Buckets commands

ipx bucket manages private and public object storage using your existing account and -O/--org or --personal context. It requires a server with Buckets enabled and the appropriate access. API keys have a fixed workspace; use an organization key or an interactive session for organization actions.

ipx bucket status
ipx bucket pricing
ipx bucket connection
ipx --org acme bucket connection
ipx bucket create models --quota-bytes 10737418240
ipx bucket list
ipx bucket info models
ipx bucket update models --quota-bytes 21474836480
ipx bucket update models --name weights
ipx bucket update weights --status read_only
ipx bucket upload models ./weights.bin --key checkpoints/weights.bin
ipx bucket ls models --prefix checkpoints/
ipx bucket download models checkpoints/weights.bin ./download.bin
ipx bucket usage models
ipx bucket rm models checkpoints/weights.bin
ipx bucket delete models

connection shows the authenticated workspace ID and its S3 endpoint. The personal endpoint contains your user UUID; an organization endpoint contains the organization UUID. Copy that complete endpoint into S3 clients. Credentials must belong to the same workspace. Different workspaces can use the same bucket name. Uploads and downloads resolve this workspace before transferring files.

update changes a bucket’s name, maximum logical size, visibility or upload status. The maximum must cover stored bytes and active upload reservations. Increasing it does not charge for unused space. Renaming preserves objects and bucket-scoped credentials; update client configurations to use the new name. read_only pauses uploads, including completion of pending uploads; set active to resume.

Buckets default to private. To publish known file URLs:

ipx bucket create shared --visibility public
ipx bucket update models --visibility public
ipx bucket public-url models checkpoints/weights.bin
ipx bucket update models --visibility private

public-url retrieves the server-issued URL and refuses private buckets. A URL is not proof of permission: a public bucket’s explicit S3 policy may still deny a particular anonymous download. Listing and mutations remain protected unless a public bucket policy explicitly grants anonymous listing; reads and transfer are billed to the workspace. With server #183 deployed, switching to private blocks new anonymous GET, HEAD and policy-granted listing even when a policy still allows them. Private does not recall files already downloaded or stop transfers already accepted.

Large transfers use bounded multipart parts; failures abort unfinished uploads. Downloads stream into a new file and do not overwrite existing paths. ls returns one page; pass its opaque token with --continuation-token to continue. --json produces machine-readable results. Empty buckets with active uploads cannot be deleted. Retained versions can remain billable after deletion.

S3 credentials

Standard S3 clients need separate signing credentials. Secrets are revealed once and saved in a new mode-0600 file; they are not printed by these commands:

ipx bucket credentials create --read-only --output ./s3-read.json
ipx bucket credentials create --bucket-id BUCKET_UUID --prefix checkpoints/ \
  --expires-in 86400 --output ./s3-checkpoints.json
ipx bucket credentials list
ipx bucket credentials rotate CREDENTIAL_UUID --output ./replacement.json
ipx bucket credentials revoke CREDENTIAL_UUID

Credential requests use a retry UUID and automatically retry one ambiguous network/server failure. If recovery still fails, the error supplies a request UUID; repeat the same command and options with --request-id UUID within ten minutes. Reusing that UUID with different options fails. Review and revoke orphaned credentials before starting a new intent after the retry window.

Rotation preserves scopes and expiration, leaving the old credential active until revoked. Use --revoke-previous for immediate replacement after checking client impact. Never commit credential files or place secrets in logs, telemetry or shell history.