lain definitions of the vocabulary behind one problem: your AI tools don't know your company. Each definition stands on its own. Where we have a full page on a term, it is linked.
Company brain
A company brain is a single, current, structured representation of how a business actually works, its definitions, rules, and institutional knowledge, captured once and made readable by both people and AI tools. Y Combinator named the company brain one of the startup ideas it most wants built in 2026. A company brain is only useful if AI can read it at the moment it answers; the architecture that makes that possible is a context layer. Full page: What Is a Company Brain?
Context layer
A context layer is the governed layer between a company's data stack and its AI tools. It holds what the data means (definitions and metrics), how the company works (rules and institutional knowledge), and who is allowed to see what, and serves all of it to any AI tool at the moment it answers, with every answer traceable to its source. Full page: What Is a Context Layer for AI?
Institutional knowledge for AI
Institutional knowledge for AI is the unwritten rules, decision patterns, and operational context that make a company work, captured in a governed form that AI tools can read and act on at the moment they answer. It is the layer between what the data says and what it means: the reasoning no database stores and no model can infer.
Context debt
Context debt is the compound cost that accrues every time a company adds an AI tool that builds its own understanding of the business instead of reading from a shared source. Each tool taught separately adds to the principal; the interest is the growing cost of maintaining, reconciling, and second-guessing tools that disagree. Unlike technical debt, it compounds multiplicatively, because each new tool multiplies the surface on which tools can disagree.
Context Catalog
The Context Catalog is the surface in Sento where a company gets written down: definitions, metrics, business logic, and policies, each with an owner, a last-validated date, and the decision history behind it. It is seeded automatically from connected tools (the warehouse, dbt, Notion, Slack) and confirmed by a person, rather than authored from a blank page.
Knowledge Capture
Knowledge Capture is the surface in Sento that lets non-data teams encode the unwritten rules of the business, the renewal flow, the discount policy, the account sensitivities, with a review queue so the data team curates what becomes canonical. It exists because the hardest company knowledge lives in people's heads, where no integration can reach it.
MCP (Model Context Protocol)
MCP is an open standard for giving AI tools access to external context and tools at the moment they run. Any MCP-compatible tool (ChatGPT, Claude, Cursor, custom agents) can query an MCP server for context while it works. MCP is the pipe, not the context: without a governed source on the other end, it gives every tool faster access to the same chaos.
MCP Server (in Sento)
The MCP Server is the surface in Sento that serves governed company context to any AI tool at inference time, filtered by who is asking, what they are allowed to see, and what is relevant to the question. It is what lets a company write its context down once and have every current and future AI tool read from it on day one.
Lineage and Audit
Lineage and Audit is the surface in Sento that logs every context resolution, so any AI output can be traced back to the exact definitions, policies, and data it used. It is what separates "the AI was wrong" from "the AI used this rule, and the rule was out of date," and it is what makes AI outputs safe to put in front of customers.
Governed context
Governed context is company knowledge that carries authority: each piece has an owner, a last-validated date, a review process that decides what is canonical, and an audit trail of where it gets used. The opposite is ambient context, text scattered across wikis and prompts that an AI reads with equal confidence whether it is current or two years stale.
Semantic layer
A semantic layer (dbt Semantic Layer, Cube, AtScale) defines a company's metrics once so BI tools compute them consistently. It covers one slice of what AI tools need, what the data means, but carries no unwritten rules or institutional knowledge, and serves dashboards rather than arbitrary AI tools. Comparison: Context Layer vs Semantic Layer vs Data Catalog
Data catalog
A data catalog (Atlan, Alation, Collibra) documents an organization's data assets, metadata, and lineage so people can find and trust data, primarily for analysts and BI governance. It describes data for humans to browse; it was not built to serve governed meaning and rules to AI tools at the moment they answer. Comparison: Context Layer vs Data Catalog
RAG (retrieval-augmented generation)
RAG is a technique where an AI system retrieves text chunks relevant to a question and adds them to the model's prompt before it answers. RAG finds text that looks similar; it does not resolve which of several conflicting definitions is canonical, carry ownership or authority, or enforce who may see what. RAG can run on top of a context layer, but it is not one.
System prompt
A system prompt is the standing instruction text given to one AI tool, often including a pasted glossary of company terms. It works for a single tool with a stable glossary and breaks at scale: it does not travel across tools, does not update when definitions change, and carries no authority or audit. Comparison: Context Layer vs System Prompt
AI that knows your company
AI that knows your company is the outcome all of the above exists to produce: every AI tool a company runs answering from the same governed definitions, rules, and knowledge, traceable back to the source, instead of guessing. How-to: How to Make AI Understand Your Business