​
Quickstart

This guide walks you through installing kvcdn and uploading your first KV cache artifact to KV Cache Store.

​
1. Install the CLI

kvcdn is an x86-64 Linux binary. ARM64 support is planned for a future release. 
curl -L https://kvcachestore.com/download/kvcdn -o kvcdn
chmod +x kvcdn
sudo mv kvcdn /usr/local/bin/
kvcdn --version
See Installation for build-from-source options.

​
2. Sign in to the portal

Open https://kvcachestore.com and sign in with the configured OIDC provider. The first time you sign in, KV Cache Store creates:
  • a customer record,
  • a default organization,
  • a default project,
  • and a storage namespace for your uploads.

​
3. Create an API key

In the dashboard, go to Settings > API Keys and create a key. Copy the secret — it is shown only once. Set it in the CLI: 
kvcdn api-key set kv_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
kvcdn api-key verify

​
4. Generate a KV cache artifact

kvcdn verify runs a model against a context, produces a reusable .kv artifact, and checks that loading the artifact gives the same result as a full prefill: 
kvcdn verify \
  --kv-path ./context.kv \
  --context-file ./context.txt \
  --question "Summarize the key claim."
The command prints the generated artifact path. The file contains the KV-cache tensors in safetensors format plus a JSON sidecar with model_name, dtype, and num_tokens.

​
5. Upload the artifact

kvcdn upload ./context.kv --name "Qwen system prompt" --visibility private
kvcdn upload reads the artifact metadata, asks the portal for a presigned upload URL, PUTs the file to object storage, and then confirms the upload.

​
6. View the artifact in the dashboard

Go to https://kvcachestore.com/app. The artifact appears with its model, dtype, token count, size, and visibility. You can toggle visibility between private and public from the artifact page.

​
7. Share or consume the artifact

For a public artifact, the dashboard shows a public URL: 
https://x.kvcdn.io/{org_slug}/{project_slug}/{artifact_id}
Anyone can fetch it anonymously: 
curl -L -o context.kv https://x.kvcdn.io/my-org-a18620ee/default/198dc22d-...
Then load the KV cache into your inference pipeline before generating tokens. The model skips re-prefilling the cached context, so you pay only for new tokens.

​
Next steps