Switchboard

Appearance

Store run artifacts ​

Let runs move large files — a 1 GiB attachment out of a run, a 300 MB video dropped on a thread into a run's workspace — through one private R2 bucket instead of the inline caps (5 MiB per image, 10 MiB per document). The bot signs URLs and records keys; the container and the bot's Worker move the bytes.

You need:

  • A Cloudflare account with R2 enabled, and CLOUDFLARE_API_TOKEN for it in your shell with Workers R2 Storage: Edit (the deploy commands already use this token; deploy creates the bucket with it).
  • A second, bucket-scoped R2 token for the bot: in the dashboard, R2 → Manage API tokens → Object Read & Write on the one bucket. Save its two halves as ARTIFACTS_R2_ACCESS_KEY_ID and ARTIFACTS_R2_SECRET_ACCESS_KEY where the profile's secretsSource reads them (~/.secrets/switchboard/<NAME> by default).
  • A bearer you mint yourself for the bot's copy route, saved as ARTIFACTS_COPY_TOKEN beside them (openssl rand -hex 32).
  • PUBLIC_BASE_URL set on the bot: the run page's file links are built from it.

Run the commands below from the operator directory with the released CLI version you operate.

Name the bucket in the deployment profile ​

json
"artifacts": { "bucket": "switchboard-artifacts" }

With this in deploy/profile.json, the bot Worker's rendered wrangler.jsonc binds the bucket, and the bot's npm run deploy creates it before the upload — idempotently, so an existing bucket reports "already exists".

Put the secrets, then turn the store on ​

bash
npx --yes @coreplane/switchboard@<version> deploy secrets bot --only ARTIFACTS_R2_ACCESS_KEY_ID,ARTIFACTS_R2_SECRET_ACCESS_KEY,ARTIFACTS_COPY_TOKEN

Secrets first, always: a configured store with a missing secret fails the bot's startup by name. Then add the section to the bot's config.yaml, push it and restart:

yaml
artifacts:
  r2:
    accountId: <your Cloudflare account id>
    bucket: switchboard-artifacts # the profile's name, exactly
  retentionDays: 30 # optional; the default
bash
npx --yes @coreplane/switchboard@<version> deploy config
npx --yes @coreplane/switchboard@<version> deploy restart

/healthz now carries "artifacts": { "bucket": "switchboard-artifacts" }. A bucket name that differs between the profile and the config refuses the bot at startup naming both.

Apply retention ​

bash
npx --yes @coreplane/switchboard@<version> artifacts lifecycle --dry-run
npx --yes @coreplane/switchboard@<version> artifacts lifecycle

The first prints the two rules — every object deleted retentionDays after it was written, an incomplete multipart upload aborted after one day — and touches nothing. The second applies them through Cloudflare's API with your token and reads them back; it refuses to claim success when the read-back differs. Run it again after changing retentionDays. The run page shows a file whose object has expired as expired after N days.

Check the bucket is private ​

bash
npx --yes @coreplane/switchboard@<version> artifacts check

… is private: the managed domain pub-….r2.dev is disabled and no custom domain is attached is the answer you want. Anything else names the setting to turn off in the dashboard (R2 → the bucket → Settings → Public access). The check reads the bucket's domain settings; nothing unsigned can reach an R2 object either way, which is why the bot does not probe at runtime.

Turn it off ​

Remove the artifacts: section, deploy config, deploy restart. Every path falls back to the inline caps; the bucket and its objects stay until the lifecycle rules expire them.

Next ​