---
url: /docs/cli/cli-reference.md
description: >-
  Explore the Kit CLI command reference. Get detailed information on commands
  available for building, versioning, pushing, pulling, and running ModelKits
  within your AI/ML projects.
---

Manage temporary files cached by Kit

### Synopsis

Manage files stored in the temporary KitOps cache dir ($KITOPS_HOME/cache)

Normally, this directory is empty, but may contain leftover files from resumable
downloads or files that were not cleaned up due to the command being cancelled.

The $KITOPS_HOME location is system dependent:
\- Linux: $XDG_DATA_HOME/kitops with a fall back to $HOME/.local/share/kitops
\- MacOS: ~/Library/Caches/kitops
\- Windows: %LOCALAPPDATA%\kitops

### Examples

```
# Get information about size of cached files
kit cache info

# Clear files in cache
kit cache clear
```

### Options

```
  -h, --help   help for cache
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit cache clear

Clear temporary cache storage

### Synopsis

Clear temporary files from cache storage.

```
kit cache clear [flags]
```

### Options

```
  -h, --help   help for clear
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit cache info

Get information about cache disk usage

### Synopsis

Print the total size of temporary files in the cache directory.

```
kit cache info [flags]
```

### Options

```
  -h, --help   help for info
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit diff

Compare two ModelKits

### Synopsis

Compare two ModelKits to see the differences in their layers.

ModelKits can be specified from either a local or from a remote registry.
To specify a local ModelKit, prefix the reference with 'local://', e.g. 'local://jozu.ml/foo/bar'.
To specify a remote ModelKit, prefix the reference with 'remote://', e.g. 'remote://jozu.ml/foo/bar'.
If no prefix is specified, the local registry will be checked first.

```
kit diff <ModelKit1> <ModelKit2> [flags]
```

### Examples

```
# Compare two ModelKits
kit diff jozu.ml/foo:latest jozu.ml/bar:latest

# Compare two ModelKits from a remote registry
kit diff remote://jozu.ml/foo:champion remote://jozu.ml/bar:latest

# Compare local ModelKit with a remote ModelKit
kit diff local://jozu.ml/foo:latest remote://jozu.ml/foo:latest

```

### Options

```
      --plain-http         Use plain HTTP when connecting to remote registries
      --tls-verify         Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings   Path to TLS cert to add to trust store (flag can be repeated)
      --cert string        Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string         Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int    Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string       Proxy to use for connections (overrides proxy set by environment)
  -h, --help               help for diff
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit import

Import a model from HuggingFace

### Synopsis

Download a repository from HuggingFace and package it as a ModelKit.

The repository can be specified either via a repository (e.g. myorg/myrepo) or
with a full URL (https://huggingface.co/myorg/myrepo). The repository will be
downloaded to a temporary directory and be packaged using a generated Kitfile.

In interactive settings, this command will read the EDITOR environment variable
to determine which editor should be used for editing the Kitfile.

This command supports multiple ways of downloading files from the remote
repository. The tool used can be specified using the --tool flag with one of the
options below:

\--tool=hf  : Download files using the Huggingface API. Requires REPOSITORY to
be a Huggingface repository. This is the default for Huggingface
repositories
\--tool=git : Download files using Git and Git LFS. Works for any Git
repository but requires that Git and Git LFS are installed.

By default, Kit will automatically select the tool based on the provided
REPOSITORY.

```
kit import [flags] REPOSITORY
```

### Examples

```
# Download repository myorg/myrepo and package it, using the default tag (myorg/myrepo:latest)
kit import myorg/myrepo

# Download repository using a different version (tag) than the default and package it
kit import myorg/myrepo --ref v1.0.0

# Download repository and tag it 'myrepository:mytag'
kit import myorg/myrepo --tag myrepository:mytag

# Download repository and pack it using an existing Kitfile
kit import myorg/myrepo --file ./path/to/Kitfile
```

### Options

```
      --ref string                  Version (tag) of repository to import (default "main")
      --token string                Token to use for authenticating with repository
  -t, --tag string                  Tag for the ModelKit (default is '[repository]:latest')
  -f, --file string                 Path to Kitfile to use for packing (use '-' to read from standard input)
      --tool string                 Tool to use for downloading files: options are 'git' and 'hf' (default: detect based on repository)
      --concurrency int             Maximum number of simultaneous downloads (for huggingface) (default 5)
      --depth int                   Maximum directory depth to process when generating a Kitfile. Setting to -1 processes all files individually
      --attestation-output string   Write SLSA Provenance v1 predicate to <path> ('-' for stdout)
  -h, --help                        help for import
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit index

Manage ModelKit indexes

### Synopsis

Create and manage ModelKit indexes.

A ModelKit index groups variants of a model under a single reference. Use
'kit pull -l' or 'kit unpack -l' to select a ModelKit from an index by label.

### Examples

```
# Create an index in local storage
kit index create my-org/my-model:all

# Add ModelKits to the index
kit index add my-org/my-model:all my-org/my-model:q4_0
kit index add my-org/my-model:all my-org/my-model:q8_0

# Remove a ModelKit from the index
kit index remove my-org/my-model:all my-org/my-model:q4_0

# List indexes in local storage
kit index list

# List the ModelKits in the index
kit index info my-org/my-model:all

# Print the index manifest
kit index inspect my-org/my-model:all

# Push the index to a remote registry
kit index push my-org/my-model:all registry.example.com/my-org/my-model:all

# Pull an index from a remote registry
kit index pull registry.example.com/my-org/my-model:all

# Delete the index
kit index delete my-org/my-model:all
```

### Options

```
  -h, --help   help for index
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit index add

Add a ModelKit to a ModelKit index

### Synopsis

Add a ModelKit to a ModelKit index, or update the labels and annotations of
a ModelKit already in it. Labels and annotations are merged into what the entry
already has.

```
kit index add [flags] INDEX MODELKIT
```

### Examples

```
# Add a ModelKit to an index
kit index add my-org/my-model:all my-org/my-model:q4_0

# Add a ModelKit to an index by digest
kit index add my-org/my-model:all my-org/my-model@sha256:44136fa355b3678a1146ad16f7e8649e94fb4fc21fe77e8310c060f61caaff8a

# Add a ModelKit, labelling the entry
kit index add my-org/my-model:all my-org/my-model:q4_0 -l quantization=q4_0 -l vram=6GB

# Add a ModelKit with a JSON label value
kit index add my-org/my-model:all my-org/my-model:q4_0 -l 'gpuArchs:=["sm_80","sm_90"]'

# Annotate an entry already in the index, and drop one of its labels
kit index add my-org/my-model:all my-org/my-model:q4_0 --annotate org.example.tested=true -l vram-
```

### Options

```
  -r, --remote                 Resolve the ModelKit in its remote registry instead of local storage
  -l, --label stringArray      Set a label on the entry as key=value, or as key:=<json> for a JSON value, or remove one as key-. Can be specified multiple times
      --annotate stringArray   Set an annotation on the entry as key=value, or remove one as key-. Can be specified multiple times
      --plain-http             Use plain HTTP when connecting to remote registries
      --tls-verify             Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings       Path to TLS cert to add to trust store (flag can be repeated)
      --cert string            Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string             Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int        Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string           Proxy to use for connections (overrides proxy set by environment)
  -h, --help                   help for add
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit index create

Create a ModelKit index

### Synopsis

Create an empty ModelKit index in local storage.

```
kit index create [flags] INDEX
```

### Examples

```
# Create an index in local storage
kit index create my-org/my-model:all

# Create an index using a full reference
kit index create registry.example.com/my-org/my-model:all
```

### Options

```
  -h, --help   help for create
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit index delete

Delete a ModelKit index

### Synopsis

Delete a ModelKit index from local storage. The ModelKits it references are
not removed. To delete an index from a remote registry, use
'kit remove --remote'.

If specified by tag, the index is only untagged while other tags refer to it,
unless --force is used.

```
kit index delete [flags] INDEX
```

### Examples

```
# Delete an index from local storage
kit index delete my-org/my-model:all

# Delete an index and every other tag that refers to it
kit index delete --force my-org/my-model:all

# Delete an index by digest
kit index delete my-org/my-model@sha256:44136fa355b3678a1146ad16f7e8649e94fb4fc21fe77e8310c060f61caaff8a

# Delete all untagged indexes
kit index delete --all

# Delete all locally stored indexes
kit index delete --all --force
```

### Options

```
  -f, --force   Delete the index and all other tags that refer to it
  -a, --all     Delete all untagged indexes
  -h, --help    help for delete
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit index info

Show the ModelKits contained in an index

### Synopsis

List the ModelKits in a ModelKit index, with their labels and annotations.

```
kit index info [flags] INDEX
```

### Examples

```
# See the contents of a local index:
kit index info my-org/my-model:all

# See the contents of a local index by digest:
kit index info my-org/my-model@sha256:44136fa355b3678a1146ad16f7e8649e94fb4fc21fe77e8310c060f61caaff8a

# See the contents of a remote index:
kit index info --remote registry.example.com/my-org/my-model:all

# Print the contents of an index as JSON:
kit index info --format json my-org/my-model:all
```

### Options

```
  -r, --remote             Check remote registry instead of local storage
      --format string      Output format: table or json (default "table")
      --plain-http         Use plain HTTP when connecting to remote registries
      --tls-verify         Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings   Path to TLS cert to add to trust store (flag can be repeated)
      --cert string        Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string         Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int    Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string       Proxy to use for connections (overrides proxy set by environment)
  -h, --help               help for info
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit index inspect

Inspect a ModelKit index's manifest

### Synopsis

Print the contents of a ModelKit index manifest.

By default, kit will check local storage for the specified index. To
inspect an index stored on a remote registry, use the --remote flag.

```
kit index inspect [flags] INDEX
```

### Examples

```
# Inspect a local index:
kit index inspect my-org/my-model:all

# Inspect a local index by digest:
kit index inspect my-org/my-model@sha256:44136fa355b3678a1146ad16f7e8649e94fb4fc21fe77e8310c060f61caaff8a

# Inspect a remote index:
kit index inspect --remote registry.example.com/my-org/my-model:all
```

### Options

```
  -r, --remote             Check remote registry instead of local storage
      --plain-http         Use plain HTTP when connecting to remote registries
      --tls-verify         Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings   Path to TLS cert to add to trust store (flag can be repeated)
      --cert string        Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string         Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int    Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string       Proxy to use for connections (overrides proxy set by environment)
  -h, --help               help for inspect
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit index list

List ModelKit indexes in local storage

### Synopsis

List ModelKit indexes in local storage, with the number of ModelKits each one
references.

```
kit index list [flags]
```

### Examples

```
# List local indexes
kit index list

# List local indexes as JSON
kit index list --format json
```

### Options

```
      --format string   Output format: table or json (default "table")
  -h, --help            help for list
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit index pull

Download a ModelKit index from a registry

### Synopsis

Pull a ModelKit index from a remote registry. Only the index is pulled; use
'kit pull -l' or 'kit unpack -l' to pull one of its ModelKits.

```
kit index pull [flags] INDEX
```

### Examples

```
# Pull an index from a remote registry
kit index pull registry.example.com/my-org/my-model:all

# Pull a specific version of an index, which is stored untagged
kit index pull registry.example.com/my-org/my-model@sha256:44136fa355b3678a1146ad16f7e8649e94fb4fc21fe77e8310c060f61caaff8a
```

### Options

```
      --plain-http         Use plain HTTP when connecting to remote registries
      --tls-verify         Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings   Path to TLS cert to add to trust store (flag can be repeated)
      --cert string        Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string         Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int    Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string       Proxy to use for connections (overrides proxy set by environment)
  -h, --help               help for pull
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit index push

Upload a ModelKit index to a specified registry

### Synopsis

Push a ModelKit index to a remote registry. Only the index is pushed; the
ModelKits it references must already be in the destination repository.

```
kit index push [flags] SOURCE [DESTINATION]
```

### Examples

```
# Push an index to a remote registry
kit index push registry.example.com/my-org/my-model:all

# Push a local index to a remote registry
kit index push my-org/my-model:all registry.example.com/my-org/my-model:all
```

### Options

```
      --plain-http         Use plain HTTP when connecting to remote registries
      --tls-verify         Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings   Path to TLS cert to add to trust store (flag can be repeated)
      --cert string        Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string         Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int    Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string       Proxy to use for connections (overrides proxy set by environment)
  -h, --help               help for push
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit index remove

Remove a ModelKit from a ModelKit index

### Synopsis

Remove a ModelKit from a ModelKit index in local storage. The ModelKit itself
is not removed.

```
kit index remove [flags] INDEX MODELKIT
```

### Examples

```
# Remove a ModelKit from an index
kit index remove my-org/my-model:all my-org/my-model:q4_0

# Remove a ModelKit from an index by digest
kit index remove my-org/my-model:all sha256:44136fa355b3678a1146ad16f7e8649e94fb4fc21fe77e8310c060f61caaff8a
```

### Options

```
  -h, --help   help for remove
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit info

Show the configuration for a modelkit

### Synopsis

Print the contents of a modelkit config to the screen.

By default, kit will check local storage for the specified modelkit. To see
the configuration for a modelkit stored on a remote registry, use the
\--remote flag.

```
kit info [flags] MODELKIT
```

### Examples

```
# See configuration for a local modelkit:
kit info mymodel:mytag

# See configuration for a local modelkit by digest:
kit info mymodel@sha256:44136fa355b3678a1146ad16f7e8649e94fb4fc21fe77e8310c060f61caaff8a

# See configuration for a remote modelkit:
kit info --remote registry.example.com/my-model:1.0.0
```

### Options

```
      --plain-http         Use plain HTTP when connecting to remote registries
      --tls-verify         Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings   Path to TLS cert to add to trust store (flag can be repeated)
      --cert string        Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string         Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int    Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string       Proxy to use for connections (overrides proxy set by environment)
  -r, --remote             Check remote registry instead of local storage
  -f, --filter string      filter with node selectors
  -h, --help               help for info
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit init

Generate a Kitfile for the contents of a directory or remote repository

### Synopsis

Examine the contents of a directory or remote repository and attempt to generate
a basic Kitfile based on common file formats. Any files whose type (i.e. model,
dataset, etc.) cannot be determined will be included in a code layer.

For local directories, the generated Kitfile is saved in the target directory by
default. Use --output to specify a different path, or --output=- for stdout.
For remote repositories (--remote), the Kitfile is printed to stdout by default.

By default the command will prompt for input for a name and description for the Kitfile.

```
kit init [flags] PATH
```

### Examples

```
# Generate a Kitfile for the current directory:
kit init .

# Generate a Kitfile for files in ./my-model, with name "mymodel" and a description:
kit init ./my-model --name "mymodel" --desc "This is my model's description"

# Generate a Kitfile, overwriting any existing Kitfile:
kit init ./my-model --force

# Generate a Kitfile for a remote HuggingFace model:
kit init https://huggingface.co/myorg/mymodel --remote

# Generate a Kitfile for a HuggingFace dataset:
kit init huggingface.co/datasets/myorg/mydataset --remote

# Generate a Kitfile for a remote repository with a specific ref:
kit init myorg/mymodel --remote --ref v1.0

# Save the generated Kitfile to a specific path:
kit init myorg/mymodel --remote --output ./Kitfile
```

### Options

```
      --name string     Name for the ModelKit
      --desc string     Description for the ModelKit
      --author string   Author for the ModelKit
  -f, --force           Overwrite existing Kitfile if present
      --remote          Generate Kitfile from a remote HuggingFace repository
      --ref string      Branch or tag for remote repository (requires --remote) (default "main")
      --token string    Auth token for remote repository (requires --remote)
  -o, --output string   Output path for generated Kitfile ('-' writes to stdout; default: Kitfile in directory for local, stdout for remote)
      --depth int       Maximum directory depth to process when generating a Kitfile. Setting to a negative number processes all files individually
  -h, --help            help for init
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit inspect

Inspect a modelkit's manifest

### Synopsis

Print the contents of a modelkit manifest to the screen.

By default, kit will check local storage for the specified modelkit. To
inspect a modelkit stored on a remote registry, use the --remote flag.

```
kit inspect [flags] MODELKIT
```

### Examples

```
# Inspect a local modelkit:
kit inspect mymodel:mytag

# Inspect a local modelkit by digest:
kit inspect mymodel@sha256:44136fa355b3678a1146ad16f7e8649e94fb4fc21fe77e8310c060f61caaff8a

# Inspect a remote modelkit:
kit inspect --remote registry.example.com/my-model:1.0.0
```

### Options

```
      --plain-http         Use plain HTTP when connecting to remote registries
      --tls-verify         Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings   Path to TLS cert to add to trust store (flag can be repeated)
      --cert string        Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string         Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int    Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string       Proxy to use for connections (overrides proxy set by environment)
  -r, --remote             Check remote registry instead of local storage
  -h, --help               help for inspect
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit list

List modelkits in a repository

### Synopsis

Displays a list of modelkits available in a repository.

This command provides an overview of modelkits stored either in the local
repository or a specified remote repository. It displays each modelkit along
with its associated tags and the cumulative size of its contents. Modelkits
comprise multiple artifacts, including models, datasets, code, and
configuration, designed to enhance reusability and modularity. However, this
command focuses on the aggregate rather than listing individual artifacts.

Each modelkit entry includes its DIGEST, a unique identifier that ensures
distinct versions of a modelkit are easily recognizable, even if they share
the same name or tags. Modelkits with multiple tags or repository names will
appear multiple times in the list, distinguished by their DIGEST.

The SIZE displayed for each modelkit represents the total storage space
occupied by all its components.

Use the --filter (-f) flag to narrow results based on modelkit contents. Only
modelkits containing at least one layer matching the filter will be shown.

Valid filters have the format
\[types]:\[filters]
where \[types] is a comma-separated list of Kitfile fields (kitfile, model, datasets,
code, docs, or prompts) and \[filters] is an optional comma-separated list of names
or paths to match against. Multiple --filter flags use OR logic (a modelkit is
listed if it matches any filter).

Use the --format flag to change how results are printed. Valid values are
"table", "json", or a Go template. When a value other than "table" or "json"
is supplied, the flag contents are treated as a Go template executed once per
listed modelkit.

Template placeholders:

* {{ .Repo }} - repository name
* {{ .Tags }} - slice of tags for the modelkit
* {{ .Digest }} - digest of the modelkit
* {{ .ModelName }} - name defined in the Kitfile
* {{ .Size }} - total size of the modelkit, in bytes
* {{ .Author }} - author from the Kitfile

```
kit list [flags] [REPOSITORY]
```

### Examples

```
# List local modelkits
kit list

# List modelkits from a remote repository
kit list registry.example.com/my-namespace/my-model

# List only modelkits that contain prompt layers
kit list -f prompts

# List only modelkits containing a model
kit list -f model

# List modelkits that have either prompts or datasets
kit list -f prompts -f datasets

# List modelkits with a specific named prompt
kit list -f prompts:pdf-processing
```

### Options

```
      --format string        Output format: table, json, or Go template string (default "table")
  -f, --filter stringArray   Filter modelkits by content type (e.g., model, datasets, code, docs, prompts, mcpservers). Can be specified multiple times
      --plain-http           Use plain HTTP when connecting to remote registries
      --tls-verify           Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings     Path to TLS cert to add to trust store (flag can be repeated)
      --cert string          Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string           Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int      Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string         Proxy to use for connections (overrides proxy set by environment)
  -h, --help                 help for list
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit login

Log in to an OCI registry

### Synopsis

Log in to a specified OCI-compatible registry. Credentials are saved and used
automatically for future CLI operations

```
kit login [flags] [REGISTRY]
```

### Examples

```
# Login to ghcr.io
kit login ghcr.io -u github_user -p personal_token

# Login to docker.io with password from stdin
kit login docker.io --password-stdin -u docker_user
```

### Options

```
  -u, --username string    registry username
  -p, --password string    registry password or token
      --password-stdin     read password from stdin
      --plain-http         Use plain HTTP when connecting to remote registries
      --tls-verify         Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings   Path to TLS cert to add to trust store (flag can be repeated)
      --cert string        Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string         Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int    Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string       Proxy to use for connections (overrides proxy set by environment)
  -h, --help               help for login
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit logout

Log out from an OCI registry

### Synopsis

Log out from a specified OCI-compatible registry. Any saved credentials are
removed from storage.

```
kit logout [flags] REGISTRY
```

### Examples

```
# Log out from ghcr.io
kit logout ghcr.io
```

### Options

```
  -h, --help   help for logout
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit pack

Pack a modelkit

### Synopsis

Pack a modelkit from a kitfile using the given context directory.

The packing process involves taking the configuration and resources defined in
your kitfile and using them to create a modelkit. This modelkit is then stored
in your local registry, making it readily available for further actions such
as pushing to a remote registry for collaboration.

Unless a different location is specified, this command looks for the kitfile
at the root of the provided context directory. Any relative paths defined
within the kitfile are interpreted as being relative to this context
directory.

If --push is set along with a --tag pointing to a remote registry, the modelkit
is streamed directly to that registry instead of being stored locally.

```
kit pack [flags] DIRECTORY
```

### Examples

```
# Pack a modelkit using the kitfile in the current directory
kit pack .

# Pack a modelkit with a specific kitfile and tag
kit pack . -f /path/to/your/Kitfile -t registry/repository:modelv1

# Pack a modelkit and push it directly to a remote registry without storing it locally
kit pack . -t registry.example.com/my-org/my-model:latest --push
```

### Options

```
  -f, --file string           Specifies the path to the Kitfile explicitly (use "-" to read from standard input)
  -t, --tag string            Assigns one or more tags to the built modelkit. Example: -t registry/repository:tag1,tag2
      --compression string    Compression format to use for layers. Valid options: 'none', 'gzip', 'gzip-fastest', 'zstd' (default "none")
      --layer-format string   Packaging format to use for layers. Valid options: 'tar', 'raw' (default "tar")
      --use-model-pack        Pack model in ModelPack format instead of ModelKit
      --push                  Stream the packed modelkit directly to the remote registry specified by --tag, without storing it locally
      --plain-http            Use plain HTTP when connecting to remote registries
      --tls-verify            Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings      Path to TLS cert to add to trust store (flag can be repeated)
      --cert string           Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string            Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int       Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string          Proxy to use for connections (overrides proxy set by environment)
  -h, --help                  help for pack
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit pull

Retrieve modelkits from a remote registry to your local environment.

### Synopsis

Downloads modelkits from a specified registry. The downloaded modelkits
are stored in the local registry.

To pull from a ModelKit index, select a ModelKit with --label (-l). It is stored
locally under the index's reference.

```
kit pull [flags] registry/repository[:tag|@digest]
```

### Examples

```
# Pull the latest version of a modelkit from a remote registry
kit pull registry.example.com/my-model:latest

# Pull one ModelKit out of a ModelKit index
kit pull registry.example.com/my-model:all -l quantization=q4_0
```

### Options

```
  -l, --label stringArray   Select the ModelKit to pull from a ModelKit index by label, as key=value. Can be specified multiple times
      --plain-http          Use plain HTTP when connecting to remote registries
      --tls-verify          Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings    Path to TLS cert to add to trust store (flag can be repeated)
      --cert string         Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string          Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int     Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string        Proxy to use for connections (overrides proxy set by environment)
  -h, --help                help for pull
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit push

Upload a modelkit to a specified registry

### Synopsis

This command pushes modelkits from local storage to a remote registry.

If specified without a destination, the ModelKit must be tagged locally before
pushing.

```
kit push [flags] SOURCE [DESTINATION]
```

### Examples

```
# Push the ModelKit tagged 'latest' to a remote registry
kit push registry.example.com/my-org/my-model:latest

# Push a ModelKit to a remote registry by digest
kit push registry.example.com/my-org/my-model@sha256:44136fa355b3678a1146ad16f7e8649e94fb4fc21fe77e8310c060f61caaff8a

# Push local modelkit 'mymodel:1.0.0' to a remote registry
kit push mymodel:1.0.0 registry.example.com/my-org/my-model:latest
```

### Options

```
      --plain-http         Use plain HTTP when connecting to remote registries
      --tls-verify         Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings   Path to TLS cert to add to trust store (flag can be repeated)
      --cert string        Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string         Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int    Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string       Proxy to use for connections (overrides proxy set by environment)
  -h, --help               help for push
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit remove

Remove a modelkit from local storage or remote registry

### Synopsis

Removes a modelkit from storage on the local disk or from a remote registry.

The model to be removed may be specifed either by a tag or by a digest. If
specified by digest, that modelkit will be removed along with any tags that
might refer to it. If specified by tag (and the --force flag is not used),
the modelkit will only be removed if no other tags refer to it; otherwise
it is only untagged.

When removing modelkits in a remote registry, if the modelkit is specified by
digest, the modelkit will be deleted on the remote (if supported), removing any
tags that might refer to it. If the modelkit is specified with a tag (and the
\--force) flag is not used, it will only be untagged. If the --force flag is
used, the modelkit and all tags referring to it will be removed (i.e. the same
as if the digest for that tag was specified).

```
kit remove [flags] registry/repository[:tag|@digest]
```

### Examples

```
# Remove modelkit by tag
kit remove my-registry.com/my-org/my-repo:my-tag

# Remove modelkit by digest
kit remove my-registry.com/my-org/my-repo@sha256:<digest>

# Remove multiple tags for a modelkit
kit remove my-registry.com/my-org/my-repo:tag1,tag2,tag3

# Remove all untagged modelkits
kit remove --all

# Remove all locally stored modelkits
kit remove --all --force

# Untag a remote modelkit
kit remove --remote my-registry.com/my-org/my-repo:my-tag

# Delete a remote modelkit and all tags referring to it
kit remove --remote my-registry.com/my-org/my-repo@sha256:<digest>

# Delete a remote modelkit and all its tags without using a digest
kit remove --remote --force my-registry.com/my-org/my-repo:my-tag
```

### Options

```
  -f, --force              remove modelkit and all other tags that refer to it
  -a, --all                remove all untagged modelkits
  -r, --remote             remove modelkit from remote registry
      --plain-http         Use plain HTTP when connecting to remote registries
      --tls-verify         Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings   Path to TLS cert to add to trust store (flag can be repeated)
      --cert string        Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string         Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int    Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string       Proxy to use for connections (overrides proxy set by environment)
  -h, --help               help for remove
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit skill

Install the kit agent skill so AI agents can use the kit CLI

### Synopsis

Install a SKILL.md that teaches an AI agent how to use the kit CLI.

The skill documents Kitfile authoring and the pack, tag, login, push, pull, and
unpack workflow. It is embedded in the kit binary, so it always matches the
installed kit version and works without network access.

The skill is written into each agent's skills directory (for example
'.claude/skills' or the shared '.agents/skills'). Without --agents, kit
auto-discovers installed agents by checking their global config directories.
With --agents, specify agents as a comma-separated list (e.g.
\--agents=claude-code,cursor). By default the skill is installed globally
(user-scoped); when -d is specified it is installed into that project
directory instead.

```
kit skill [flags]
```

### Examples

```
# Install the kit skill for auto-detected agents (user-scoped)
kit skill

# Install for specific agents
kit skill --agents=claude-code,cursor

# Install into a project directory
kit skill -d /path/to/project

# Overwrite an existing installation
kit skill -o
```

### Options

```
  -d, --dir string        Install the skill into this project directory instead of globally (user-scoped)
      --agents string     Agents to install the skill for, as a comma-separated list (e.g. claude-code,cursor). Without a value, auto-discovers installed agents
  -o, --overwrite         Overwrite the skill if it already exists
  -i, --ignore-existing   Skip installation if the skill already exists
  -h, --help              help for skill
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit tag

Create a tag that refers to a modelkit

### Synopsis

Create or update a tag {target-modelkit} that refers to {source-modelkit}

This command assigns a new tag to an existing modelkit (source-modelkit) or
updates an existing tag, effectively renaming or categorizing modelkits for
better organization and version control. Tags are identifiers linked to
specific modelkit versions within a repository.

A full modelkit reference has the following format:

\[HOST\[:PORT_NUMBER]/]\[NAMESPACE/]REPOSITORY\[:TAG]

* HOST: Optional. The registry hostname where the ModelKit is located.
  Defaults to localhost if unspecified. Must follow standard DNS rules
  (excluding underscores).

* PORT_NUMBER: Optional. Specifies the registry's port number if a hostname
  is provided.

* NAMESPACE: Represents a user or organization's namespace, consisting of
  slash-separated components that may include lowercase letters, digits, and
  specific separators (periods, underscores, hyphens).

* REPOSITORY: The name of the repository, typically corresponding to the
  modelkit's name.

* TAG: A human-readable identifier for the modelkit version or variant.
  Valid ASCII characters include lowercase and uppercase letters, digits,
  underscores, periods, and hyphens. It cannot start with a period or hyphen
  and is limited to 128 characters.

Tagging is a powerful way to manage different versions or configurations of
your modelkits, making it easier to organize, retrieve, and deploy specific
iterations. Ensure tags are meaningful and consistent across your team or
organization to maintain clarity and avoid confusion.

```
kit tag SOURCE_MODELKIT[:TAG] TARGET_MODELKIT[:TAG] [flags]
```

### Examples

```
kit tag myregistry.com/myrepo/mykit:latest myregistry.com/myrepo/mykit:v1.0.0
```

### Options

```
  -h, --help   help for tag
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit unpack

Produce the components from a modelkit on the local filesystem

### Synopsis

Produces all or selected components of a modelkit on the local filesystem.

This command unpacks a modelkit's components, including models, code,
datasets, and configuration files, to a specified directory on the local
filesystem. By default, it attempts to find the modelkit in local storage; if
not found, it searches the remote registry and retrieves it. This process
ensures that the necessary components are always available for unpacking,
optimizing for efficiency by fetching only specified components from the
remote registry when necessary.

To unpack from a ModelKit index, select a ModelKit with --label (-l).

The content that is unpacked can be limited via the --filter (-f) flag. For example,
use
\--filter=model
to unpack only the model, or
\--filter=datasets:my-dataset
to unpack only the dataset named 'my-dataset'.

Valid filters have the format
\[types]:\[filters]
where \[types] is a comma-separated list of Kitfile fields (kitfile, model, datasets
code, docs, prompts, or mcpservers) and \[filters] is an optional comma-separated list of additional filters
to apply, which are matched against the Kitfile to further restrict what is extracted.
Additional filters match elements of the Kitfile on either the name (if present) or
the path used.

The filter field can be specified multiple times. A layer will be unpacked if it matches
any of the specified filters.

Use --as-skill to install SKILL.md prompt layers as agent skills instead of unpacking
them to their original paths. Without a value, kit auto-discovers installed agents by
checking their global config directories. With a value, specify agents as a comma-
separated list (e.g. --as-skill=claude-code,cursor). By default, skills are installed
globally (user-scoped). When -d is specified, skills are installed into that project
directory instead.

```
kit unpack [flags] [registry/]repository[:tag|@digest]
```

### Examples

```
# Unpack all components of a modelkit to the current directory
kit unpack myrepo/my-model:latest -d /path/to/unpacked

# Unpack one ModelKit out of a ModelKit index
kit unpack myrepo/my-model:all -l quantization=q4_0 -d /path/to/unpacked

# Unpack only the model and datasets of a modelkit to a specified directory
kit unpack myrepo/my-model:latest --filter=model,datasets -d /path/to/unpacked

# Unpack only the dataset named "my-dataset" to the current directory
kit unpack myrepo/my-model:latest --filter=datasets:my-dataset

# Unpack only the docs layer with path "./README.md" to the current directory
kit unpack myrepo/my-model:latest --filter=docs:./README.md

# Unpack the model and the dataset named "validation"
kit unpack myrepo/my-model:latest --filter=model --filter=datasets:validation

# Unpack a modelkit from a remote registry with overwrite enabled
kit unpack registry.example.com/myrepo/my-model:latest -o -d /path/to/unpacked

# Install SKILL.md prompt layers as agent skills (auto-detect agents)
kit unpack myrepo/my-model:latest --as-skill

# Install skills for specific agents (= required when specifying agents)
kit unpack myrepo/my-model:latest --as-skill=claude-code,cursor,windsurf

# Install skills into a project directory
kit unpack myrepo/my-model:latest --as-skill -d /path/to/project

# Overwrite existing skills
kit unpack myrepo/my-model:latest --as-skill -o
```

### Options

```
  -d, --dir string                 The target directory to unpack components into. This directory will be created if it does not exist
  -o, --overwrite                  Overwrites existing files and directories in the target unpack directory without prompting
  -i, --ignore-existing            Skip unpacking files if a file with that name already exists
  -f, --filter stringArray         Filter what is unpacked from the modelkit based on type and name. Can be specified multiple times
  -l, --label stringArray          Select the ModelKit to unpack from a ModelKit index by label, as key=value. Can be specified multiple times
      --include-remote             Include remote datasets in unpacked files
      --kitfile                    Unpack only Kitfile (deprecated: use --filter=kitfile)
      --model                      Unpack only model (deprecated: use --filter=model)
      --code                       Unpack only code (deprecated: use --filter=code)
      --datasets                   Unpack only datasets (deprecated: use --filter=datasets)
      --docs                       Unpack only docs (deprecated: use --filter=docs)
      --as-skill string[="auto"]   Install SKILL.md prompt layers as agent skills. Without a value, auto-discovers installed agents. With a value, specify agents as a comma-separated list (e.g. --as-skill=claude-code,cursor)
      --plain-http                 Use plain HTTP when connecting to remote registries
      --tls-verify                 Require TLS and verify certificates when connecting to remote registries (default true)
      --tls-cert strings           Path to TLS cert to add to trust store (flag can be repeated)
      --cert string                Path to client certificate used for authentication (can also be set via environment variable KITOPS_CLIENT_CERT)
      --key string                 Path to client certificate key used for authentication (can also be set via environment variable KITOPS_CLIENT_KEY)
      --concurrency int            Maximum number of simultaneous uploads/downloads (default 5)
      --proxy string               Proxy to use for connections (overrides proxy set by environment)
  -h, --help                       help for unpack
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```

## kit version

Display the version information for the CLI

### Synopsis

The version command prints detailed version information.

This information includes the current version of the tool, the Git commit that
the version was built from, the build time, and the version of Go it was
compiled with.

```
kit version [flags]
```

### Options

```
  -h, --help                        help for version
      --show-update-notifications   Enable or disable update notifications for the Kit CLI
```

### Options inherited from parent commands

```
      --config string      Alternate path to root storage directory for CLI
      --log-level string   Log messages above specified level ('trace', 'debug', 'info', 'warn', 'error') (default "info")
      --progress string    Configure progress bars for longer operations (options: none, plain, fancy) (default "plain")
  -v, --verbose count      Increase verbosity of output (use -vv for more)
```
