Community Packages - Agent Environments Built with Strands for RL Post-training and Evaluation

[Lawhy]

Hi! we're a group exploring Strands as a scaffold for agentic RL training, and we've built a small set of community packages that may align with the Strands Labs direction.

First, we created strands-sglang: an SGLang model provider designed for RL needs, including token-level trajectories and other customizable hooks for post-training.

Building on top of that, we introduced strands-env, which provides a unified abstraction for RL environments with Strands. The goal is an "environments-as-services" setup for both training and evaluation, enabling stable, high-concurrency rollouts.

These packages are already being used in several of our agentic RL training and benchmarking efforts.

We've been following the Strands Labs initiative and wanted to ask: would these projects be interesting to host under Labs (or collaborate more closely in that direction)?