Configuration
Kertex AI is a transparent proxy, and that single fact explains how configuration works: whatever generation settings your application sends on a request are forwarded to the model exactly as you set them. Kertex does not inject or override them.
What you control — per request
Set these in your SDK request (or in the Playground for quick tests). They pass straight through to the model:
- Model / tier — the alias you call (
kertexai-fast,kertexai-advanced,kertexai-light,kertexai-embed). - Temperature, top-p, top-k — sampling controls.
- Max output tokens — response length ceiling.
- System instruction — the system prompt / persona.
- Tools / function calling — tool declarations and tool results.
- Response schema / JSON mode — structured-output controls.
- Safety settings — within Google's provider policy.
Example (Python GenAI SDK):
from google.genai import types
resp = client.models.generate_content(
model="kertexai-advanced",
contents="Draft a release note for v2.1.",
config=types.GenerateContentConfig(
temperature=0.3,
max_output_tokens=800,
system_instruction="You write concise, friendly product release notes.",
),
)
Because configuration travels with each request, two different calls can use completely different settings against the same environment and token.
What Kloudbean manages
Some things are handled for you and are shown read-only in the environment's Configuration tab:
- The Google Cloud project, region, and service-account credentials.
- The underlying model each tier alias maps to (so tiers can be upgraded without breaking your code).
- The safety baseline and data-handling posture.
The Configuration tab reflects these managed defaults for transparency; you change runtime behaviour by setting parameters in your request, not by editing the tab.
Knowledge / RAG
Grounded retrieval over your own documents (RAG) is provisioned by Kloudbean rather than self-serve. If an environment hasn't had it enabled, the Knowledge tab shows a "not enabled" state. Contact Kloudbean to enable grounding for an environment.
Next steps
- Developer Guide → — full request examples that use these settings.
- Limits & Alerts → — the spend and rate controls that are enforced by the gateway.