Build and benchmark agents against GridWorld, CartPole, and MountainCar with a stable, vendor-neutral C ABI. Opt into Python bindings via cffi — no CPython extensions required.
Modular Envs
Swap dynamics & reward models with a stable ABI.
Bare‑metal Speed
Tight C loops. Zero Python in hot paths.
Bindings by Choice
Use C directly or import via cffi in Python.
Cross‑tooling
Works with C, C++, Zig, Rust FFI, and more.
C (source)
git clone https://github.com/your-org/rl-c-envs && cd rl-c-envs && makePython (cffi)
pip install cffi && pip install rl-c-envs// env.h — stable C ABI (excerpt)
// Initialize an environment by ID (gridworld, cartpole, mountaincar)
typedef struct rl_env rl_env;
// Create / destroy
tl_rl_api rl_env* rl_create(const char* env_id);
tl_rl_api void rl_destroy(rl_env* env);
// Reset / step
// action is discrete (int); returns reward, done, and writes next state into user buffer
// returns 0 on success
tl_rl_api int rl_reset(rl_env* env, float* state_out, int* state_len);
tl_rl_api int rl_step(rl_env* env, int action, float* state_out, int* state_len, float* reward_out, int* done_out);
Environments
3
gridworld • cartpole • mountaincar
ABI Stability
v1
semantic, backwards compatible
Bindings
cffi
optional Python integration
Minimal RL environments with a stable C API, example agents, optional Python bindings, and a small Raylib-based viewer. Portable, reproducible, and easy to embed.
Provide a minimal set of RL environments exposed via a stable C API (ABI-safe), easy to call from any language.
Most RL stacks are heavy and Python-centric. TinyGym‑C is small, fast, and portable.
Focused feature set to start simple and reproducible.
Plain C99 with CMake, tested and formatted for CI and releases.
Watch agents play and script experiments from Python if you want to.
In short: TinyGym‑C is a minimal C library for reinforcement learning environments and example agents, with a Python bridge and a simple viewer. It’s a stripped‑down, embeddable version of Gym — small enough to understand fully, but practical for learning, experimenting, or extending.