Guillaume Nodet created CAMEL-23728:
---------------------------------------
Summary: camel-jbang - camel ask returns 404 when using Vertex AI
endpoint
Key: CAMEL-23728
URL: https://issues.apache.org/jira/browse/CAMEL-23728
Project: Camel
Issue Type: Bug
Components: camel-jbang
Reporter: Guillaume Nodet
Fix For: 4.x
When using `camel ask` through Google Cloud Vertex AI, the LLM client sends the
model name as-is (e.g., `claude-sonnet-4-6`) in the Vertex AI URL path:
{code}
/v1/projects/<project>/locations/global/publishers/anthropic/models/claude-sonnet-4-6:rawPredict
{code}
This returns a 404 because Vertex AI expects a versioned model identifier
(e.g., `claude-sonnet-4-6@20250514`).
The HTML error response from Google is dumped directly to the terminal:
{code}
LLM returned status: 404
<!DOCTYPE html>
<html lang=en>
...
<p><b>404.</b> <ins>That's an error.</ins>
<p>The requested URL <code>/.../models/claude-sonnet-4-6:rawPredict</code> was
not found on this server.
{code}
Two issues:
1. *Model name mapping* — LlmClient should map model names to Vertex
AI-compatible identifiers when the endpoint is a Vertex AI URL
2. *Error handling* — Non-JSON error responses (like HTML 404 pages) should be
parsed gracefully instead of dumping raw HTML to the user
*Workaround*: Use the Anthropic API directly with `--api-type=anthropic
--api-key=$ANTHROPIC_API_KEY`
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
