Concepts
This page explains the concepts and terminology used across the Suvereo documentation and codebase.
Data set
A data set is a workspace. It holds the agents and integrations of one company, brand or department, together with the content they work on. It also carries the settings of the language model that agents use and of the embedding model, which turns content into numbers so it can be searched by meaning.
Data sets are separate from each other: an agent only uses content that belongs to its own data set, and access can be limited to a single data set. A single data set is enough to get started. Further data sets are useful for separate brands, for separate content languages or for trying out different models.
Agent
An agent is an AI assistant set up for a specific job. It can be defined on top of a domain definition or directly on top of Suvereo. Its definition specifies the tools it can call and the settings it accepts. Each agent has a name, a description and values for those settings, and belongs to one data set.
An agent does its job by calling tools. People talk to agents in conversations, and other systems start them in agentic executions.
Tool
A tool is a capability an agent can call while it works: searching content, reading a file, calling an external service or handing a task to another agent. Tools can come with the agent definition, from integrations such as MCP servers, and from the domain definition the agent is built on.
Conversation
A conversation is an interactive exchange between a person and an agent in the web UI. Answers are streamed as the agent writes them, and the history is saved. People can attach files for the agent to work with.
Agentic execution
An agentic execution is an autonomous run of an agent, started through the API with structured input and returning a structured result. No one takes part in the run. It happens in the background and moves through the statuses pending, in_progress and then finished or failed.
An agent definition can offer several kinds of execution, and the caller picks one with an execution_key. Executions suit work that arrives in batches or from other systems, such as processing a stream of documents.
Evaluation
An evaluation measures how well an agent does its job. A suite of plain files names the agent, the models and prompts to run it with, and a set of cases: inputs, each with what a right result looks like. Every run is scored by metrics, such as whether the output matches the expected one, which tools the agent called and how a judge model grades its work. Mocks stand in for integrations, so runs work on fixture data.
An evaluation runs an agent’s agentic executions or conversations from the command line and writes reports that compare models, prompts and code changes side by side. See Evaluate an Agent.
Integration
An integration is a configured connection to an external system on a data set. Integrations serve two purposes:
- Bringing content in. A source integration syncs content from an external system. The content is stored as entities and indexed for search.
- Letting agents act. A tool-providing integration gives agents tools for working in the external system. It syncs no content; each agent picks the ones it uses under Tool sources.
Suvereo includes a ready-made integration for MCP servers. An MCP server is an external server that exposes tools through the Model Context Protocol. Once it is connected to a data set by signing in with OAuth, its tools are available to the data set’s agents that select it under Tool sources in their settings.
Domain
A domain is a field of business with its own data, systems and workflows, such as e-commerce or healthcare. Suvereo is not tied to a domain. A domain definition describes one: the kinds of records it works with, the external systems it relies on and how agents reach them. Agents are then set up for a job within that domain.
Entity
An entity is a kind of record in a domain, such as an article or a support ticket. Suvereo defines no entities of its own: they come from a domain definition. Suvereo stores the entities, lists them in the UI and indexes their content, so agents can retrieve what is relevant to a question.
Domain definition
A domain definition adds a domain to Suvereo. It can define the domain’s entities, its integrations with external systems, and how both are handed to agents as tools. Everything else, including conversations, agentic executions, permissions, models and the UI, comes from Suvereo.
A domain definition is enabled in the instance settings, and agents are then built on top of it. Suvereo also runs without a domain definition, with agents that use the basic tools it provides and the tools of any connected MCP server.
Enthusiast is the domain definition for e-commerce. It defines products and documents.
Plugin
A plugin is a standalone package that extends Suvereo without changing it. Plugins can supply language and embedding model providers, integrations, agent definitions and domain definitions. All of them are built against shared contracts, which is how Suvereo loads them and how they stay independent of each other.