Model metadata and fallback review
Record the delivered model value for each workload and define fallback behavior before customer-facing traffic grows.
Product / Model Policy
Document GPT and Codex model values from your delivered setup.
Keep GPT-5.5 and Codex 5.5, 5.4, 5.3-style model values, fallback plans, endpoint defaults, and package assumptions visible before traffic grows.
Control plane
Model Policy
Model familyGPT + Codex
Rule scopePer endpoint
Package limit4 concurrent
Budget-aware policies
Plan lower-risk requests for efficient model choices while preserving GPT-5.5 or Codex-style options for high-value tasks.
Endpoint-level control
Prepare chat, embedding, and image policy defaults separately instead of forcing one rule across every feature.
Request visibility
Inspect model, latency, status, and estimated cost in request logs so policy choices stay measurable.
Workflow
From setup to launch prep.
Each module is designed to give developers fast integration paths while giving operators the controls they need before traffic scales.
1
Choose the workload
Start with the request shape your product uses most: chat completions, embeddings, or images.
2
Define the model policy
Record the default GPT-5.5 or Codex-style model value from your delivered setup files, fallback plan, and whether the policy should optimize for quality, speed, or cost.
3
Connect traffic intentionally
Call the delivered OpenAI-compatible base URL with an explicit model value, then review token usage, 4-connection pressure, and rate-limit behavior.
4
Review and adjust
Use request logs and usage reporting to adjust policy metadata before traffic volume grows.
Request with an explicit model value
POST /v1/chat/completions
Authorization: Bearer $UCX_API_KEY
Content-Type: application/json
{
"model": "gpt-5.5",
"messages": [
{ "role": "user", "content": "Summarize this support ticket." }
]
}Why it matters
AI products change quickly. Model policy setup keeps delivered model values visible so developers can review usage, latency, and fallback plans before access becomes a dependency.
FAQ
Questions teams ask before rollout.
Can I use a specific model from my setup files?
Yes. Send a specific GPT-5.5 or Codex-style model name in the request. Policy defaults, fallback plans, and cost preferences live in the control plane as metadata.
Do model choices affect usage reporting?
No. Requests still appear in usage and request logs with the requested model, status, latency, token count, and estimated cost.
Can policy defaults vary by endpoint?
Yes. Chat, embeddings, and images can each document different defaults, fallback plans, and cost preferences.
Ready to build?