Overview
Splash CLI is a command-line interface for administrators to manage your Poolside deployment. You can use it to:- Authenticate with the Poolside API
- Manage models (create, list, edit, scale, delete)
- Configure teams and permissions
- Bootstrap new deployments
- Manage platform configurations
Installation
You can download the Splash CLI from the Poolside team. To install:Authentication
Splash CLI authenticates through your organization’s identity provider (IdP). Runningsplash login opens a browser window to complete authentication and stores the resulting token in your local configuration.
Configuration
Manage your Splash CLI configuration:Model management
Models are the core of the Poolside platform. Splash CLI provides commands for managing models.List models
Get model details
Create models
Important model creation flags
| Flag | Description | Default |
|---|---|---|
--capabilities | Model capabilities (line_numbers, empty_system_prompt, multi_turn) | None |
--checkpoint | S3 path to model checkpoint | Required |
--lora-checkpoint | S3 path to LoRA checkpoint | None |
--context-length | Maximum context length | 8192 |
--default | Set as default model for tenant per type | false |
--description | Model description | None |
--gpus | Number of GPUs required | 2 |
--max-tokens | Maximum tokens to generate | 8192 |
--min-p | Minimum probability for sampling | 0.10 |
--mode | Model mode (kubernetes, static) | kubernetes |
--public | Make model available across tenants | false |
--replicas | Number of model replicas | 1 |
--temperature | Sampling temperature | 0.40 |
--top-p | Top-p sampling parameter | 0.90 |
--type | Model type (chat, oss, completion) | chat |
--url | URL for static model | None |
Example: Chat model (Malibu)
Example: Completion model (Point)
Edit models
- Replicas count
- Model description
- Queue length (via
extra_envsettings) - Resource allocations
Manage prompt templates
Scale models
Clone models
Delete models
Team management
Teams control access to Poolside features and models.List teams
Create teams
Create teams using CEL (Common Expression Language) conditions to specify membership:Edit teams
Delete teams
Bootstrap configuration
The first step is setting up a tenant:Shell completion
Splash supports shell completion for easier command-line use:Common workflows
Manage model queue length
To adjust the maximum pending requests for a model:extra_env section:
Add administrators
- Identify the administrator team (usually named
admin). - Edit the team to include new administrators: