prime lab setup

Initialize a development workspace for building RL environments

Overview

The prime lab setup command initializes a new workspace for developing environments with Verifiers. It creates the recommended directory structure, downloads starter configurations, and optionally installs prime-rl for local training.

Usage

prime lab setup [OPTIONS]

What It Does

1. Initializes a Python project (if not already present) using uv init
2. Installs verifiers via uv add verifiers
3. Creates directory structure:
   configs/
   ├── endpoints.toml      # API endpoint configuration
   ├── rl/                 # Training configs for Hosted Training
   ├── eval/               # Multi-environment eval configs
   └── gepa/               # Prompt optimization configs
   .prime/
   └── skills/             # Bundled workflow skills
   environments/
   └── AGENTS.md           # Documentation for AI coding agents
   AGENTS.md               # Top-level agent documentation
   CLAUDE.md               # Claude-specific pointer
   
4. Downloads starter files from the verifiers repository
5. Optionally installs prime-rl for local training

Options

--prime-rl · flag

Install prime-rl and download prime-rl training configurations. Use this if you want to train models locally.

--skip-install · flag

Skip uv init and uv add verifiers steps. Use when adding to an existing project.

--skip-agents-md · flag

Skip downloading AGENTS.md, CLAUDE.md, and environments/AGENTS.md files.

--agents · string

Comma-separated list of coding agents to scaffold (codex, claude, cursor, opencode, amp).

Example: --agents codex,claude

--no-interactive · flag

Disable interactive prompts for coding agent selection.

Examples

Basic Setup

Initialize a new workspace in an empty directory:

prime lab setup

Output:

No pyproject.toml found, initializing uv project...
Running: uv init
Running: uv add verifiers
Downloaded configs/endpoints.toml from https://github.com/primeintellect-ai/verifiers
Downloaded AGENTS.md from https://github.com/primeintellect-ai/verifiers

Add to Existing Project

Add verifiers to an existing Python project:

# First add verifiers to your project
uv add verifiers

# Then run setup without the install step
prime lab setup --skip-install

Setup with Local Training

Install prime-rl for training models locally:

prime lab setup --prime-rl

This will:

• Clone and install prime-rl
• Download prime-rl training configurations
• Install all environments from environments/ into the prime-rl workspace

Specify Coding Agents

Configure skill directories for specific coding agents:

prime lab setup --agents codex,opencode

Next Steps

After running prime lab setup, you can:

1. Create an environment:

   prime env init my-env
   

2. Run an evaluation:

   prime eval run my-env -m gpt-4.1-mini -n 5
   

3. View results:

   prime eval tui
   

4. Train a model (if you used --prime-rl):

   uv run prime-rl configs/prime-rl/wiki-search.toml
   

Configuration Files

The setup command downloads several starter configuration files:

Endpoints Configuration

configs/endpoints.toml defines model API endpoints:

[[endpoint]]
endpoint_id = "gpt-4.1-mini"
model = "gpt-4.1-mini"
url = "https://api.openai.com/v1"
key = "OPENAI_API_KEY"

[[endpoint]]
endpoint_id = "claude-sonnet"
model = "claude-sonnet-4-5-20250929"
url = "https://api.anthropic.com"
key = "ANTHROPIC_API_KEY"
api_client_type = "anthropic_messages"

Evaluation Configs

configs/eval/ contains multi-environment evaluation configurations:

# configs/eval/multi-env.toml
model = "openai/gpt-4.1-mini"
num_examples = 50

[[eval]]
env_id = "gsm8k"
num_examples = 100

[[eval]]
env_id = "math-python"

Training Configs

configs/rl/ contains training configurations for use with Hosted Training or prime-rl.

Troubleshooting

uv Not Found

If uv is not installed:

curl -LsSf https://astral.sh/uv/install.sh | sh

Permission Errors

Ensure you have write permissions in the target directory.

Git Conflicts

If you’re in a git repository, the setup may create files that conflict with existing files. Review changes carefully before committing.