Orbax

JAX is a Python library for accelerator-oriented array computation and program transformation, designed for numerical computing and large-scale machine learning. It transforms Python and NumPy-style functions through automatic differentiation with `jax.grad`, automatic batching and vectorization with `jax.vmap`, and just-in-time compilation with `jax.jit`. It uses XLA to compile computations for GPUs, TPUs, and other accelerators. JAX supports compiler-driven multi-device execution: `Mesh`, `PartitionSpec`, and `NamedSharding` declare array placement across device meshes, allowing distributed training code to include automatically managed parallel work such as gradient synchronization. Its compilation tools also include ahead-of-time compilation, portable artifact export with `jax.export`, and conversion to TensorFlow SavedModels through `jax2tf` for serving integrations. The project describes itself as a research project rather than an official Google product.

Flax NNX is a neural-network API for JAX developed by the Flax team in collaboration with the JAX team. Released in 2024, it uses Python reference semantics so models can be expressed as regular Python objects with reference sharing and mutability, supporting inspection, debugging, and analysis. Its API includes layers and modules such as Linear, convolution, normalization, attention, recurrent cells, and dropout, along with patterns for replicated training, serialization, and checkpointing. In distributed training, it can manage model state while JAX handles computation across a device mesh.

Axolotl is a free, open-source framework for post-training and fine-tuning large language models. It uses YAML configuration to define workflows for preprocessing, training, evaluation, quantization, inference, preference tuning, reinforcement learning, and reward modeling. The project is hosted by axolotl-ai-cloud and documents support for distributed training, mixture-of-experts models, multimodal models, and multiple fine-tuning methods.

AX is Google's open agentic orchestration runtime for declaring and running autonomous agent workloads in Kubernetes clusters. It defines Tasks, Workspaces, Gateways, and Models as ax.io/v1alpha1 manifests: Tasks run agent code in isolated sandboxes, Workspaces preconfigure Git repositories, MCP servers, and skill packages, Gateways restrict outbound traffic with host allowlists, and Models configure the platform's LLM access through Kubernetes secrets. The control plane runs on Agent Substrate and is operated through a kubectl-shaped CLI that applies manifests, watches task status, opens SSH sessions into sandboxes, and suspends or resumes checkpointed agents. AX is described as being designed for high-throughput cluster operation and is still subject to breaking changes before a stable release. It is distributed under the Apache License 2.0.

Blaxel is an infrastructure platform for autonomous AI agents that provides isolated microVMs (which boot in milliseconds and resume in ~25ms), persistent shared memory, and programmable networking to run, connect, and operate agents at scale. It offers a distributed shared-storage feature (presented in videos as "Agent Drive") that can be mounted as a POSIX-style drive so agents in separate sandboxes can share files, cache dependencies, exchange context/memory, and coordinate parallel work in real time.