Cloudflare KV

Store artifacts in a Workers KV namespace, within KV's 25 MiB limit per artifact.

The Worker stores each artifact as one entry in the namespace bound as KV_STORE. The artifact is the value, and its size, upload time, and headers are the entry’s metadata.

Check the limits first

An artifact can be at most 25 MiB. The server rejects a larger upload with a 400 error, so that task never reaches the remote cache.
KV is eventually consistent. A new artifact is readable at once in the region that wrote it, but other regions can take 60 seconds or more to see it.

Bind a namespace

In apps/api/wrangler.jsonc, add a KV_STORE binding:
```
{
    "kv_namespaces": [{ "binding": "KV_STORE" }],
}
```
With no id, wrangler deploy creates the namespace and links it to the Worker. To use a namespace you already have, add its id.
Remove the r2_buckets entry, unless you want to keep the binding. When both are bound, the server uses KV.
Build and deploy the Worker:
```
pnpm build
pnpm run deploy
```

How entries expire

The server writes each entry with a TTL of BUCKET_OBJECT_EXPIRATION_HOURS, so KV deletes artifacts on its own. The daily expiry task also runs, and deletes any entry that has outlived that age.

KV fixes the TTL when the server writes an entry. After you change BUCKET_OBJECT_EXPIRATION_HOURS, entries written earlier keep their old TTL.

If you used KV with version 4, read Upgrade from v4 for what happens to its entries.