ARGI Agent Framework

What's Agent Framework

ARGI Agent Framework is created for Java developers to quickly and easily building, orchestrating, and deploying AI agents. This framework is built upon the ReactAgent concept which features a Reasoning-Acting loop while at the same time supports multi-agent orchestration.

Core Features

• ReactAgent
• Multi-Agent Orchestration
• Context Engineering
• Human In The Loop
• A2A
• Rich Model, Tool and MCP Support

Model Call Error Handling

By default, exceptions caught during a synchronous model call or synchronous streaming setup are converted into an AssistantMessage prefixed with Exception:. To propagate those exceptions instead, opt in when building an agent:

ReactAgent agent = ReactAgent.builder()
    .name("assistant")
    .model(chatModel)
    .throwOnModelError(true)
    .build();

throwOnModelError defaults to false and is also available on AgentLlmNode.builder(). When enabled, model interceptors receive the original exception with its type and cause intact; graph execution may wrap it while preserving the cause chain. This allows applications to distinguish model failures from normal responses without parsing error text.

This setting does not add retries or change tool exception handling. Errors emitted by a streaming publisher after setup continue to propagate in either mode, including errors after partial output. Existing retry interceptors can still retry or wrap exceptions according to their own configuration.

Related Projects

ARGI Agent Framework depends on the following projects:

• argi-graph-core: https://github.com/agentic-ai-java/argi/tree/main/argi-graph-core
• spring-ai-extensions: https://github.com/agentic-ai-java/argi-extensions
• spring-ai: https://github.com/spring-projects/spring-ai