Glossary

A unified, alphabetical reference for the terms used throughout this documentation. Where a term has a full reference page, it's linked.

A

• Agent — An LLM-driven reasoning step that an agent node runs. You configure an agent by choosing its type, writing its Instructions, and wiring its inputs, knowledge bases, and Tools. See Agent reference.
• Agent type — The kind of work an agent does. Built-in types are search_respond, intent_classification, extraction, rule_validation, and respond; the platform also supports code_execution agents and custom agents.
• AI Employee (AIE) — The unit of automation you build, test, publish, and run. It bundles a workflow with an interface, knowledge bases, Tools, and access controls. See What Is Ema?
• AI Employee builder — The visual workspace where you assemble and configure an AI Employee's workflow.
• Audit trail — The chronological record of actions and changes across the workspace, used for compliance and review. Surfaced under Governance. See Administration.
• Autopilot — Ema's in-app assistant. Describe what you want in plain language and Autopilot builds and edits the workflow, agents, knowledge bases, and Tools for you. See Autopilot.

C

• Citation — A reference to the source passage an agent used to ground an answer. Returned by search_respond agents and surfaced in the UI.
• code_execution agent — An agent type that runs your JavaScript or TypeScript in a sandbox, for logic better expressed in code than in Instructions.
• Condition (edge) — A test attached to an edge. When the source node completes, the engine evaluates each outgoing edge's condition and follows only the edges whose condition is true. This is how a workflow branches — there is no separate "conditional" node. See Conditions and Expressions.
• Custom agent — A purpose-built agent dispatched to a dedicated execution service, for work outside the built-in agent types.

D

• DAG (directed acyclic graph) — The structure of a workflow: nodes connected by edges, with no cycles. Because it never loops back on itself, every run has a well-defined path the engine validates before execution.

E

• Edge — A connection between two workflow nodes that defines execution order and how data flows downstream.
• EmaFusion™ — Ema's model layer. Every LLM call flows through it; it selects an appropriate model per request across multiple providers, tracks token usage, and applies privacy controls at the model boundary. See EmaFusion™.
• end node — The node that terminates a workflow run.
• extraction agent — An agent type that extracts structured data from source text, conforming to a JSON Schema you supply.

F

• feedback_router agent — An agent type (set on an agent node) that routes a run based on end-user feedback, used for re-work or escalation paths.
• Form (interaction type) — An AI Employee interface driven by a defined set of inputs, suited to structured, repeatable runs.

H

• Human-in-the-loop (HITL) — A capability that pauses a workflow to bring in a person — to answer a question, approve a step, or fill a form — then resumes. Enabled on an agent node (there is no separate node type). See Human-in-the-loop.

I

• Instructions — The plain-language system guidance you give an agent to shape its behavior. (Not called "prompts.")
• Integration — A connection between your tenant and an external application. Once installed, it exposes Tools. See Integrations and data.
• intent_classification agent — An agent type that classifies input text into one of a set of intents you define; the label is commonly used in edge conditions to branch the workflow.
• Interaction type — How an AI Employee is surfaced to users: chat, form, or voice. Chosen when the AI Employee is created.

K

• Knowledge base — A collection of documents and synced data sources that an AI Employee retrieves from. Content is chunked and embedded so agents can find and cite relevant passages. See Knowledge bases.

N

• Node — A single step in a workflow's DAG. Each node has a type: start, agent, transform, publish, or end. Branching and human-in-the-loop are not node types — they are configured on edges and on agent nodes respectively.

P

• PII (personally identifiable information) — Sensitive data (such as names, emails, and account numbers) that Ema detects and tokenizes before it leaves for a model or an external Tool. Governed per tenant. See Administration.
• publish node — A workflow node that emits a result: a final response, a record write-back, or a generated document.
• Publish (a workflow) — To release a version of a workflow so it can run. Workflows are versioned and can be compared and reverted.

R

• respond agent — An agent type that generates a free-form natural-language reply from your Instructions, with no type-specific behavior added.
• Role — A permission level that governs what an account can do. Common roles include builder, builder_admin, user, user_admin, and system_admin. See Administration.
• rule_validation agent — An agent type that evaluates documents and inputs against a list of rules and returns a per-rule verdict with cited evidence.
• Run — A single execution of an AI Employee's workflow, from a start node to an end node. Runs stream their progress and can pause at an agent node that has human-in-the-loop enabled.

S

• search_respond agent — An agent type that searches knowledge bases and external Tools in a reasoning loop, then returns a synthesized, citable answer.
• SSO (single sign-on) — Authentication through your organization's identity provider (SAML or OIDC). Configured by administrators. See Administration.
• start node — The entry point of a workflow; receives the run's inputs.

T

• Template — A pre-built AI Employee blueprint you can use as a starting point. See Templates.
• Tool — A discrete read or write operation an agent can call, exposed by an installed integration (for example "look up a customer" or "create a ticket"). (Not called "actions.")
• transform node — A deterministic, non-LLM workflow node that extracts and coerces named fields from an upstream value using a list of rules. Never calls a model.

V

• Version (workflow) — A published snapshot of a workflow. You can compare versions and revert to an earlier one.
• Voice (interaction type) — An AI Employee interface for phone and voice channels. Voice AI Employees run the same workflow with voice-specific configuration. Beta.

W

• Workflow — The definition behind an AI Employee: a DAG of typed nodes that the workflow engine validates and runs. See Workflows.
• Workflow engine — The component that validates a workflow's DAG and executes it node by node, streaming progress and pausing durably at human-in-the-loop checkpoints. (This is the engine Ema runs; older material may call it the "Generative Workflow Engine" or "GWE.")
• Workspace — Your organization's tenant in Ema: the AI Employees, integrations, members, and settings that share one permission and policy context. Shown as Workspace in the Admin sidebar.