The Libra Dictionary
Clear, one-sentence definitions for the concepts behind AI.
A
- AI AgentAn AI system that carries out a multi-step task across tools, rather than returning a single answer.
- AI GovernanceThe permissions, approval rules and audit requirements that determine what an AI system may do.
- Ambient AgentAn agent that runs in the background on a signal rather than a prompt, surfacing work instead of waiting to be asked.
- Approval WorkflowA checkpoint that requires human sign-off before an AI agent completes a consequential action.
- Audit TrailThe durable record of what an AI system did, in which system, on whose authority, and whether a person approved it.
B
C
- Company BrainThe connected layer of company knowledge, context and systems that gives AI the information it needs to reason and act across the business.
- ConnectorThe link between an AI platform and a system like Slack, Gmail or Salesforce.
- Context GraphA structured map of how a company’s information connects: documents, decisions, people, conversations.
- Context LayerThe part of an AI system that holds a company’s connected knowledge for every agent to reason from.
- Context WindowThe amount of text an AI model can process in a single request, and why it is not the same as persistent memory.
- Context-Aware ReasoningAn AI system’s ability to interpret a request using surrounding company knowledge, not just the prompt.
D
- Data ResidencyThe requirement that company data is stored and processed in a specific country or region, and not moved outside it.
- Desktop AIAI embedded directly in an employee’s everyday computer workflow, rather than a separate app to visit.
- DiarizationWorking out who spoke when in a recording, so a transcript is attributed rather than one undifferentiated block of text.
E
- EmbeddingA numeric representation of what content means, letting a system compare passages by sense rather than by the words they use.
- Enterprise AIAI systems built for the access controls, deployment and scale of a company, not a consumer tool.
- Evaluation (Evals)Measuring an AI system’s output against known-good cases, so a change can be shown to help rather than assumed to.
F
G
H
I
K
L
M
- MCP (Model Context Protocol)An open standard that lets AI systems connect to external tools and data sources in a consistent way.
- Meeting BotThe participant an AI platform sends into a call, so the recording and transcript do not depend on someone remembering to press record.
- Multi-Agent SystemSeveral narrow agents coordinating on one outcome, instead of a single agent holding every tool and every instruction.
N
O
P
Q
R
S
- Semantic SearchSearch that matches meaning rather than wording, so a result can be relevant without sharing any words with the question.
- Structured OutputConstraining a model to return data in a fixed shape, so the result can be used by a program rather than read by a person.
- System of RecordThe authoritative source for a specific type of company data, like a CRM for customers.
T
- TaskA unit of work delegated to an AI system and tracked to completion, rather than a question answered and forgotten.
- Tenant IsolationThe guarantee that one customer’s data and compute cannot reach another’s inside a shared platform.
- TokenThe unit a language model reads and writes — roughly a word fragment — and the unit AI usage is measured and billed in.
- TriggerThe event that starts a piece of automated work — a message arriving, a record changing, a meeting ending — instead of a person starting it.