Release history, stack versions, and upgrade guidance for NVIDIA Cloud Functions. Each stack keeps its own artifact version. The compatibility matrix records the stack releases that are qualified to run together.
NVCF 1.1.0
NVCF 1.1.0 selects Self-Managed 1.1.5, Compute Plane 1.1.0, and Observability 1.1.0. The artifact manifest records the selected artifacts. Download the product archive and checksums from the NVCF 1.1.0 release, then follow Consume an NVCF release.
The 1.1.0 docs edition is available in the version selector. Version 1.0.4 remains the default while the 1.0.1 to 1.1.0 upgrade procedure and its validation are completed. Publication does not establish a qualified upgrade path. Track completion in the upgrade follow-up. Reviewed documentation corrections can update this edition without changing the product tag or release assets. Making 1.1.0 the default requires a separate reviewed docs change after upgrade validation is complete.
NVCF 1.0.4
NVCF 1.0.4 selects Self-Managed 1.0.10, Compute Plane 1.0.5 and Observability 1.0.0. It combines security patches and runtime fixes across the three stacks. The selected stack versions and artifact manifest record the complete selection. Changes below are relative to docs edition 1.0.2.
- Updates selected service, worker and telemetry dependencies for security maintenance, including NICLLS 2.118.2, Samba 1.0.9 and both compute-plane collectors at 0.160.0-nv-0.2.9.
- Patches the Java service runtime images and updates OpenBao server and agent
to 2.6.3 with both old and target JWT plugins installed. Existing mounts retain
their selection; new mounts use target
v1.4.1. Existing-mount migration is a separate procedure. The migration image includes patched Kubernetes clients. - Corrects Samba share ownership setup so non-root Helm tasks can create their results directory. Samba retains its existing root runtime.
- Returns a prompt not-found response when an LLM function is sent to the unsupported classic invocation endpoint.
- Preserves existing readable secret-file ownership during non-root ESS refresh.
- Processes new NVCA requests after a restarted stream reuses a sequence number, and fixes NATS authentication reconnect and account selection.
- Replaces legacy task-worker references with published init and task images. Follow the task image prerequisites when upgrading a deployment that runs tasks.
- Includes ICMS reconnect and instance-event fixes and an opt-in legacy lookup backfill. Pylon allows queued LLM requests more time for first output.
- Adds upstream image override support for NATS, cert-manager and the Vault injector. NATS uses the upstream Synadia server by default.
Release 1.0.3
Release 1.0.3 is a docs-only patch. It keeps Self-Managed 1.0.1, Compute Plane 1.0.0, and Observability 1.0.0. No stack upgrade is required for this documentation update.
- Add the legacy Self-Managed 0.6.2 release entry and patch upgrade procedure.
- Update the staged 1.0.1 upgrade to start from 0.6.2 and state the rehearsal’s availability limits.
- Repair the KAI template download and Dynamo topology guide links.
Self-Managed 0.6.2 (legacy patch)
Self-Managed 0.6.2 is a Helmfile bundle patch for the legacy 0.6.x stack. It updates the OpenBao chart from 0.30.25 to 0.32.6 and refreshes the JWT plugin catalog before migrations. Runtime image versions remain at their 0.6.1 pins.
Follow the 0.6.1 to 0.6.2 upgrade notes before the staged migration to 1.0.1. See the stable release and its artifact inventory for the patch contents.
Release 1.0.1
Starting with NVCF 1.0.1, the three Helm stacks are versioned and released independently.
Release 1.0.1 documents this qualified stack combination:
These notes describe what changed since NVCF self-hosted 0.6.1, the last legacy release. In 0.6.1, all components shipped as a single bundle.
Release 0.6.0
NVCF Self-Hosted 0.6.0 is a stable release. It adds asynchronous task workloads, multi-cluster control-plane management, OpenAI-compatible LLM routing, single-mode Multi-Instance GPU (MIG) support, and control-plane observability. It also adds Early Access support for optional NVIDIA Dynamo and Grove operator integration.