Python API
Look up a Python symbol or choose a module below. For a working example, start with the docs.
$ pip install rlmesh==0.1.0This is the Python symbol reference. For a complete first run, follow the
quickstart. When you need an exact signature,
start with EnvServer to serve an
environment, RemoteEnv to connect, and
run() to evaluate a model. The pages below are
generated from the package version shown above.
Native values or a backend
The top-level classes (rlmesh.RemoteEnv, rlmesh.Model, and their peers)
keep RLMesh-native tensor values and need no array library. The backend modules
(rlmesh.numpy, rlmesh.torch,
rlmesh.jax) expose corresponding classes that
decode tensor leaves to framework arrays. Choose the module that matches
your model’s values.
import rlmesh # native values, no dependencies
from rlmesh.numpy import RemoteEnv # the same client, NumPy arrays in and outManaged evaluations are created and inspected in the dashboard. Python and TypeScript SDK guides will follow.
Start here
- EnvFactoryclassAuthoring base that *builds* environment(s): set
tagsand implementmake. - RemoteEnvclassDependency-free client for one remote environment endpoint.
- ModelclassDependency-free model over RLMesh-native values.
- run()functionDrive
modelagainstenvto completion and return aRunResult.
Modules
Core
Serve environments, connect clients, run models, and record results.
Contracts
Describe what an env publishes and what a model ingests, and resolve the two.
Spaces and types
RLMesh-native spaces, Gymnasium conversion, and shared value types.
Value backends
The same clients and models, decoding tensor leaves to NumPy, Torch, or JAX.
More references
- Native specs — Native spec and value classes exposed through the Python wrappers.
- Gymnasium space support — Which Gymnasium spaces convert, and how.
- rlmesh (Rust) — The Rust SDK facade.
- rlmesh-spaces (Rust) — Space specs and values.
- rlmesh-proto (Rust) — Wire schemas.