Documentation

See the documentation for a technical overview of the platform and train your first agent

Quick Start

1. Install uv (Python package manager)

# macOS/Linux:
$ curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows:
PS> powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

2. Install ReinforceNow

uv init && uv venv --python 3.11
source .venv/bin/activate  # Windows: .\.venv\Scripts\Activate.ps1
uv pip install rnow

3. Authenticate

rnow login

4. Create & Run Your First Project

rnow init --template sft
rnow run

That's it! Your training run will start on ReinforceNow's infrastructure. Monitor progress in the dashboard.

Core Concepts

Go from raw data to a reliable AI agent in production. ReinforceNow gives you the flexibility to define:

1. Reward Functions

Define how your model should be evaluated using the @reward decorator:

from rnow.core import reward, RewardArgs

@reward
async def accuracy(args: RewardArgs, messages: list) -> float:
    """Check if the model's answer matches ground truth."""
    response = messages[-1]["content"]
    expected = args.metadata["answer"]
    return 1.0 if expected in response else 0.0

→ Write your first reward function

2. Tools (for Agents)

Give your model the ability to call functions during training:

from rnow.core import tool

@tool
def search(query: str, max_results: int = 5) -> dict:
    """Search the web for information."""
    # Your implementation here
    return {"results": [...]}

→ Train an agent with custom tools

3. Training Data

Create a train.jsonl file with your prompts and reward assignments:

{"messages": [{"role": "user", "content": "Balance the equation: Fe + O2 → Fe2O3"}], "rewards": ["accuracy"], "metadata": {"answer": "4Fe + 3O2 → 2Fe2O3"}}
{"messages": [{"role": "user", "content": "Balance the equation: H2 + O2 → H2O"}], "rewards": ["accuracy"], "metadata": {"answer": "2H2 + O2 → 2H2O"}}
{"messages": [{"role": "user", "content": "Balance the equation: N2 + H2 → NH3"}], "rewards": ["accuracy"], "metadata": {"answer": "N2 + 3H2 → 2NH3"}}

→ Learn about training data format

Contributing

We welcome contributions! ❤️ Please open an issue to discuss your ideas before submitting a PR