crab configure

Set up a new Crab repository in one guided flow.

crab configure [REMOTE] [OPTIONS]

crab configure combines the normal crab init and crab setup steps. It:

1. Selects Amazon S3, Google Cloud Storage, or Azure Blob Storage
2. Connects the current directory to a remote bucket/container
3. Detects credentials from environment variables and cloud SDK profiles
4. Installs the Git filter and diff drivers
5. Scans for large files and writes tracking rules
6. Prints the next commands to verify and ship the repository

Crab never copies long-lived cloud secrets into crab.toml. When credentials are missing, the setup output gives the provider-specific login command; run crab doctor afterward to verify access.

Interactive setup

Run with no arguments in a terminal:

mkdir models && cd models
crab configure

Crab asks for the cloud provider and a bucket/container plus repository path. Amazon S3 is the default choice.

Explicit setup

Use flags in scripts or when you already know the remote:

# Amazon S3
crab configure s3://team-data/models

# S3-compatible storage (after exporting credentials and AWS_ENDPOINT_URL)
crab configure crab://team-data/models --provider s3

# Google Cloud Storage
crab configure gs://team-data/models --provider gcs

# Azure Blob Storage
crab configure azure://team-data/models --provider azure

Provider-prefixed URLs let Crab infer the provider, so --provider is optional in those examples. Use it with a canonical crab:// URL when the provider is not S3.

Select a named AWS profile without committing the profile name:

crab configure s3://team-data/models --aws-profile ml-team

Crab stores the selector in .crab/local.toml and uses the AWS SDK default credential chain. Without the flag it honors AWS_PROFILE, then shared AWS configuration, web identity, ECS credentials, and EC2 instance credentials.

S3-compatible services use the s3 provider. Export the service's access key, secret key, region, and AWS_ENDPOINT_URL before configuration; do not put credentials in the remote URL. See Static Credentials for Cloudflare R2 and local MinIO/RustFS examples.

Choose the local bucket-GC listing policy during setup when desired:

crab configure s3://team-data/models --gc-list-profile adaptive

The default adaptive profile minimizes LIST streams for smaller namespaces and switches larger namespaces to hash-partition parallelism. cost always prefers fewer LIST streams; latency immediately prefers partition scans.

Tracking options

Pass explicit patterns instead of relying only on the repository scan:

crab configure s3://team-data/models \
  --track '*.safetensors' \
  --track 'datasets/**'

Install the integration without scanning:

crab configure s3://team-data/models --no-auto-track

Preview the plan without creating .git, .crab/, crab.toml, or tracking rules:

crab configure s3://team-data/models --provider s3 --dry-run

Existing project configuration

With no REMOTE, crab configure reuses a discovered crab.toml. This is a convenient way to repair local Git integration and refresh tracking after a regular Git clone.

Related commands

• crab init — configure only the repository connection
• crab setup — rescan or change large-file tracking
• crab doctor — verify credentials and repository health
• Static credentials — provider credential discovery