Keep the interface you know.
Send OpenAI Chat Completions or Anthropic Messages requests. Use the supported text and tool-call formats with your existing SDK.
INFERENCE INFRASTRUCTURE EARLY ACCESS / 01
Every request.
A better route.
One interface for your models. Route across providers, handle failures, and see what each request actually costs.
OpenAI & Anthropic request formats. MCP tools for agents.
ROUTING ENGINEINTERACTIVE / 3D
X 032 / Y 018 / Z 064
Groq route selected in this illustration.
Concept view · not live traffic. Availability depends on configured providers.
01 Compatible requests02 Provider fallback03 Measured usage
BUILT FOR THE REQUEST PATH
Less plumbing.
More control.
Keep your application focused on the work. Bring routing, credentials, and request history into one place.
API COMPATIBILITY
A failed provider is a decision.
Eligible providers are ranked for the requested model. If a provider fails before streaming content, the router can try the next eligible route.
Know where the tokens went.
Review request attempts, provider selection, latency, and recorded usage. Missing token usage stays unknown instead of becoming a made-up number.
SMALL CHANGE. CLEAR BOUNDARY.
Your model.
Your code.
One entry point.
Create a key, set your router URL, and make your first request. The dashboard gives you the history to inspect what happened next.
Connect an MCP clientfrom openai import OpenAI
client = OpenAI(
base_url="YOUR_ROUTER_URL/v1",
api_key="YOUR_OMNIDEPLOY_KEY",
)
response = client.chat.completions.create(
model="gpt-oss-120b",
messages=[{
"role": "user",
"content": "Explain your routing decision."
}],
)
print(response.choices[0].message.content)Replace the URL and key with your workspace credentials.
START WITH ONE REQUEST
Build first.
Measure as you go.
OmniDeploy is in early access. Model availability and usage costs depend on the providers configured for your workspace.
Check your workspace for access and billing details before sending traffic.
Create an account