CLI

The ReinforceNow command-line interface for training AI models.

Usage

rnow [OPTIONS] <COMMAND>

Commands

rnow login

rnow init

Initialize a new ReinforceNow project

rnow run

Submit project for training

rnow stop

Stop an active training run

rnow status

Check current authentication status

rnow orgs

List organizations or select one

rnow logout

Logout and remove stored credentials

rnow test

Test RL rollouts locally before submitting

rnow download

Download a trained model checkpoint

Usage

rnow login [OPTIONS]

Options

--force

Force new login even if already authenticated

--api-url api-url

API base URL (default: https://www.reinforcenow.ai/api)

May also be set with the REINFORCE_API_URL environment variable.

--help

Display the help for this command

rnow init

Initialize a new ReinforceNow project in the current directory.

Usage

rnow init [OPTIONS]

Options

--template template

Project template to use

Possible values:

start: Default single-turn RL template (alias for rl-single)
rl-single: Single-turn reinforcement learning (OpenMathReasoning dataset)
rl-tools: Multi-turn RL with tool calling (BrowseComp dataset)
tutorial-reward: Tutorial for creating reward functions
tutorial-tool: Tutorial for training agents with tools
sft: Supervised fine-tuning
new: Minimal template
blank: Empty project (config only)

--name name

Project name (prompts if not provided)

--help

Display the help for this command

Created Files

config.yml - Project configuration
train.jsonl - Training data
rewards.py - Reward functions (RL only)
env.py - Tool definitions (optional)
requirements.txt - Python dependencies (optional)

rnow run

Submit project for training on ReinforceNow platform.

Usage

rnow run [OPTIONS]

Options

--dir directory

Directory containing project files (default: current directory)

--name name

Custom name for the training run

--help

Display the help for this command

Required Files

config.yml or config.json - Project configuration
train.jsonl - Training data
rewards.py - Reward functions (RL projects only)

Optional Files

env.py - Custom tools/environment
requirements.txt - Additional Python dependencies

Multi-Model Training

You can train multiple models with the same configuration by specifying model.path as a list in your config.yml:

model:
  path:
    - Qwen/Qwen3-4B-Instruct-2507
    - Qwen/Qwen3-8B
    - Qwen/Qwen3-30B-A3B
    - openai/gpt-oss-20b
  qlora_rank: 32

When you run rnow run with a list of models, the CLI will:

Validate LoRA rank compatibility for all models before starting
Submit a separate training run for each model
Display all run IDs and dashboard links

Example Output

$ rnow run

Runs submitted:
  • Qwen3-4B-Instruct-2507: run_abc123
    https://www.reinforcenow.ai/dashboard/runs/run_abc123
  • Qwen3-8B: run_def456
    https://www.reinforcenow.ai/dashboard/runs/run_def456
  • Qwen3-30B-A3B: run_ghi789
    https://www.reinforcenow.ai/dashboard/runs/run_ghi789
  • openai/gpt-oss-20b: run_jkl012
    https://www.reinforcenow.ai/dashboard/runs/run_jkl012

rnow stop

Stop an active training run.

Usage

rnow stop <RUN_ID>

Arguments

RUN_ID

Run ID from rnow run output

Prompts for confirmation before stopping. Shows duration and charges after stopping.

Options

--help

Display the help for this command

rnow status

Check authentication status and running jobs.

Usage

rnow status

Shows whether you're authenticated, your active organization, and any currently running training jobs.

Options

--help

Display the help for this command

Example Output

$ rnow status
✓ Authenticated
Organization: org_abc123

Running jobs:
  • run_xyz789 - My Project (training)

rnow orgs

List all organizations you have access to, or select one as active.

Usage

rnow orgs [ORG_ID]

Without arguments, lists all organizations with their IDs and your role in each. The active organization is highlighted. With an ORG_ID argument, sets that organization as active.

Arguments

ORG_ID (optional)

Organization ID to set as active

Options

--help

Display the help for this command

Example

# List all organizations
rnow orgs

# Set active organization
rnow orgs org_abc123

rnow logout

Logout from ReinforceNow and remove stored credentials.

Usage

rnow logout

Options

--help

Display the help for this command

rnow test

Test RL rollouts locally before submitting to the platform.

Usage

rnow test [OPTIONS]

Options

--dir, -d directory

Project directory containing config.yml, rewards.py, env.py, train.jsonl (default: current directory)

--num-rollouts, -n count

Number of rollouts to run (default: 3)

--multi-turn / --single-turn

Allow multi-turn rollouts or force single-turn (default: --multi-turn)

--with-tools / --no-tools

Enable or disable tool use during rollout (default: --with-tools)

--model model

Override model name for sampling (otherwise uses config.model.path)

--api-url url

Base URL of the Next.js backend (default: https://www.reinforcenow.ai)

May also be set with the RNOW_API_URL environment variable.

--truncate, -t chars

Truncate message content to N characters (default: no truncation)

--verbose, -v

Show detailed output for each rollout turn

--help

Display the help for this command

Example

# Run 5 test rollouts with truncated output
rnow test -n 5 -t 200

# Test single-turn without tools
rnow test --single-turn --no-tools

# Test with a different model
rnow test --model Qwen/Qwen3-8B

Output

The command displays a spinner while running, then shows each rollout result with:

Message history with colored role tags ([system], [user], [assistant], [tool])
Reward scores for each reward function
Number of turns and tool calls

rnow download

Download a trained model checkpoint after a run completes.

Usage

rnow download <RUN_ID> [OPTIONS]

Arguments

RUN_ID

The run ID to download the model from

Options

--output, -o directory

Output directory for the downloaded model (default: ./model)

--help

Display the help for this command

Example

# Download model to default ./model directory
rnow download run_abc123

# Download to a custom directory
rnow download run_abc123 -o ./my-trained-model

CLI

On this page