Nistro AI CLI

Create and manage Nistro AI Inference and Nistro AI Postgres deployments from your terminal, scripts and CI with the nistro CLI. Install it with one command, sign in, and everything you do in the console is a command away. Its source and releases are on GitHub.

Install#

Install the latest release on macOS, Linux or Windows (Git Bash):

curl -fsSL https://raw.githubusercontent.com/nistroai/ctl/main/install.sh | sh

The installer downloads the build for your system, verifies its checksum and puts nistro in ~/.nistro/bin. On macOS and Linux it also adds that folder to your PATH in your shell profile, so open a new terminal before you run nistro. On Windows, add the folder to your PATH yourself.

To install a specific release, pass its version:

curl -fsSL https://raw.githubusercontent.com/nistroai/ctl/main/install.sh \
  | sh -s -- --version <version>

Releases are listed on GitHub.

Run the installer again to update. nistro --version shows the installed version.

Sign in#

nistro login

nistro login asks for your email and password and saves your session privately in ~/.nistro/sessions.json. When your session expires, run nistro login again. nistro logout clears the saved session.

Signed up with Google or GitHub? Set a password first with Forgot password.

In scripts and CI, pipe the password in instead of typing it:

printf '%s' "$NISTRO_PASSWORD" | nistro login --email "$NISTRO_EMAIL" --password-stdin

nistro login also reads NISTRO_EMAIL and NISTRO_PASSWORD from the environment.

The CLI manages the deployments in your own team.

Commands#

nistro inference and nistro postgres share these commands:

Inference adds recommend-hardware, and Postgres adds offers and placements, to check your options before you create. Add --help to any command for its options, for example nistro postgres create scale --help.

Output#

Commands print tables, with passwords and API keys hidden. Add -o json for the full response, credentials included:

nistro postgres describe <id> -o json

Configuration files#

create, update, offers and recommend-hardware take their settings as flags or from a JSON file passed with --config-file. JSON keys use underscores where flags use hyphens (storage_gib for --storage-gib), and flags override the file. --config-file - reads the JSON from stdin:

echo '{"vcpu": 2, "storage_gib": 50, "location": "ANY"}' \
  | nistro postgres create on-demand --config-file -

Inference On-Demand#

nistro inference create on-demand --model-url google/gemma-4-26B-A4B \
  --max-model-len 32768 --location ANY

• --model-url: a Hugging Face model id or huggingface.co URL.
• --max-model-len: the context length in tokens, up to the model's maximum.
• --location: US, EU, ASIA or ANY.
• Optional: --quantization. The default, Auto, uses what the repository provides.

Based on your model, context length and location, Nistro AI finds the most cost-effective GPU that meets your requirements. When the deployment is ready, nistro inference describe <id> -o json shows its endpoint and API key; see Call the API to send requests.

Inference Scale#

Find the machines that can run the model in a location, then create on one of them:

nistro inference recommend-hardware scale --model-url google/gemma-4-26B-A4B \
  --max-model-len 32768 --location US
nistro inference create scale --model-url google/gemma-4-26B-A4B \
  --max-model-len 32768 --cloud <cloud> --region <region> \
  --instance-type <instance-type> --replicas 1

• --location: US, EU or ASIA.
• --cloud, --region and --instance-type: a row from recommend-hardware scale.

recommend-hardware scale lists the machines in that location that fit the model, with each one's cloud, region, GPU and instance type. create scale sets the number of replicas with --replicas (required), and takes an optional --quantization and --gpu-type (keep the default, Any, or name a specific GPU).

Change the number of replicas later with:

nistro inference update scale <id> --replicas 2

Postgres On-Demand#

nistro postgres create on-demand --vcpu 2 --storage-gib 50 --location ANY

• --vcpu: 1 to 8. Memory follows at 4 GiB per vCPU.
• --storage-gib: 10 to 1024.
• --location: US, EU, ASIA or ANY.
• Optional: --database-name and --database-username, both app by default, and --database-password, generated if you leave it out.

nistro postgres offers on-demand takes the same capacity and location and shows the matching offers before you create. To scale, pass a new vCPU count or a larger storage size:

nistro postgres update on-demand <id> --vcpu 4 --storage-gib 100

Postgres Scale#

List the clouds, regions and instance types you can use, then create on one of them:

nistro postgres placements scale
nistro postgres create scale --cloud <cloud> --region <region> \
  --instance-type <instance-type> --storage-gib 100

• --cloud, --region and --instance-type: a row from placements scale. The instance type sets the CPU and memory.
• --storage-gib: 10 to 4096.
• --replicas: standby replicas, 0 to 8. The default is 1.
• Optional: the database flags above, --extensions as a comma-separated list such as vector,pg_trgm, and --parameters name=value, once per parameter. See Extensions and Parameters for the allowed names.

nistro postgres offers scale takes the same placement and storage and shows the matching offers before you create.

After you create a deployment, you can grow its storage, change the number of standbys and set parameters:

nistro postgres update scale <id> --replicas 2
nistro postgres update scale <id> --parameters work_mem=32MB

--parameters replaces the deployment's whole parameter set, so pass every parameter you want to keep.