# Kvick > Kvick is the company behind Spegel Enterprise Edition (SEE), the production version of the open source project Spegel. SEE is peer-to-peer container image and AI model distribution for Kubernetes. It cuts image pull time, model cold starts, and registry egress for model serving, ML model deployment and LLM inference clusters. Spegel is the open source stateless registry mirror for Kubernetes (https://github.com/spegel-org/spegel). SEE is its production version. Every node shares the image layers and model weights it already has with its peers, so new GPU nodes pull from inside the cluster instead of an external registry or Hugging Face. Main topics: AI model cold start, Kubernetes cold start, model distribution, thundering herd. Problems SEE solves: - AI model cold starts and Kubernetes cold starts: slow image pulls and model downloads that delay time to first token (TTFT) - Thundering herd pulls when hundreds of GPU nodes scale up at once - Docker Hub and registry rate limits, and registry egress cost - Moving Hugging Face models in-cluster as OCI artifacts - Air-gapped and regulated inference with SBOM and FIPS compliant builds - Multi-region AI model distribution - Multi-tenant clusters: Multi-Tenancy Enforcement sets fallback levels and enforces artifact access cluster wide, so only tenants that should have access can access their image, even on a shared node - Cross-zone traffic: Topology Aware Routing keeps artifact traffic inside a zone and only sends it elsewhere when necessary - Pulling the same artifact over and over: SEE fetches the artifacts you need before you need them, once and only once Trusted Integrations: Hugging Face, vLLM, kServe. The site also covers model serving with llm-d, NVIDIA Dynamo, and Ray. Editions: Development (evaluation, limited nodes) and Enterprise (unlimited nodes, Clairvoyant Caching, multi-region coordination, Multi-Tenancy Enforcement, Topology Aware Routing, SBOM and FIPS builds, SLA support). A self-hosted, air-gapped option is available. ## Product - [Spegel Enterprise Edition (SEE)](https://kvick.dev/spegel-enterprise-edition): Plans, capabilities, and sales enquiries. - [Pricing](https://kvick.dev/pricing): Development and Enterprise editions compared. - [Pull and distribute](https://kvick.dev/how-it-works/pull-and-distribute): How peer-to-peer layer distribution works. - [Clairvoyant Caching](https://kvick.dev/how-it-works/clairvoyant-caching): Predictive pre-caching of images and models before nodes need them. - [Pull cost calculator](https://kvick.dev/pull-cost-calculator): Estimate the cost of slow image and model pulls. ## Solutions - [Multi-Region AI Model Distribution](https://kvick.dev/use-cases/multi-region-ai-model-distribution): Serve model weights peer-locally across regions. - [Air-Gapped Inference](https://kvick.dev/use-cases/air-gapped-inference): Model serving with no external registry. - [Thundering Herd Elimination](https://kvick.dev/use-cases/thundering-herd-elimination): Scale GPU nodes without overwhelming the registry. - [SBOM and FIPS Compliant Builds](https://kvick.dev/use-cases/sbom-fips-compliant-builds): Hardened, auditable binaries for regulated environments. ## Blog - [Bring It All In: Keep Hugging Face Models Next To Your Workloads](https://kvick.dev/blog/keep-hugging-face-models-next-to-your-workloads): Cuddle, the Hugging Face integration. By Philip Laine; includes the demo video and GitHub repository. - [Model caching for faster inference](https://kvick.dev/blog/model-caching-faster-inference): Fixing LLM serving bottlenecks and inference cold starts. - [Introducing Clairvoyant Caching](https://kvick.dev/blog/introducing-clairvoyant-caching): Predictive caching for Kubernetes images. - [Tårta på Tårta: Too Much of a Good Thing](https://kvick.dev/blog/tarta-pa-tarta): Batu Uslu on registry mirrors, Spegel, and the founding of Kvick. ## Docs - [SEE for yourself](https://docs.kvick.dev/spegel-enterprise/getting-started/): Getting started with Spegel Enterprise Edition. - [Enterprise documentation](https://docs.kvick.dev/spegel-enterprise/overview/): Spegel Enterprise Edition overview. - [Hugging Face usage](https://docs.kvick.dev/spegel-enterprise/usage/huggingface/): Serve Hugging Face models through SEE. ## Optional - [Security](https://kvick.dev/security): Security practices. - [Phone a Friend](https://kvick.dev/request-demo): Tell us about your clusters and your models. - [Kvick on GitHub](https://github.com/kvick-org): Kvick repositories. - [Cuddle demo repository](https://github.com/kvick-org/cuddle-demo): Code accompanying the Cuddle article. - [Kvick on LinkedIn](https://www.linkedin.com/company/kvick-dev/): Company page.