SkyRL

Trillium Labs is a nonprofit AI research lab founded by Nathan Lambert that focuses on fostering open science in frontier AI. Its stated plans include releasing open-source reinforcement-learning environments for high-leverage tasks and documenting how AI is used to automate AI research, including work related to recursive self-improvement. The lab aims to make parts of frontier AI research that are difficult to study externally more accessible.

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.

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.

Reflexio is a learning platform for AI agents that turns user corrections, failed paths, and successful outcomes into reusable behavioral changes without retraining the underlying model. Its SDK and integration loop publish interaction outcomes, extract actionable feedback, store learned behaviors, and retrieve only relevant learnings during later inference; integrations are available through Python, REST, a CLI, and a portable coding-agent skill. Reflexio evaluates learnings against control responses and user-defined success criteria, tracking whether they solved the user's problem, required correction, or escalated to a human. It makes learnings auditable and revocable: they can be reviewed, rewritten, approved, rejected, or deleted, with rejected learnings removed from retrieval. Background processes consolidate duplicate signals and resolve conflicts or outdated lessons, while evidence from later sessions is used to revise learnings. The service supports managed, bring-your-own-key, customer-owned database, bring-your-own-cloud, and fully self-hosted deployments. The page identifies an official repository at github.com/ReflexioAI/reflexio and states that the same API can be used across deployment modes.

RSIAgent is an open-source, training-free multi-agent framework for recursive self-improvement in unfamiliar digital environments. It uses a Curriculum Agent, Actor Agent, and Verifier Agent while keeping model parameters fixed. Its learning loop has two stages: broad recursive self-exploration, in which agents perform diverse projects and consolidate verified experience, followed by deep recursive self-exploration, in which the Curriculum Agent selects focused practice around gaps and fragile successes. The Actor Agent interacts with software through executable Python or Bash programs and visual observations; the Verifier Agent independently checks task requirements and resulting environments. Procedures, scripts, successful experiences, and failure lessons are stored in persistent memory, which is frozen and reused for downstream task execution. The repository provides runtimes and benchmark integrations for OSWorld-V2 and Agents' Last Exam, along with setup, smoke-test, batch execution, reporting, recovery, and validation utilities. It requires Python 3.12 and a Linux host with Docker and /dev/kvm for the documented benchmark runs, and is licensed under Apache License 2.0.