Failover
When Edgaze switches to a backup model, your price stays the same.
Buyers running workflows and creators configuring failover.
Overview#
When a primary model degrades, Edgaze can route the run to an equivalent model on another provider. You pay the pinned displayed price only. Any extra compute cost is absorbed by Edgaze, not added to your bill.
Creators can disable failover. When off, degradation causes the run to fail instead of switching models. You are not charged margin on that failure.
Failover enabled#
The run starts against the primary model. If degradation is detected, execution routes to a fallback provider. Your displayed price does not change. Edgaze covers the cost difference internally.
On success after failover, the creator still earns 80% of their margin. You pay the same pinned price as a run that never failed over.
Failover disabled#
If the creator opted out in publish settings, degradation causes run failure. Margin is not charged. Compute absorption follows the standard failure policy.
Special cases#
- Failover then failure: No margin charge. Absorbed compute includes the failover segment.
- BYOK graphs: Provider keys bill tokens to your account. The orchestration fee still applies per BYOK policy.
- API and bundle paths: Same failover billing rules as UI wallet runs.
What you see in Edgaze#
- Publish settings: failover toggle for creators
- Buyer UI: no visible price change when failover occurs internally
- Failure with failover off: standard failure message, no margin charge
Related policies#
Was this useful?
Your response helps us improve the documentation.
Related documentation
How workflow runs are priced, funded, settled, and refunded. Step-by-step guides for buyers, creators, and developers.
How to open a workflow, fund a run, and what happens when it completes.
How the per-run price is built, what the creator earns, and what Edgaze keeps.