A2A Artifact

What is an A2A Artifact?

An A2A Artifact is an output generated by a remote agent as the result of an [A2A Task](/glossary/a2a-task/). It is composed of one or more content parts, which can carry text, files, or structured data.

Artifacts represent deliverables rather than conversational turns. A report, generated image, code archive, table of findings, or structured booking option can all be modeled as an artifact when produced for a task.

How it works

An A2A server associates artifacts with a task. Clients can receive them in a task response or through artifact update events when the agent supports incremental delivery. The protocol provides a shared representation; it does not prescribe how the remote agent created the output.

A task can produce more than one artifact. For example, a research task might return a summary document, a source-data file, and a chart as separate outputs. Artifact names and descriptions can help a client present them clearly.

Simple example

A design agent receives an A2A task to prepare campaign assets. It generates two artifacts: a presentation file and a structured JSON file containing approved copy. Status messages describe progress, while the artifacts carry the final deliverables.

If the client asks for a revision, the continued task may produce an updated artifact according to the server’s behavior.

A2A Artifact versus tool result

A [tool result](/glossary/tool-result/) is the response returned from a particular tool invocation inside an agent execution flow. An A2A Artifact is a task-level output shared across an A2A boundary.

An agent might call several tools, inspect their results, and synthesize one artifact. A raw search response used internally is usually a tool result; the completed research brief delivered to the requesting agent is an artifact.

Artifacts also differ from [A2A messages](/glossary/a2a-message/). Messages carry communication turns, including questions and explanations. Artifacts carry outputs intended as task deliverables.

Why it matters

Artifacts give agent systems a stable way to exchange rich results without reducing every output to chat text. They support clear product presentation, storage, auditing, and downstream processing.

Clients should validate file types, sizes, schemas, provenance, and access rights. A protocol-defined artifact is still external content and should not automatically be trusted or executed.