# NVCF (NVIDIA Cloud Functions)

Self-managed GPU functions, streaming and batch workloads on Kubernetes

Source: https://wheredoirunthis.com/products/nvcf · JSON: https://wheredoirunthis.com/api/products/nvcf · Read on 2026-10-15

## At a glance

- Type: Serverless
- Kind: Open source tool
- Isolation: Container
- Free tier: Not documented
- Starting price: Not disclosed
- GPU: Yes
- Regions: Not disclosed
- Coverage: 14 of 61 assessed capabilities supported
- Agent readiness: read 9 of 10 signals, drive 9 of 10 signals

## Links

- Homepage: https://github.com/NVIDIA/nvcf
- Docs: https://docs.nvidia.com/nvcf/overview
- Repository: https://github.com/NVIDIA/nvcf

## Website design

Desktop public website · assessed 2026-10-15

Complete assessment.

The public entry point is GitHub's dense repository interface, while NVIDIA's docs use a white, three-column technical layout with restrained green branding, clear heading hierarchy and a prominent NVCF banner. Pricing is not applicable to this self-host-only OSS subject.

- Styles: Corporate
- Landing page: captured
- Pricing: not applicable
- Docs: captured

## Runtime

- **GPU & accelerators**: Yes — "NVCF GPU-backed functions and asynchronous tasks; multiple GPU types/cluster fleets." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) (confidence: high)
- **BYOC (Bring your own cloud)**: na — "Pure oss-tool: operator self-hosts infrastructure; vendor-managed hosted BYOC is not a runtime facet here." [source](https://docs.nvidia.com/nvcf/1.0.3/self-managed/installation-overview.md) (confidence: high)
- **Secrets management**: ESS tenant-scoped encrypted secrets plus ESS Agent workload injection; OpenBao wraps namespace keys with AES-256-GCM master keys. — "MEK stored in OpenBao wraps namespace encryption keys; ESS is primary consumer. ESS public README documents encrypted secret API and Agent source renders secret templates." [source](https://docs.nvidia.com/nvcf/1.0.3/self-managed/key-rotation.md) (confidence: high)
- **Storage options**: Platform state in Cassandra; OpenBao Raft persistence; shared read-only model cache through SMB/CSI/RWX; Helm functions may use PVCs; self-hosted task result upload is unsupported. — "Tasks must set resultHandlingStrategy NONE and preserve outputs themselves to mounted volume/external storage. These are not a managed object-store or per-function volume-service guarantee." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/task-creation.md) (confidence: medium)
- **Supported workloads**: gpu-inference, gpu-streaming, batch-gpu-jobs, fine-tuning, evaluation, data-preparation — "Functions are long-running invokable workloads; NVCT tasks run to completion. Container or coordinated Helm workloads; no blanket arbitrary-code sandbox claim." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) (confidence: high)
- **Persistence**: Operator-managed persistent infrastructure; Helm functions can request PVC-backed volumes. No universal per-function durable-disk lifecycle guarantee. — "Allowed objects include PersistentVolumeClaims and allowed volumes include persistentVolumeClaim. Internal Cassandra/OpenBao state persistence is separate from workload data." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/helm-functions.md) (confidence: medium)
- **Data jurisdiction**: na — "Self-managed placement is chosen by operator; no hosted jurisdiction guarantee." [source](https://docs.nvidia.com/nvcf/1.0.3/self-managed/installation-overview.md) (confidence: high)
- **Resource limits & quotas**: Per-function-version autoscaling with minInstances>=0 and maxInstances>0; default idle timeout 1800s (operator-configurable); API-key scopes and account quotas are administered locally. — "Instance counts/policy apply to function version, not independently per GPU spec. Quota runbook documents admin accounts update limits; task max runtime/queued duration and heartbeats are documented." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/configure-autoscaling.md) (confidence: high)
- **SDK languages**: python, go — "Official Python workload helpers and in-repo Go shared libraries. Not evidence of public multi-language management SDKs; do not count Rust Stargate routing libraries as client SDKs." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/container-functions.md) (confidence: medium)
- **Network & egress controls**: NVCA-managed customizable Kubernetes NetworkPolicy: default public-internet egress with common private ranges blocked; CNI enforcement required. — "Policies are supplied through nvca-namespace-networkpolicies ConfigMap and can be overridden. Unsupported CNI means no enforcement." [source](https://docs.nvidia.com/nvcf/1.0.3/compute-plane/cluster-configuration.md) (confidence: high)
- **Snapshots & forks**: Optional beta NvSnap GPU checkpoint/restore; single-GPU full process restore, any-GPU-count warm-cache restore — "NvSnap integration is off by default/enabled per cluster. Single-GPU full state includes CPU process state and GPU memory; multi-GPU process restore currently blocked upstream." [source](https://github.com/NVIDIA/nvcf/blob/main/src/compute-plane-services/nvsnap/README.md) (confidence: high)
- **Deployment methods**: container-image, cli, api — "Create container or Helm-based function/task through API or nvcf-cli. Helm describes workload packaging/operator installation; helm is not a current controlled deploy-interface token." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/function-creation.md) (confidence: high)
- **Isolation**: Container — "Default standard container runtime/runc on Kubernetes. Optional KataRuntimeIsolation forces cluster workloads into Kata lightweight VMs; not a universal VM guarantee." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/multi-tenancy.md) (confidence: high)

## Capabilities

### AI & agents

| Capability | Status | Evidence |
| --- | --- | --- |
| Agent skills | Supported | "Public installable user/developer skills cover self-managed installation, CLI lifecycle, cluster prerequisites, topology exploration and documentation style; external coding agents discover these skills." [source](https://github.com/NVIDIA/nvcf/blob/main/ai-tooling/README.md) |
| AI app builder | Not supported | "No prompt-to-app generation product advertised." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Autonomous QA agent | Not supported | "No autonomous first-party app QA agent advertised." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Built-in coding agent | Not supported | "No first-party coding agent advertised; agent skills assist external coding tools." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| BYOK for LLMs | Not supported | "No external LLM provider-key gateway product advertised; the core serves operator-deployed model workloads." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Credential-injection proxy | Not supported | "ESS renders secrets inside the workload; worker credential sidecar rotation is not on-the-wire secret substitution keeping real API keys outside customer code." [source](https://github.com/NVIDIA/nvcf/blob/main/src/compute-plane-services/ess-agent/README.md) |
| LLM gateway | Partially supported | "LLM API Gateway routes OpenAI-compatible chat/completions, Responses and embeddings by function/model, applies token limits and load-balances NVCF model backends. No documented general external multi-provider/BYOK routing; narrower than the canonical multi-provider gateway." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/llm-gateway.md) |
| MCP client | Not supported | "No first-party MCP client advertised. The docs MCP server is read-only documentation search." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| MCP-server hosting | Not supported | "No documented MCP-server workload hosting recipe or first-class hosting surface. Serving a documentation-search MCP endpoint is not MCP hosting." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| MCP server | Supported | "Discoverable streamable-HTTP fern-docs-mcp-server exposes read-only searchDocs over NVCF documentation. Documentation search only, not deployment/control-plane tools." [source](https://docs.nvidia.com/nvcf/_mcp/server) |
| Model inference API (first-party) | Not supported | "Self-managed NVCF exposes an OpenAI-compatible gateway for the operator's own models; no first-party hosted model catalogue/pay-per-token service in this SKU. build.nvidia.com remains a separate sibling." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/llm-gateway.md) |
| Preinstalled coding agent | Not supported | "No provision-and-launch third-party coding harness inside an NVCF workspace advertised." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |

### Networking & delivery

| Capability | Status | Evidence |
| --- | --- | --- |
| Automatic HTTPS | Partially supported | "Production HTTPS requires operator DNS and TLS configuration. cert-manager can automatically provision certificates after setup; not turnkey automatic per-workload HTTPS." [source](https://docs.nvidia.com/nvcf/1.0.3/self-managed/gateway-routing.md) |
| Custom domains | Partially supported | "Production deployments use operator-controlled domains with A/CNAME records and configured route hostnames. Optional Vanity Gateway custom host/path mapping is available only in stack packages that include that addon." [source](https://docs.nvidia.com/nvcf/1.0.3/self-managed/gateway-routing.md) |
| Direct public IPv6 | Not supported | "No stable public per-instance IPv6/raw socket addressing advertised." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Edge KV store | Not supported | "No first-party colocated edge key-value store advertised." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Egress controls | Supported | "NVCA preconfigures NetworkPolicy including allow-egress-internet-no-internal-no-api; operators customize via nvca-namespace-networkpolicies ConfigMap. Enforced only with a NetworkPolicy-capable CNI." [source](https://docs.nvidia.com/nvcf/1.0.3/compute-plane/cluster-configuration.md) |
| Email send / receive | Not supported | "No email send/receive service advertised." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Public preview URLs | Not supported | "No per-branch/deploy preview URL feature is advertised; invocation hostnames are workload addressing, not preview environments." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Private networking | Partially supported | "Per-instance namespaces allow pod-to-pod communication within the same namespace and Helm workloads support ClusterIP Services; not an advertised cross-function/cross-region private service network." [source](https://docs.nvidia.com/nvcf/1.0.3/compute-plane/cluster-configuration.md) |
| Built-in reverse proxy | Supported | "nvcf-gateway-routes creates HTTPRoute/GRPCRoute/TCPRoute/UDPRoute resources to route external traffic to NVCF and workload invocation services." [source](https://docs.nvidia.com/nvcf/1.0.3/self-managed/gateway-routing.md) |

### Organisation & security

| Capability | Status | Evidence |
| --- | --- | --- |
| Built-in auth | Supported | "Self-hosted NVCF uses scoped bearer credentials NVCF_TOKEN, NVCF_API_KEY and NVCF_NVCT_API_KEY. API Keys service issues/authenticates/introspects/rotates/revokes keys; per-endpoint scope table is documented." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/api.md) |
| BYOC | Supported | "Self-managed NVCF runs on operator-owned Kubernetes clusters (cloud-provider agnostic or on-prem); control-plane and GPU targets may be one cluster or separate clusters." [source](https://docs.nvidia.com/nvcf/1.0.3/self-managed/installation-overview.md) |
| Deploy into customers' VPCs | Not supported | "Self-hosting spans the operator's clusters; no vendor-managed deployment of a SaaS into the vendor's customers' accounts advertised." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Secrets management | Supported | "ESS provides tenant-scoped encrypted secret storage and authorization using a Vault KV-v2-like API. OpenBao holds AES-256-GCM master wrapping keys; ESS Agent renders/injects workload secrets." [source](https://github.com/NVIDIA/nvcf/blob/main/src/control-plane-services/encrypted-secret-store/README.md) |
| Self-host / OSS core | Supported | "Apache-2.0 self-managed platform with public service source, CLI, Helm deployment stacks and build instructions. Issue #12 remains open but its missing-source list is no longer current evidence that APIs are absent." [source](https://github.com/NVIDIA/nvcf/blob/main/README.md) |
| Sharing | Not supported | "No share-by-link running-resource primitive advertised; API client authorization is an access-control surface rather than collaborative sharing." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/service-keys.md) |
| SSO / SAML | Not supported | "No enterprise SSO/SAML login flow advertised for NVCF. Scoped bearer keys, service OAuth/JWT utilities, and issuer overrides alone do not establish workforce SSO." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/api.md) |
| Teams | Partially supported | "Multi-tenant NCA account scoping and account isolation exist; no full team/member invitation/role management workflow established on the self-managed public surface." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/multi-tenancy.md) |
| Workload identity | Not supported | "Infrastructure workers rotate router/ESS credentials, but customer containers do not mount those credentials; no user-facing OIDC/SPIFFE/cloud-IAM workload federation feature established." [source](https://github.com/NVIDIA/nvcf/blob/main/src/compute-plane-services/worker-llm-credentials/README.md) |

### Developer access

| Capability | Status | Evidence |
| --- | --- | --- |
| CLI | Supported | "Official nvcf-cli supports self-hosted installation, cluster registration, token/key management, function/task lifecycle, invocation, registry credentials and account administration; source build or packaged NGC download." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/cli.md) |
| E2B-compatible API | Not supported | "NVCF APIs are its own compute/function APIs, not drop-in E2B sandbox SDK compatibility." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Filesystem API | Not supported | "No remote workload filesystem list/read/write/watch API advertised; invocation assets and model caches are not a general filesystem API." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Git deploy | Not supported | "Function creation accepts a prebuilt container or Helm chart, not git push-to-deploy." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/function-creation.md) |
| HTTP data API | Not supported | "The REST API manages/invokes compute, not a stored-data query/read/write service." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/api.md) |
| OpenAPI spec | Supported | "Fetchable OpenAPI 3.1.0 Cloud Functions spec, info.version 2.261.2. Self-hosted callers replace the hosted domain with their own gateway." [source](https://api.nvcf.nvidia.com/v3/openapi) |
| Private container registry | Not supported | "Integrates with private registries and manages pull credentials; it does not provide a hosted private image registry." [source](https://docs.nvidia.com/nvcf/1.0.3/self-managed/third-party-registries.md) |
| Public API | Supported | "Documented function management, deployment, invocation, task management and cluster APIs, served on the operator's gateway." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/api.md) |
| SDK | Partially supported | "Official Python workload helpers and in-repository Go shared libraries exist; no documented multi-language public management SDK release/package surface established. Rust Stargate is runtime routing infrastructure, not a client SDK." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/container-functions.md) |
| SSH access | Not supported | "No direct workload SSH feature advertised; workloads are Kubernetes functions and tasks." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |

### Compute & storage

| Capability | Status | Evidence |
| --- | --- | --- |
| Cloud-init / ignition config | Not supported | "No VM cloud-init/ignition user-data interface advertised; this is Kubernetes workload deployment." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Desktop / computer-use sandbox | Not supported | "No virtual-desktop/computer-use sandbox advertised; streaming GPU application workloads alone do not establish desktop sandboxes." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| GPU compute | Supported | "GPU-accelerated functions and asynchronous tasks run across GPU clusters; deployment specifications select GPU and instance types." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Live-state (memory) snapshots | Partially supported | "Full process checkpoint/restore uses CRIU plus NVIDIA cuda-checkpoint for CPU state, open FDs, io_uring/epoll and GPU memory, resuming a live engine. Currently single-GPU only; multi-GPU process restore is blocked upstream. Beta, per-cluster opt-in." [source](https://github.com/NVIDIA/nvcf/blob/main/src/compute-plane-services/nvsnap/README.md) |
| Multi-region | Supported | "The unified control plane manages and routes requests across multi-region GPU clusters; placement is operator-provided, not an NVIDIA-hosted region guarantee." [source](https://github.com/NVIDIA/nvcf/blob/main/README.md) |
| Object storage | Not supported | "No managed object-store offering. Task outputs must go to operator-provided mounted volumes or external storage; internal NvSnap blob storage is not a user bucket service." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/task-creation.md) |
| Persistent disk | Partially supported | "Helm functions allow PersistentVolumeClaims and persistentVolumeClaim volumes; operator-provisioned storage, not a general persistent disk API. Task default validation permits only configMap/secret/projected/emptyDir volumes, and retention across instance deletion is not promised." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/helm-functions.md) |
| Sandbox templates / custom images | Not supported | "Container/Helm function packaging is not advertised as reusable sandbox/VM base images from which sandbox instances are forked." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/function-creation.md) |
| Scale to zero | Supported | "minInstances: 0 allows a function to scale to zero when idle. Platform default idle timeout is 1,800 seconds; self-managed operators can override the platform setting." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/configure-autoscaling.md) |
| Selectable isolation runtime | Partially supported | "Default is runc containers; operator can enable KataRuntimeIsolation for a cluster, forcing all function/task pods there into Kata lightweight VMs. Cluster-level environment choice, not a per-workload runtime switch." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/multi-tenancy.md) |
| Snapshots / forking | Partially supported | "Optional beta NvSnap integration checkpoints/restores GPU workloads; warm-cache capture works at any GPU count, full process checkpoint/restore is currently single-GPU. Disabled by default and enabled per cluster." [source](https://github.com/NVIDIA/nvcf/blob/main/src/compute-plane-services/nvsnap/README.md) |

### Data

| Capability | Status | Evidence |
| --- | --- | --- |
| Database branching | Not supported | "No user database branch primitive advertised." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Ephemeral instant DB (no-signup) | Not supported | "No no-signup temporary database provisioning advertised." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| First-party database | Not supported | "Cassandra stores platform state; no first-party managed database for user workloads advertised." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Vector search | Not supported | "No native vector database/index/search feature advertised; deploying an embedding model is not a vector store." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/llm-gateway.md) |

### Platform & operations

| Capability | Status | Evidence |
| --- | --- | --- |
| Managed CI/CD runners | Not supported | "No platform-hosted drop-in CI runner product advertised; repository CI workflows are not a user runtime feature." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Built-in observability | Partially supported | "Official separately versioned observability stack bundles NVCF collectors/monitors and default VictoriaMetrics, with example dashboards and metrics docs. Early Access; operators still own long-term storage, log collection, dashboards and alerting." [source](https://docs.nvidia.com/nvcf/1.0.3/observability/observability.md) |
| Preview environments | Not supported | "No full per-PR/branch ephemeral environment lifecycle advertised." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Scheduled tasks | Partially supported | "Helm function charts explicitly allow Kubernetes CronJobs, but no dedicated NVCF cron schedule API is documented. NVCT by itself is run-to-completion, not scheduling." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/helm-functions.md) |
| Project template marketplace | Not supported | "Function examples and Helm samples are not an advertised starter-template marketplace/gallery." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/overview.md) |
| Durable workflows | Not supported | "NATS invocation buffering and one-shot NVCT task lifecycle are not a general durable workflow execution/replay engine." [source](https://docs.nvidia.com/nvcf/1.0.3/overview/task-creation.md) |

## Agent readiness

### Read: 9 of 10 signals present

9/10 signals (5 full, 4 partial)

- Llms txt: pass — "docs-1.0.3-llms.txt is a substantive complete current page index, discovered via docs-llms.txt version selector."
- Llms full txt: fail — "Both root and version /llms-full.txt return HTTP 200 after redirect to the corresponding llms.txt index, not aggregated full content. Retained docs-llms-full.txt and docs-1.0.3-llms-full.txt; do not count 200 as full corpus."
- Robots AI policy: pass — "docs-robots.txt explicitly allows major AI bots and clean .md/llms paths; no blanket Disallow /."
- Sitemap: pass — "docs-sitemap.xml returns substantive NVCF URL set (28713 bytes)."
- Markdown negotiation: pass — "Accept: text/markdown on /overview/overview redirects to /overview/overview.md; Content-Type text/markdown; full page 2172 bytes retained docs-markdown-negotiation.txt."
- Docs discoverable: pass — "github-readme.md has top-level Docs link, installation/API links and repository documentation map."
- Pricing machine readable: na — "Pure oss-tool with no hosted commercial price tiers; Apache-2.0 + operator infrastructure cost, not missing hosted pricing."
- JSON LD: partial — "Recovered rendered HTML contains BreadcrumbList and TechArticle JSON-LD, but no Product/SoftwareApplication schema."
- Meta quality: partial — "Rendered docs HTML has description and OpenGraph/Twitter image/title metadata; sampled description is 12 whitespace-delimited words, below rubric 20-word threshold."
- Content efficiency: partial — "Recovered rendered HTML text/HTML ratio measured approximately 1.45%; large Fern scripts/markup reduce content efficiency."
- Anti bot: partial — "Plain urllib HTML returned HTTP 403; foxcrawl HTML/browser and native Markdown succeed. Intermittent access friction, not an unresolved content gap."

### Drive: 9 of 10 signals present

9/10 signals (6 full, 3 partial)

- Public API: pass — "docs-overview-api.md documents API groups and concrete endpoints/scopes on operator gateway."
- Openapi spec: pass — "openapi.json is valid JSON with openapi=3.1.0 and info.version=2.261.2, 116822 bytes."
- MCP server: pass — "docs-mcp-response.txt and docs-llms.txt expose streamable-HTTP read-only documentation search endpoint https://docs.nvidia.com/nvcf/_mcp/server. Not deployment tools."
- SDK packages: partial — "Python workload helpers are official; Go libraries are shared infrastructure; no public multi-language management SDK packages established. docs-overview-container-functions.md and github library inventories."
- Auth docs: pass — "docs-overview-api.md and CLI describe three bearer types, programmatic generation, scopes per endpoint and expirations."
- Agent self serve creds: na — "No hosted NVCF signup/API account; operator installs service. Local nvcf-cli init/api-key generate is programmatic and documented, but should not be scored as hosted self-registration."
- Webhooks: fail — "No documented signed user-facing event webhook delivery in retained current docs index/API. Internal auth webhooks and GitHub CI are not such a feature."
- CLI tool: pass — "docs-overview-cli.md describes official install, lifecycle, keys, account/registry administration."
- Error model: partial — "OpenAPI contains documented 403/429 responses and quota runbook shows JSON title/detail-style failure, but no unified public JSON error schema established. Do not preserve old unsupported full score."
- Rate limit docs: partial — "docs-overview-llm-gateway.md documents tokenRateLimit syntax, units and request behavior; spec includes 429. No full Retry-After/rate-limit-header contract established."
- Agent config: pass — "github-AGENTS.md plus public user/developer skills in github-ai-tooling-README.md and CLI SKILL. Multiple surfaces, not old one-surface partial."

## Open source

- Repository: https://github.com/NVIDIA/nvcf
- License: Apache-2.0
- Stars: 237
- Forks: 93
- Self-host methods: helm, kubernetes, source-build
- Self-host dependencies: kubernetes, nats, cassandra, openbao, envoy-gateway, cert-manager

## Company & presence

- Founded: NVIDIA founded 1993; NVCF public repository created 2026-04-01
- GitHub: https://github.com/NVIDIA/nvcf
- Docs: https://docs.nvidia.com/nvcf/overview/overview
- OpenAPI: https://api.nvcf.nvidia.com/v3/openapi
- Documentation MCP: https://docs.nvidia.com/nvcf/_mcp/server

## Nearby in the atlas

- [Beam](/products/beam.md)
- [Chutes](/products/chutes.md)
- [Deno Deploy](/products/deno.md)
- [Koyeb](/products/koyeb.md)
- [Kyushu](/products/kyushu.md)
- [Modal](/products/modal.md)

## Method

Facts come from the product's own public pages. Each claim carries its evidence: a quote and the page it came from. We count capabilities rather than grade products, and "Not assessed" means we have not checked, never that the product lacks it.