[
https://issues.apache.org/jira/browse/CAMEL-23965?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18103067#comment-18103067
]
Karol Krawczyk commented on CAMEL-23965:
----------------------------------------
PR: https://github.com/apache/camel/pull/25410
Implemented as proposed, with the pass-through body plus verdict headers:
- {{CamelOpenAIModerationFlagged}} is always a Boolean (for a batch: true when
at least one input was flagged), so the same content-based routing works for a
single input and for a list.
- {{CamelOpenAIModerationCategories}} / {{CamelOpenAIModerationCategoryScores}}
are a Map for a single input and a List of maps for a batch, mirroring
{{CamelOpenAIOriginalText}} in the embeddings operation.
- New {{moderationModel}} option, default {{omni-moderation-latest}},
overridable per message with {{CamelOpenAIModerationModel}}.
- {{storeFullResponse=true}} stores the SDK response in a dedicated
{{CamelOpenAIModerationResponse}} property rather than reusing
{{CamelOpenAIResponse}}, whose declared javaType is ChatCompletion.
Two notes worth flagging:
- The issue was written against openai-java 4.41.0, but the project is now on
4.49.0. The API used here was verified against 4.49.0.
- {{Moderation.Categories.illicit()}} / {{illicitViolent()}} are Optional,
because only the omni-moderation-* models return them, while the same
categories are plain doubles in CategoryScores. The category map therefore
omits them for legacy models; documented and covered by a test.
There is no integration test: the ITs in camel-openai run against Ollama via
testcontainers, and Ollama does not expose /v1/moderations. Instead,
camel-test-infra-openai-mock was extended with the /moderations endpoint, and
the operation is covered by 10 mock-based test cases.
[~croway] could you review it, since you reported this one? I cannot request
reviewers on GitHub myself.
_Reported by Claude Code on behalf of Karol Krawczyk_
> camel-openai: add moderation operation
> --------------------------------------
>
> Key: CAMEL-23965
> URL: https://issues.apache.org/jira/browse/CAMEL-23965
> Project: Camel
> Issue Type: New Feature
> Components: camel-openai
> Affects Versions: 4.21.0
> Reporter: Federico Mariani
> Priority: Major
> Labels: ai
>
> The pinned openai-java SDK (4.41.0) exposes
> {{moderations().create(ModerationCreateParams)}}, but camel-openai has no
> moderation operation. Moderation is the canonical pre-filter for untrusted
> input on public-facing AI routes — rejecting policy-violating content
> *before* spending chat tokens or triggering tool calls — and fits Camel's
> routing model perfectly (content-based routing on the moderation verdict).
> Proposal: new operation {{openai:moderation}}:
> * body: String (or List of Strings for batch);
> * options/headers: {{moderationModel}} (e.g. {{omni-moderation-latest}});
> * output: body unchanged (pass-through style) with headers
> {{CamelOpenAIModerationFlagged}} (Boolean),
> {{CamelOpenAIModerationCategories}} (Map or JSON),
> {{CamelOpenAIModerationCategoryScores}}; full response via
> {{storeFullResponse}}.
> Pass-through body + verdict headers lets users do content-based routing on
> the verdict without content juggling:
> {code:java}
> // guard a public-facing chat route: reject flagged input before spending
> chat tokens
> from("platform-http:/chat")
> .to("openai:moderation?moderationModel=omni-moderation-latest")
> .choice()
> .when(header(OpenAIConstants.MODERATION_FLAGGED).isEqualTo(true))
> .setBody(constant("Your message violates our usage policy."))
> .otherwise()
> .to("openai:chat-completion?model=gpt-5")
> .end();
> {code}
> Cross-reference: CAMEL-23954 proposes moderation for camel-langchain4j-agent
> via LangChain4j {{@Moderate}}; a native operation in camel-openai is simpler
> (no annotated-interface machinery) and would give the OpenAI stack parity.
> _This issue was drafted by Claude Code on behalf of Federico Mariani_
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
