Skip to main content

Glossary

This page unifies the core terms used across the AI4J documentation, so the same concept is not confused across different topics.

A​

AI Foundation​

In this documentation, "AI foundation" is not a marketing term but a structured positioning.

It indicates that AI4J does more than offer single-point model calls — it places these layers into one unified system:

  • Model calls
  • Tool / Function Call
  • Skill
  • MCP
  • Upper layers: Spring Boot / Agent / Coding Agent / FlowGram

If you want to see where this layering starts, read these first:

ACP​

ACP is the host integration protocol for the Coding Agent, used by IDEs, desktop apps, or custom front-ends to communicate with ai4j-cli acp via structured JSON-RPC.

It is not a model protocol, and it is not MCP.

Related docs:

Agent​

In AI4J, Agent refers to an agent framework built around model, runtime, tools, memory, and orchestration.

Related docs:

AiService​

AI4J's unified service factory, used to obtain service interfaces such as Chat, Responses, Embedding, Audio, Image, and Realtime by PlatformType.

Related docs:

C​

Chat​

Refers to the Chat Completions-style, message-based model interface.

In AI4J it typically corresponds to:

  • IChatService
  • ChatCompletion
  • ChatCompletionResponse

CodeAct​

A code-driven runtime in the AI4J Agent.

Suitable for:

  • Generating code first
  • Then calling tools multiple times through code
  • Handling complex structured tasks

Related docs:

Coding Agent​

An engineering entry point in AI4J aimed at local repository delivery, comprising:

  • CLI
  • TUI
  • ACP
  • Sessions, commands, tools, Skills, and MCP integration

It is not a synonym for a general-purpose Agent framework, but rather a product layer biased toward "local coding interaction".

Related docs:

F​

Function Call​

Function Call refers to the call semantics by which a model can select and invoke a local capability according to a tool schema.

In the AI4J context, it is usually the entry point through which most users first understand the Tool track.

Related docs:

FlowGram​

A low-code workflow orchestration integration direction provided by AI4J, aimed at flowchart-style node execution, backend task execution, and node extension.

Related docs:

Function Tool​

A local Java function tool, exposed to the model via annotations or registration.

It differs from an MCP Tool in that:

  • A Function Tool usually lives directly inside the local application
  • An MCP Tool comes from an MCP Server

G​

Gateway​

In the MCP context it usually refers to McpGateway, used to manage multiple MCP Clients uniformly, aggregate tools, and apply routing governance.

Related docs:

M​

MCP​

Model Context Protocol, the standard protocol layer for a model to reach external capabilities.

In AI4J it covers:

  • MCP Client
  • MCP Gateway
  • MCP Server

Related docs:

Memory​

The context memory mechanism an Agent or Coding Agent uses within a persistent session.

It usually includes:

  • History messages
  • Tool call records
  • Compaction summaries
  • checkpoint

Model Client​

The model adaptation interface at the Agent layer, used to turn the AgentPrompt built by the runtime into a concrete model request.

Common implementations:

  • ChatModelClient
  • ResponsesModelClient

P​

PlatformType​

The platform enum in AI4J, used to declare which model platform you are calling.

For example:

  • OPENAI
  • DOUBAO
  • DASHSCOPE
  • OLLAMA

Profile​

In the Coding Agent, a provider profile is a reusable combination of model configuration, for example:

  • provider
  • protocol
  • model
  • baseUrl
  • apiKey source

Related docs:

Prompt Assembly​

In the Coding Agent context, this refers to how the final context sent to the model is composed from:

  • systemPrompt
  • workspace instructions
  • instructions
  • session memory
  • current input
  • tool schemas

Related docs:

R​

ReAct​

The default general-purpose runtime of the AI4J Agent, suitable for:

  • Text tasks
  • Multi-turn reasoning
  • Calling tools on demand

Related docs:

Responses​

Refers to the event-based response model interface.

In AI4J it typically corresponds to:

  • IResponsesService
  • ResponseRequest
  • Response
  • ResponseSseListener

The difference from Chat is not only that the interface name differs — the event model is more powerful.

S​

Session​

A persistent session instance.

In the Coding Agent, a session usually contains:

  • Current context
  • History events
  • Branch relationships
  • In-memory compaction info
  • Process state

Skill​

A Skill is first and foremost an instructional-asset capability in the AI4J foundation, further productized inside the Coding Agent.

It usually takes the form of a SKILL.md.

It is not a tool protocol, but rather a task instruction, template, or workflow guidance that the model can read and reuse on demand.

Related docs:

StateGraph​

The state-graph orchestration capability within an Agent Workflow, suitable for branching, loops, and conditional routing.

Related docs:

Stream​

Means the model response arrives incrementally, rather than being returned all at once as a single payload.

note

Keep in mind:

  • A streaming event is not the same as a token
  • Different platforms have different chunk granularity

T​

Tool Registry​

The registration layer that decides "which tools are exposed to the model".

It differs from ToolExecutor:

  • ToolRegistry decides visibility
  • ToolExecutor decides how a tool is executed

Trace​

Refers to the process-observation capability of an Agent or Coding Agent.

Usually used to record:

  • Model calls
  • Tool calls
  • Per-step latency
  • Errors and fallbacks

Related docs: