How a hybrid agent works
A hybrid agent is served by the runtime in your own cluster. Its provider uses a credential reference that your cluster resolves at request time, and its endpoint is your deployment’s ingress URL followed by/{agent_id}/v1.
Register a deployment
1
Create it
In Settings › Deployments, choose New deployment. Give it a Name. Ingress URL is needed before developers connect, and Data plane URL (HTTPS, with a trusted certificate) is needed for viewing traces; both can be added later. Choose Create deployment.
2
Copy the bootstrap token
The token is shown once and can be redeemed once, before it expires. Store it in your cluster as the
SYNC_PLANE_BOOTSTRAP_TOKEN secret.3
Deploy the sync plane
Deploy the sync plane from
deploy/k8s in the install pack. The deployment shows Connected once it checks in, in about 30 seconds.Runtime settings
Settings for Neon and model routing
Settings for Neon and model routing
These runtime settings govern guardrails and model routing in a hybrid deployment:
Limitations
- A hybrid runtime can’t load fallback credentials yet, so it skips every fallback and serves from the primary.
- A provider that uses a credential reference is saved without the connection test, because its key stays in your cluster.
- Agents on a hybrid runtime can’t be simulated.
- Keys scoped to an inline sub-agent work only against the Xenovia cloud runtime, and a hybrid agent can’t be linked as a sub-agent. See Sub-agents.
What next?
Runtime
What the runtime does with each request.
Providers
Add a provider for a hybrid agent.
Neon
What happens when Neon can’t be reached.