Model deployment
A model deployment, called a serving in the API and the CLI, is a model running behind an endpoint on one of your devices. The platform starts it, keeps it up, and proxies calls to it, so a client never has to reach the device directly.
It is a service whose command the platform supplies rather than you: the model runs through the CLIKA runtime on the device, and the endpoints a client calls are derived from the model's type rather than configured by hand.
The commands sit under the servings group.
Managing a serving
| Command | Purpose |
|---|---|
servings list | List servings. |
servings get <id> | Get one, with its state and its endpoint. |
servings create --body '<json>' | Create one. |
servings update <id> --body '<json>' | Change one. |
servings start <id> | Start it. |
servings stop <id> | Stop it without removing it. |
servings restart <id> | Restart it. |
servings delete <id> | Delete it. |
servings key <id> | Reveal the deployment key the proxy wants. |
servings fit | Which devices a model can be deployed to. |
projects servings <id> | List one project's servings. |
clika-cli servings list
clika-cli servings get 4b7e0c1a-2d3e-4f50-6172-8394a5b6c7d8
Starting and stopping are separate from creating and deleting for the same reason they are on a service: stopping frees the device's accelerator while keeping the deployment and its configuration, so you can bring the same endpoint back without redefining it.
servings create creates the serving pending and starts nothing; run servings start <id> next. A serving that has sat in pending for minutes with no device state has not been started. A start reads starting for about two to three minutes when a benchmark has already put the engine on the device, and for five to nine minutes on a device that has never run it; on a device busy with a job the start waits behind that job.
Calling a serving through the platform
clika-cli servings devices-proxy <id> <device_id> <upstream>
The proxy is how a client reaches a deployed model without a route to the device. You address the platform, name the serving, the device and the upstream path, and the platform forwards the call to the model and the response back.
<upstream> is the path on the model's own endpoint. Which paths exist depends on the model's type, because the runtime exposes the endpoints that type implies rather than a fixed set.
Four variants cover the HTTP methods, since a proxied call has to carry the caller's own verb:
| Command | Method it proxies |
|---|---|
servings devices-proxy | GET |
servings devices-proxy-create | POST |
servings devices-proxy-update | PUT |
servings devices-proxy-update-patch | PATCH |
servings devices-proxy-delete | DELETE |
The proxy wants the deployment's own key on the Authorization header, which servings key <id> prints; the proxy subcommands send no request body. For anything beyond a check that the endpoint answers, call the proxy endpoint from your own client rather than through the CLI. The CLI's value here is discovery: servings get tells you the serving is up and what its endpoint is.
What it needs
A model deployment consumes a licensed inference engine on the device, so the same entitlement gate that applies to a job definition applies here. A device without the engine package cannot host one, and the refusal is not forceable. Provision the engine with devices engine-package <name-or-id>, and see runtime licenses for the entitlement that permits it.
Full command reference
Every command below is generated from the deployment's own API description, so
one subcommand is exactly one platform operation. Each entry names the method,
the endpoint and the MCP tool name, so the same operation is
identifiable whichever surface you drive it from. Path parameters are positional
arguments, query parameters are flags, and a request body is --body or
--body-file. The hand-written commands, the ones that stream, propagate an
exit code, or hand your terminal to ssh, carry no operation line.
The prose above covers the commands most people reach for. This section is the complete surface, for when you need the flag you have not used before.
clika-cli servings
servings has 17 subcommands.
clika-cli servings api-reference
Download a model serving's offline API reference
clika-cli servings api-reference <id> [flags]
Positional arguments: required <id>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--format | string | none | json (OpenAPI 3, the default), yaml (OpenAPI 3) or markdown |
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings create
Create model serving
clika-cli servings create [flags]
| Flag | Type | Default | Meaning |
|---|---|---|---|
--body | string | none | request body (inline JSON) |
--body-file | string | none | request body (path to a JSON file) |
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings delete
Delete model serving
clika-cli servings delete <id> [flags]
Positional arguments: required <id>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--purge_model | string | none | Also remove the deployment's model from each device's engine model cache where no other deployment on that device uses it. Default false: every device keeps its cached model. |
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings devices-proxy
Call a model serving on one device through the platform
clika-cli servings devices-proxy <id> <device_id> <upstream> [flags]
Positional arguments: required <id>, <device_id>, <upstream>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings devices-proxy-create
Call a model serving on one device through the platform
clika-cli servings devices-proxy-create <id> <device_id> <upstream> [flags]
Positional arguments: required <id>, <device_id>, <upstream>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings devices-proxy-delete
Call a model serving on one device through the platform
clika-cli servings devices-proxy-delete <id> <device_id> <upstream> [flags]
Positional arguments: required <id>, <device_id>, <upstream>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings devices-proxy-update
Call a model serving on one device through the platform
clika-cli servings devices-proxy-update <id> <device_id> <upstream> [flags]
Positional arguments: required <id>, <device_id>, <upstream>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings devices-proxy-update-patch
Call a model serving on one device through the platform
clika-cli servings devices-proxy-update-patch <id> <device_id> <upstream> [flags]
Positional arguments: required <id>, <device_id>, <upstream>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings fit
Which devices a model can be deployed to
clika-cli servings fit [flags]
| Flag | Type | Default | Meaning |
|---|---|---|---|
--model_id | string | none | Model ID (UUID) |
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings get
Get model serving
clika-cli servings get <id> [flags]
Positional arguments: required <id>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings key
Reveal a model serving's deployment key
clika-cli servings key <id> [flags]
Positional arguments: required <id>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings key-rotate
Rotate a model serving's deployment key
clika-cli servings key-rotate <id> [flags]
Positional arguments: required <id>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings list
List model servings
clika-cli servings list [flags]
| Flag | Type | Default | Meaning |
|---|---|---|---|
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings restart
Restart model serving
clika-cli servings restart <id> [flags]
Positional arguments: required <id>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings start
Start model serving
clika-cli servings start <id> [flags]
Positional arguments: required <id>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings stop
Stop model serving
clika-cli servings stop <id> [flags]
Positional arguments: required <id>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--raw | bool | false | print raw response without pretty-printing |
clika-cli servings update
Update model serving
clika-cli servings update <id> [flags]
Positional arguments: required <id>.
| Flag | Type | Default | Meaning |
|---|---|---|---|
--body | string | none | request body (inline JSON) |
--body-file | string | none | request body (path to a JSON file) |
--raw | bool | false | print raw response without pretty-printing |
Related
- Services: the free-form counterpart, where you supply the command.
- Runtime licenses: the entitlement a deployed model runs under.
- Devices: the device that hosts the deployment.
- Artifacts and models: registering the model to deploy.
- Model deployment concept: what a deployment is, in prose.