness, and correctness is hard to check in the moment. Compiling a subject into a page ahead of time makes it possible to review both. ## Quality Hyperspell measures the quality of a company brain along three axes: 1. Correctness. Is the information true? 2. Relevance. Does it matter to the task at hand? 3. Completeness. Is everything needed for the task included? Relevance and completeness depend on context. So does correctness, more often than you would expect. Take "We closed the deal with Microsoft." It is true for the sales team, false for finance until the money arrives, and unclear for legal while signatures are still outstanding.