Agent Inspector

See what your Spring AI app sends to the model and what comes back: requests, responses, tool calls, tokens and timing, live in your browser. Works with OpenAI, Claude (Anthropic) and Docker Model Runner.

Use it in your app

For Spring Boot 4 apps that use Spring AI 2.0, on Java 21 or later.

1. Add the starter to your pom.xml. It's on Maven Central, so nothing else is needed:

<dependency>
    <groupId>io.github.osman-butt</groupId>
    <artifactId>agent-inspector-spring-boot-starter</artifactId>
    <version>0.2.1</version>
</dependency>

2. Run your app and open http://localhost:7070. The app's log shows the address:

Agent Inspector UI: http://localhost:7070 (running in this app)

That's it. Nothing else in your app changes.

To try the latest unreleased version, use 0.3.0-SNAPSHOT and add Central's snapshot repository:

<repositories>
    <repository>
        <id>central-snapshots</id>
        <url>https://central.sonatype.com/repository/maven-snapshots/</url>
        <releases><enabled>false</enabled></releases>
        <snapshots><enabled>true</enabled></snapshots>
    </repository>
</repositories>

What you see

• Sessions: each question your app handles, with every round-trip to the model.
• Messages and tool calls: what was sent, what the model answered, which tools it called and how long they ran.
• Spring AI advisors: what chat memory and other advisors added to your prompt.
• The raw JSON of every call, one click away, and Copy as cURL to run a call again.

Settings

Property	Default
agent.inspector.enabled	true	false turns it off
agent.inspector.url	http://localhost:7070	where the inspector runs
agent.inspector.embedded.enabled	true	false to only use an inspector that is already running
agent.inspector.embedded.keep-alive	false	true keeps the in-app inspector running after a command-line app finishes

Good to know

• It is for development: leave it out of production, or set agent.inspector.enabled=false there.
• A command-line app should set spring.main.web-application-type=none (a WebFlux app reactive), because the starter brings a web server for the inspector.
• An inspector started inside a command-line app stops when the app has nothing left running, so the app exits on its own. Set agent.inspector.embedded.keep-alive=true to keep it up and browse the results after the run.
• If you build your own ToolCallingManager, pass it .observationRegistry(observationRegistry) to see how long tools take.