Your connection details
Name, enabled state, URL, and authentication stay editable for manual Endpoints you own.
Hybrid RPC
Run your own node without giving up a backup
Register a self-hosted node as a manual Endpoint, order it ahead of vendor Endpoints in a Route, and keep one API Key in the application.
A self-hosted node is just another Endpoint
Create a manual Endpoint with the chain, network, and protocol your node actually serves; those three fields are fixed after creation. The URL must use http or https and pass the Gateway's outbound destination policy. Authentication covers no authentication, Bearer, an API Key in a named header, in a query parameter, or in a URL path, so a node behind your own reverse proxy fits without changing how it is exposed.
Secrets stay structured
Credentials belong in the authentication fields rather than in application source or plain URLs.
Checked on demand
A manual health check runs the protocol probe and records whether the upstream answered.
Order it ahead of the vendors
Put the node first on a priority Route and leave vendor Endpoints behind it. The Route uses your node while it is usable and advances to the next target when the node is filtered out, rejected by its Circuit, or returns a retryable failure. Vendor-synchronized Endpoints from the supported integrations sit in the same Account inventory as your manual one, so the ordering decision is the only thing you configure.
Protection when the node struggles
Each Endpoint has its own Circuit, tracked per Endpoint version and per workload class, and debug and trace family methods carry a separate Circuit from ordinary traffic. Overloading a node with tracing therefore does not block plain reads on the same target. While a Circuit is open the Route skips that Endpoint entirely; after a cooldown it admits one probe, closes on consecutive successes, and reopens with a longer cooldown if the probe fails.
Per-Endpoint isolation
One struggling target is skipped without changing how the other targets on the Route are treated.
Separate trace class
Heavy debug and trace work has its own Circuit, so it does not close the Endpoint to normal reads.
Probing recovery
Recovery is gradual: one probe first, then normal traffic once the upstream answers consistently.
The application keeps one integration
Your code holds an App API Key and an Access Point, not a node URL. Moving traffic between your own node and a vendor Endpoint is a Route change, so adding a backup, taking the node out for maintenance, or rotating an upstream credential does not require a redeploy. Upstream credentials stay in the Endpoint's structured authentication fields rather than in application configuration, and deleting a manual Endpoint removes it from every Route it was attached to in one operation.
What the Gateway will not decide for you
Routing does not read chain tip or block height, so a node that is responsive but lagging keeps receiving traffic until it fails outright. Health status and latency are observational, idle Endpoints carry no fresh observation because there is no scheduled probing, and neither value feeds selection. Method coverage, archive depth, and vendor namespaces are yours to validate against the upstream, since a successful probe only proves the connection works.
Next step
Add your node as an Endpoint
Create a manual Endpoint for the node, run a health check against it, then place it first on the Route that already points at a vendor Endpoint.