Feature
Fallbacks
Prepare backup model and provider routes for important workloads before the primary path fails.
Primary
Provider A
Backup
Provider B
Policy
Configured
Primary and backup routes for each workload.
Support
Claude direct
Primary route for summaries.
Backup
Claude on Azure
Fallback route for the same workload.
App
One request
Fallback behavior stays outside the handler.
New capabilities
What your team gains with Concentrate
A backup route, ready in advance
Pick the approved model or provider a workload should fail over to, so an outage triggers an automatic reroute instead of a page in the middle of the night.
No fallback branches in your code
You never write per-provider retry or branching logic. List your backup models or providers in the request's routing config, and Concentrate handles the reroute.
Test the backup before you trust it
Compare a backup's output shape, latency, and cost against the primary before depending on it, so failover doesn't quietly change what your app returns.
Who Concentrate is designed for
How fallbacks keep AI features online when a provider fails
An LLM fallback is a backup model or provider route used automatically when the primary route fails, times out, or slows down. Provider outages and rate limits are routine, so a single-provider setup means a single point of failure. Fallbacks define the backup ahead of time and reroute traffic at the gateway, so a degraded provider becomes a config-defined recovery instead of a customer-facing incident.
Primary and backup, per workload
Each important workload gets a primary route and an approved backup, so the response stays in a shape the app already expects.
Reroute happens at the gateway
Failover logic lives in Concentrate, not in each service, so apps don't carry provider-specific error handling.
Validate before you depend on it
Compare the backup's output, latency, and cost against the primary so failover doesn't silently change behavior under load.
Declared in your request
List the backup models or providers to fail over to in your request's routing config — Concentrate runs the failover. No custom fallback logic, and no provider-specific error handling.
Feature basics