Federico Mariani created CAMEL-23954:
----------------------------------------
Summary: camel-langchain4j-agent: moderation support
(ModerationModel + @Moderate)
Key: CAMEL-23954
URL: https://issues.apache.org/jira/browse/CAMEL-23954
Project: Camel
Issue Type: New Feature
Components: camel-langchain4j-agent
Reporter: Federico Mariani
LangChain4j supports content moderation on AI services via
{{AiServices.moderationModel(...)}} combined with the {{@Moderate}} annotation
on the service interface methods: the moderation call runs concurrently with
the chat call, and flagged input raises {{ModerationException}} before the
answer is returned.
The agent component exposes neither. Unlike most builder options this is not a
pure {{AgentConfiguration}} addition: {{@Moderate}} must be present on the
internal service interfaces ({{AiAgentWithMemoryService}} /
{{AiAgentWithoutMemoryService}}), so it needs either annotated method variants
or separate moderated service interfaces, selected when a {{moderationModel}}
is configured.
*Use case*: a public-facing support chatbot exposed via {{platform-http}}.
Untrusted end-user text goes straight into an agent with tool access; the
operator must reject policy-violating input (harassment, self-harm,
prompt-injection bait wrapped in abusive content) *before* the agent spends
tokens and invokes tools, and must return a controlled refusal instead of a
stack trace:
{code:java}
// AgentConfiguration bean
new AgentConfiguration()
.withChatModel(chatModel)
.withModerationModel(OpenAiModerationModel.builder().apiKey("{{openai.key}}").build());
// Route
onException(ModerationException.class)
.handled(true)
.setHeader(Exchange.HTTP_RESPONSE_CODE, constant(422))
.setBody(constant("Sorry, your message violates our usage policy."));
from("platform-http:/support/chat")
.to("langchain4j-agent:support?agentConfiguration=#supportAgentConfig&tags=orders,billing");
{code}
*Proposal*: {{AgentConfiguration.withModerationModel(ModerationModel)}} +
moderated service interface variants + {{ModerationException}} surfaced as the
Camel exchange exception with a dedicated header carrying the flagged content
details.
_This issue was drafted by Claude Code on behalf of Federico Mariani._
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
