OPEN-SOURCE ENVIRONMENT FRAMEWORK
Give reasoning
room to work.
Chalk is a framework for RL environments focused on coding and mathematical reasoning. Write a task, give an agent tools, and see how it performs.
LOCAL FIRST / PYTHON 3.12+ / OPEN SOURCE
A clear space for
the work that matters.
Chalk brings tasks, tools, runs, and scoring into one environment. Its current framework gives you the pieces to build and inspect agent tasks; specialized coding and mathematics environments are still to come.
From task to transcript.
Write a task
Define steps in Python, supply data and tools, and choose how work will be scored.
Run in isolation
Use a supported container or virtual machine runtime to give the agent a working environment.
Inspect the result
Review scores, tool calls, and transcripts in the terminal dashboard or saved files.
Begin with
the example.
Create the default environment from your local Chalk checkout. Its example task asks an agent to find its Python executable and write its Python version to a file.
Read the complete walkthrough$ uv tool install .
$ chalk create-env my_env
$ cd my_env
$ uv sync --extra dev
$ uv run setup_data.py
$ uv run chalk tasks listUse Linux or macOS (WSL on Windows). Running the task also requires a supported runtime and model configuration.