Service capacity and availability
Web and background workloads run as containers. For a customer-managed Kubernetes environment, agree on the deployment manifests and service dependencies before installation. Size and scale each workload according to its role. Voice concurrency also depends on telephony capacity and the speech and language-model services used for each call. Define acceptance targets for response time, concurrent calls, transaction throughput, and recovery. Run representative load tests, adjust replicas, resource limits, queues, and provider capacity, and repeat the tests against the target environment. Application configuration alone is not a benchmark result. For a private deployment, include every dependency in the plan: telephony, speech recognition, speech synthesis, model inference, databases, object storage, and identity. Choose private or local endpoints where the environment requires them.Private model endpoints
The voice runtime supports configured model endpoints with a base URL, model identifier, and authentication settings. A deployment administrator can point a compatible endpoint at an organization-hosted model or an approved provider route. The service must support the streaming and tool-calling behavior used by the agent. Configure the endpoint’s required bearer token, API-key header, or subscription-key authentication and test a complete conversation, including tool calls and failures. Selecting a private language-model endpoint does not also relocate speech recognition, speech synthesis, or telephony; configure those dependencies separately.Release pipelines
Revve’s deployment workflow builds container images, runs automated integration checks, and updates deployment workloads. Establish release gates for the target environment, including code coverage, vulnerability scanning, and any organization-required analysis tools. Additional scanners can be integrated into the CI/CD pipeline; their presence in a project plan does not mean that a particular scanner has already run against a release. Record the image and configuration versions for each release. Include rollback and database-change handling in the runbook, and keep application deployment separate from publishing an agent’s conversational draft.Secrets and encryption
Choose the secret store for your environment, such as AWS Secrets Manager or Google Secret Manager. If you use External Secrets to synchronize values into Kubernetes, configure and test that integration as part of deployment. Workloads should reference the configured secrets; keep secret values out of operational documentation. Set a refresh policy and a rollout procedure for rotated credentials. Updating a Kubernetes secret does not automatically refresh an environment variable in an already-running process; restart or roll out the affected workload when required. If your infrastructure tooling stores deployment state in an AWS bucket, configure encryption, versioning, and the required KMS key and rotation policy. State protection does not cover application databases or recordings; configure those storage services independently. Use TLS on external endpoints and document certificate ownership and renewal. Requirements for a particular hardware security module, per-user signing key, or cryptographic compliance profile need an explicit integration design.Logs and monitoring
Collect application and container logs in your organization’s logging system. Loki and Grafana are possible log and dashboard components; confirm the selected stack and its configuration in the deployment plan. Preserve timestamps and service identifiers, and correlate integration requests with call and request identifiers. To connect an enterprise SIEM, configure your log collector or centralized logging layer to forward the required records to the SIEM’s supported ingestion endpoint. Agree on fields, retention, access, and alert routing with the security team. Test collector connectivity and confirm that a sample event reaches the SIEM with the expected fields. Combine application audit records with identity-provider sign-in logs and infrastructure logs for operational review. Define the event coverage required by your organization rather than assuming that one source records every action.Recovery and handover
Include the following in the environment’s runbook:- Service ownership, escalation contacts, and the support channel.
- Deployment and rollback steps, including configuration and database changes.
- Backup schedules, retention, restore procedures, and a tested recovery exercise.
- Health checks, capacity dashboards, alert thresholds, and common failure responses.
- Integration endpoints, authentication, expected responses, retry behavior, and reconciliation identifiers.
- Certificate renewal, secret rotation, and access-review procedures.