KServe provider can now serve models packaged as OCI images
The KServe provider can serve LLMs packaged as container images via oci:// modelcar or oci+native:// image volumes, with node-level caching and private registry support.
kserve
llm
releases
Latest OpenEverest releases, features, and announcements
The KServe provider can serve LLMs packaged as container images via oci:// modelcar or oci+native:// image volumes, with node-level caching and private registry support.
The KServe provider now exposes vLLM engine parameters — tool calling, reasoning parsers, context length, quantization and more — without hand-written config.
The TiDB provider's latest releases add an HTAP analytics preset, S3-compatible backups for more storage providers, and a series of reliability fixes for Secrets, DNS, and small-instance sizing.
OpenEverest Valkey Instances can now be scraped by Prometheus and/or VictoriaMetrics, with no extra exporter setup.
The OpenEverest TiDB provider can now deploy TiProxy as an optional component for stable, proxy-managed connections to TiDB clusters.
plugin-metrics v0.2.2 adds pod CPU and memory panels plotted against resource limits on every dashboard, plus a prebuilt SeaweedFS dashboard and Explore fixes for VictoriaMetrics.
provider-seaweedfs v0.2.0 secures component traffic with mTLS, requires S3 credentials by default, and ships replication policies, storage expansion, disruption budgets, and metrics.
Percona Server for MongoDB provider v0.5.0 brings TLS with cert-manager, native Prometheus and VictoriaMetrics exposure, richer backup imports, UI schema validation, scheduling controls, and MongoDB 8.0 as the default.
AI Gateway API keys in the KServe provider are now scoped to their own namespace, and key collisions fail closed instead of leaking across namespaces.
A new timescaledbCommunity toggle enables TimescaleDB Community Edition — compression, continuous aggregates, and retention policies — on CloudNativePG instances, with PostgreSQL 18.6 as the new default.
CloudNativePG instances gain required TLS for clients, synchronous replication, separate WAL storage, time-based backup retention with deletion, and read endpoints in the connection Secret.
You can now restore another instance's backup — or any point in time within its retention window — into a brand-new PostgreSQL instance.
The KServe provider now publishes TypeSafe decision models through the Envoy AI Gateway, so model traffic flows through the gateway like any other workload.
The Inspector plugin's pod list is now organized by instance component, with each component's Services shown inline.
Percona XtraDB Cluster instances can now be configured with pod scheduling policies — including affinity — directly from the OpenEverest UI.
The TiDB provider now creates a PodMonitor and a VMPodScrape, exposing metrics from every TiDB component to Prometheus and VictoriaMetrics.
The TiDB provider issues TLS certificates through cert-manager and encrypts all TiDB traffic, with backups running as a provider-managed account.
The Altinity ClickHouse provider can now require TLS for client connections, encrypts distributed queries between replicas, and secures replication, parts fetches, and Keeper traffic between pods.
Rolling restarts now wait for your approval, upgrades run preflight checks before anything is applied, Ready reflects the actual rollout state, and deleted instances retain their data by default.
The OpenEverest metrics plugin automatically discovers VictoriaMetrics operator monitors, so Prometheus-class databases are scraped without extra configuration.
The MariaDB provider now creates a VMServiceScrape, so MariaDB metrics land in VictoriaMetrics without any manual setup.
The AI Gateway in provider-kserve no longer rejects batched embeddings, long RAG prompts, or long completions; its Envoy buffer limit is now configurable and defaults to 50 MiB.
Prometheus monitoring for Valkey clusters can now be enabled or disabled directly from the topology UI.
You can now choose the StorageClass used by the bundled dependency volumes when installing provider-milvus.
PMM metrics exported by the Percona PostgreSQL, Percona Server for MongoDB, and Percona XtraDB Cluster providers now carry the instance and namespace labels.
Replication topologies can now be scaled out, and broken replicas are recovered automatically.
New MariaDB instances can be seeded directly from an existing logical backup by setting spec.dataSource.
Embedding models can now be served through the Envoy AI Gateway, giving embedding traffic the same routing and management as other model endpoints.
Predictors can serve any model packaged as a container image, not just the models from the built-in catalog.
pgvector, pgAudit, TimescaleDB, PostGIS, pgRouting, pg_ivm, wal2json, and Apache AGE can now be enabled straight from the instance spec.
Turning on Monitoring → Prometheus creates a PodMonitor that scrapes CloudNativePG instance metrics on port 9187 — no hand-written scrape configs.
KServe instances gain predictor topology in the UI, CR, and API, unlocking Hugging Face embedding models on both CPU and GPU, with a model catalog and Prometheus monitoring built in.
You can now define pod scheduling policies for Percona PostgreSQL clusters, including affinity and placement rules.
Exposed PgBouncer endpoints now publish the correct address — the load balancer address when exposed, the direct host when PgBouncer runs without replicas.
The provider can now expose the PostgreSQL superuser account, for operations that need full privileges on a Percona PostgreSQL cluster.
provider-milvus 0.2.0 adds backup and restore for Milvus via milvus-backup, plus scheduled backups with retention.
provider-milvus 0.2.0 scrapes Milvus metrics via a Prometheus PodMonitor, so clusters are observable out of the box.
everestctl upgrade now upgrades the everest-crds release instead of refusing it, and honors --helm.reuse-values and --skip-wizard as expected.
OpenEverest 1.17.0 adds support for the latest PostgreSQL operator (3.1.0) and Percona Server for MongoDB operator (1.23.1).
Single sign-on in OpenEverest now accepts opaque tokens from identity providers and reliably completes the login redirect.
Valkey instances can now require mutual TLS, so only clients presenting a trusted certificate can connect.
New Valkey instances can run Valkey 9.1, and existing 9.0 deployments move to the latest 9.0.6 patch release.
Semi-synchronous replication nodes in MariaDB clusters now start in read-only mode to protect replication consistency.
The Milvus provider adds Milvus 2.6.15 plus GPU-ready bundles, with per-component images so you can pin exactly the pieces you need.
Multi-replica MixCoord deployments now run in active-standby mode, keeping coordination available if a replica fails.
External S3 storage and etcd for Milvus deployments are now hardened for production use.
You can now deploy MariaDB 10.11 clusters through the OpenEverest MariaDB provider.
The first release of the Milvus provider brings vector database deployments to OpenEverest with auth, network exposure, and full component control.
A new plugin adds an Inspector tab to database pages with pod status, live logs and kubectl-describe-style details.
A new plugin runs pgbench against your PostgreSQL instances from the cluster page and shows the results, no kubectl needed.
The Valkey provider now applies engine-level scheduling policies to determine where pods are placed across the cluster.
Cassandra instances now honour node selectors, tolerations and affinity, and new clusters prefer separate nodes instead of requiring them.
Galera and replication clusters can now sit behind MariaDB MaxScale, which sends writes to the primary and spreads reads across the other nodes.
MariaDB instances now honour node selectors, tolerations, affinity and topology spread constraints, all editable from a new pod scheduling section in the UI.
Backup schedules support time-based retention, such as keeping 30 days of backups, with the PostgreSQL provider supporting it first.
OpenEverest instances now expose per-component pod status with PodsScheduled and PodsReady conditions, surfaced in the UI and everestctl.
The instance creation wizard and overview can now edit each component's affinity rules, and scheduling defaults are explicit - omitted fields use the provider default, empty values mean none.
Backup schedules can now keep backups for a period of time, not just a fixed number of copies.
The new @openeverest/plugin-theme package themes a plugin's own MUI bundle from the host's design tokens, including live dark mode.
Plugin compatibility manifests are now enforced - a plugin that fails compatibleHostVersions or the new compatibleUiContractVersions is skipped.
Backup and restore grants now follow the owning instance, reading instance credentials needs a new read-connection action, and previously unguarded restore endpoints are now protected.
Provider UI schemas can mark form sections as toggleable, hiding optional settings behind an Enable switch until they are needed.
The new BackupImport API and provider-runtime support let a provider create an instance directly from a backup in S3 storage.
TiDB instances can now back up on a cron schedule and prune old backups by count or by age.
Initial release brings SeaweedFS distributed object storage support to the OpenEverest platform.
Serve every model behind one HTTPS, OpenAI-compatible endpoint with a per-model API key and token quotas - your own private AI API on your own GPUs.
Restore PostgreSQL clusters to any specific point in time using backup archival data.
Schedule automated PostgreSQL backups via storage-level cron expressions.
Enable TimescaleDB OSS on PostgreSQL 18 clusters directly through the OpenEverest provider.
Deploy TiDB clusters faster with pre-configured resource size presets that eliminate manual CPU and memory tuning.
The TiDB provider now supports deploying TiFlash, enabling real-time analytical queries alongside transactional workloads.
You can now restore TiDB clusters from existing backups and create new instances from backup.
You can now trigger snapshot backups of TiDB clusters to S3-compatible storage on demand.
Expanded preset library reduces time needed to configure new MariaDB clusters.
Automatic data-plane updates keep MariaDB clusters current without manual intervention.
The TiDB provider now allows expanding Placement Driver and TiKV storage volumes, and rejects shrinking to prevent data loss.
The Percona PostgreSQL provider now lets you set Kubernetes CPU and memory resource requests through the OpenEverest UI.
The Percona PostgreSQL provider now exposes a custom configuration field, letting operators tune database settings directly from the OpenEverest UI or API.
The MongoDB Explorer plugin now surfaces cluster health at a glance and lets users expand truncated table cells inline without leaving the table view.
OpenEverest now supports TiDB, a distributed HTAP database, through a new provider built on TiDB Operator v2.
The KServe provider now supports distributed LLM inference with separate head and worker nodes for larger models that exceed single-node memory.
The KServe provider now supports automatic scaling of LLM inference workloads via Workload Variant Autoscaler, Horizontal Pod Autoscaler, and KEDA triggers.
The KServe provider's built-in model catalog has been refreshed with new presets, making it faster to deploy popular open-source LLMs.
The MongoDB Explorer plugin now automatically discovers and lists your OpenEverest-managed MongoDB instances directly in the sidebar.
The Percona Server for MongoDB provider now supports importing existing backups through an updated Backup CRD.
The Plugin Hub now automatically detects when installed plugins or providers are out of date and shows update alerts directly in the catalog.
OpenEverest now gates provider upgrades behind an explicit maintenance approval system, so disruptive changes like rolling restarts require operator consent.
OpenEverest now lets you create database instances in a single click by selecting from predefined InstancePresets in the UI.
OpenEverest now supports seeding an engine's bootstrap credentials from a user-provided Secret when creating an instance.
The Instance and BackupStorage APIs now accept PATCH requests, letting you update individual fields safely without read-modify-write races.
Instance scheduling in OpenEverest now supports tolerations, node selectors, topology spread constraints, and custom scheduler names alongside affinity rules.
Automatic secure password generation is now available when creating Valkey instances through OpenEverest.
The OpenEverest MariaDB provider now supports point-in-time recovery, letting you restore a database to any moment within the backup retention window.
The Altinity ClickHouse provider now exposes engine configuration parameters in the OpenEverest UI, letting you fine-tune settings at creation and edit time.
The Altinity ClickHouse provider now enables TLS encryption and sets a secure default user configuration out of the box.
The Altinity ClickHouse provider now exposes cluster topology and connection details through the OpenEverest API for easier integration and monitoring.
The Altinity ClickHouse provider now supports OpenEverest's built-in monitoring integration, automatically exposing metrics for collection.
The Altinity ClickHouse provider now runs in cluster-wide mode and reports accurate instance status, including health and readiness conditions.
No updates match the selected topic.
