rl-env-use

rLLM is an open-source framework from UC Berkeley's Sky Computing Lab for training language agents with reinforcement learning. It connects agent code to evaluation and training through a decorator-based Python API or CLI workflow: agents run in sandboxes, the model gateway captures token IDs and log probabilities, traces are organized into episodes and trajectories, evaluators compute rewards, and a training backend updates the model. It supports existing or custom agent harnesses, Docker, Daytona, Modal, and local sandboxes, with training backends including verl, tinker, and Fireworks, and methods including GRPO, REINFORCE, RLOO, supervised fine-tuning, and on-policy distillation. The project also provides integrated benchmarks and requires Python 3.11 or later.

RSI-Anything is a domain-general framework by Gen-Verse for recursive self-improvement using world-model predictions and interaction feedback. Its shared workflow proposes candidate actions, predicts their consequences with a JEPA-based world model, selects actions, evaluates them externally, learns from the resulting experience, and retains improvements for later rounds. The repository provides a Python command-line workflow and task adapters for code repair, GUI interaction, laboratory planning, reaction optimization, and immune-science workflows, with domain-specific states, actions, prediction targets, and feedback. It can be installed from the repository with Python 3.10 or newer and includes configuration-driven planning, execution, reporting, run resumption, tests, and task guides. The project-owned code is licensed under Apache-2.0.

SkyRL is an open-source reinforcement-learning training framework for AI models. Nathan Lambert said Trillium plans to use it as part of its open post-training stack.

RouteLLM is an open-source framework from lm-sys for serving and evaluating routers that direct requests between stronger and cheaper language models. It provides trained routers, a drop-in replacement for the OpenAI client, and an OpenAI-compatible server; routing decisions use configurable cost thresholds that can be calibrated for a target share of requests sent to the stronger model. The framework supports comparing routers across benchmarks and extending it with additional routers, model pairs, and providers.

Ratel is an in-process context-engineering platform for AI agents. It catalogs tools, skills, and persistent facts, then searches the catalog on each turn and progressively discloses only the capabilities relevant to the request instead of placing every tool schema and instruction in the context window. Its retrieval supports BM25, semantic, and hybrid ranking without requiring a vector database. The project includes a Rust core, TypeScript and Python SDKs, an MCP server, and a CLI. Tools can be registered directly or ingested from an MCP server; skills can associate instructions with tools, while registered facts are re-injected when they are no longer fresh in the conversation. The repository describes the system as reducing token usage and mitigating tool overload.