Switching AI gateways?Here's what moves, and what doesn't.

One name in model, one workflow behind it

What a workflow bundles, and why model is its name.

2 lines of code + 1 workflow

Keep your OpenAI SDK. On the client line, change the base URL and key. On the request line, set model to the name of a workflow.

Step 0: create a workflow in the console with your primary model and backups. Its name is what you send as model.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://your-gateway.example/v1",
    api_key=os.environ["GATEWAY_API_KEY"],
)

resp = client.chat.completions.create(
    model="openai/gpt-4o",
    messages=[{"role": "user", "content": "Summarize this ticket."}],
)

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.modelriver.com/v1",
    api_key=os.environ["MODELRIVER_API_KEY"],
)

resp = client.chat.completions.create(
    model="support_bot",  # workflow name
    messages=[{"role": "user", "content": "Summarize this ticket."}],
)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://your-gateway.example/v1",
  apiKey: process.env.GATEWAY_API_KEY,
});

const resp = await client.chat.completions.create({
  model: "openai/gpt-4o",
  messages: [{ role: "user", content: "Summarize this ticket." }],
});

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.modelriver.com/v1",
  apiKey: process.env.MODELRIVER_API_KEY,
});

const resp = await client.chat.completions.create({
  model: "support_bot", // workflow name
  messages: [{ role: "user", content: "Summarize this ticket." }],
});

curl https://your-gateway.example/v1/chat/completions \
  -H "Authorization: Bearer $GATEWAY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "openai/gpt-4o", "messages": [{"role": "user", "content": "Summarize this ticket."}]}'

curl https://api.modelriver.com/v1/chat/completions \
  -H "Authorization: Bearer $MODELRIVER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "support_bot", "messages": [{"role": "user", "content": "Summarize this ticket."}]}'

What maps to what, by platform

• OpenRouter

  Your models[] fallbacks and @preset/ configs become workflows. Your own provider keys replace credits.

  model: "openai/gpt-4o"

  Workflow name

  models: [...]

  Up to 2 backup models

  @preset/slug

  Workflow with Draft and Live

  Moving from OpenRouter

• LiteLLM

  Each model_name alias and its fallbacks become a workflow. No proxy to run.

  model_name alias

  Workflow name

  fallbacks: [{"a": ["b"]}]

  Up to 2 backup models

  os.environ/OPENAI_API_KEY

  Provider credential in the project

  Moving from LiteLLM

• Portkey

  The fallback targets in your x-portkey-config become one workflow. No gateway headers on the request.

  @openai-prod/gpt-4o

  Workflow name

  strategy.mode: "fallback" + targets

  Up to 2 backup models

  x-portkey-config header

  Nothing: the name in model is enough

  Moving from Portkey

• Vercel AI Gateway

  Model fallbacks and vmc/ virtual models become workflows. BYOK without buying credits.

  anthropic/claude-sonnet-5

  Workflow name

  vmc/<slug>

  Workflow with Draft and Live

  providerOptions.gateway.models

  Up to 2 backup models

  Moving from Vercel AI Gateway

Side by side, in plain text

How each one is set up. Checked against each platform’s own docs; see Sources.

What ModelRiver doesn’t do yet

If you need one of these today, keep it where it is.

• Unified billing or credits. You bring a key for each provider.

• Per-request model strings or an auto-router. Every call names a workflow you created first.

• Load balancing, key rotation, and routing by latency or cost.

• Circuit breaking, configurable retries or timeouts. Retries cover dropped connections; timeouts are 120 seconds.

• Semantic caching. Response caching is exact-match only.

• Output, PII or prompt-injection guardrails. Guardrails check input only.

• Image, audio and vision requests, the Responses API and the Anthropic Messages API.

• Embeddings from providers other than OpenAI, Google, Cohere and Mistral.

• More than 2 backup models per workflow.

• A catalog beyond about 139 models across 13 providers. Custom OpenAI-compatible base URLs are supported.

• SSO, SAML or SCIM, audit logs, OpenTelemetry and log export.

• Self-hosting, or a choice of region.

• Third-party compliance attestations and a contractual SLA. We target 99.9% uptime and publish status at status.modelriver.com; a DPA covering GDPR, UK GDPR and CCPA is available.

Related

• Gateway migration docsConcept mapping, snippets and first errors.