Model Routing & Failover

Intelligent model routing and failover

Set defaults by team, route by complexity, and failover the instant a provider goes down, so every request lands on the right model.

Maintain three levels of control for every request.

The Model Gateway decides which model serves each request, by team, by complexity, and by which providers are healthy right now.

Set a default model for every team.

Give each team the model that fits how they work: Sales starts on Claude Sonnet, Engineering on GPT-5, Product on Claude Opus. Set it once in the gateway; every agent that team runs inherits it, and you can change it for everyone in one place.

per-team defaults
inherited by every agent
changed in one place
Default model by team editing
TEAM DEFAULT MODEL ON
Sales Claude Sonnet
Engineering GPT-5
Product Claude Opus · med
Support Claude Haiku
set once per team inherited by every agent

What intelligent routing gets you

Every request on the right model, a lower bill by default, and uptime that survives a provider outage.

01 Right model

The right model, without anyone choosing.

Defaults by team and complexity-aware routing mean each request reaches the model that fits it, no engineer wiring model selection into every agent.

02 Lower cost

A lower bill, by default.

When a cheaper model can return the same correct answer, routing sends the request there automatically, capturing a spread of up to 178× between models.

03 Uptime

100% uptime through an outage.

Continuous health checks and instant failover keep agents answering when a provider degrades, with context intact, so the answer never changes.

CData Labs Study
178×

The same correct, safe answer—at up to 178× less cost, depending only on the model that served it.

We tested 22 models, from budget to frontier, on more than a thousand real questions against live enterprise systems. With Connect AI's data layer handling accuracy and safety, every model got the same right answer. The only difference was the price.

Read the Study

Routing that knows your data, not just your models

Defaults, complexity routing, and failover run in the layer that carries your context, so the model can change without the answer changing.

Team defaults · applied 4 teams
TEAMDEFAULT MODELON
SalesClaude Sonnet
EngineeringGPT-5
ProductClaude Opus · med
set onceapplied everywhere

Change a team’s model in one place; every agent it runs picks it up.

Cost per correct answer, economy vs frontier
Economy model
1×
Mid-tier model
~40×
Frontier model
178×
Same correct answer — up to 178× apart on price alone.

CData Labs, 22 models against live Salesforce, Snowflake, and ServiceNow.

Provider health & failover monitoring
Anthropic · Claude healthy
OpenAI · GPT-5 offline
Google · Gemini healthy
Rerouted · context intact
continuous health checksinstant failover

21 of 22 models returned identical results through the same governed tools, so failover doesn’t change the answer.

FAQ

Questions teams ask about routing.

  • How does the gateway decide which model runs?
  • What happens when a provider goes down?
  • Do we lose context when a request switches models?
  • How do we add or swap a model?
  • Does routing on cost change the result?
  • Do our provider keys get exposed to agents?
  • What does migrating to the gateway take?

Route smarter. Never go down.

Set defaults, route by complexity, and fail over across providers, all from the Model Gateway, with your context intact.