Dev Tools
Agent-Skills
Curated AI agent skills for Microsoft and Azure, sourced from Microsoft Learn.
★ 772 starsCC-BY-4.0
View guide →Microsoft's Physical AI Toolchain: tiered pipelines from robot data capture to edge deployment.
physical-ai-toolchain is an open-source dev tools project by microsoft (microsoft/physical-ai-toolchain on GitHub). It has 123 GitHub stars and 62 forks, is released under the MIT license, and was last updated on GitHub on 8 October 2026.
Choose your tier honestly: start at T0 (one laptop, one robot, no cloud) unless your scale already demands more.
Clone the repository and run ./setup-dev.sh, which installs Python 3.12 via uv, creates a virtual environment, and installs training dependencies.
Capture demonstrations on the robot with ROS 2 bag recording and move the data to your laptop.
Curate the data, train an imitation policy with LeRobot tooling, validate it locally, and run the policy back on the robot.
Six tiers from a zero-cloud laptop setup to multi-site fleet delivery, each stating the minimum infrastructure it needs so teams adopt only what their scale demands.
Simulation and synthetic data, edge data capture, cloud data pipelines with ROS-to-LeRobot conversion, training infrastructure, evaluation, deployment packaging, and edge inference are all addressed.
Native support for Isaac Sim and Isaac Lab simulation, OSMO workflow orchestration, and Jetson edge inference.
Infrastructure as Code with Terraform, MLflow experiment tracking, Entra ID with managed identities, and GitHub Actions CI/CD enter at the tiers that need them.
Inference can be offloaded to an on-premise GPU through Kubernetes operators, saving robot battery and cost.
Optional agents can turn a high-level instruction into executed collection, training, evaluation, and deployment stages, within the same security boundaries as manual workflows.
Get the framework, tier recipes, and documentation locally.
git clone https://github.com/microsoft/physical-ai-toolchain.git
cd physical-ai-toolchainThe script installs Python 3.12 via uv, creates the virtual environment, and installs training dependencies.
./setup-dev.shWork through the Tier 0 Dev recipe in the docs: capture, curate, train, validate, and run locally. # see README Start on a Laptop section
# see the Tier 0 - Dev recipe linked from the READMEIt is maintained by Microsoft on GitHub.
Yes, the toolchain is MIT-licensed and its default T0 path needs no Azure subscription at all; cloud tiers use your own Azure account and incur normal Azure costs.
No key is needed for the local T0 path. Cloud tiers authenticate with your Azure and Entra ID setup, and agent features use your configured model access.
Standalone tools cover one stage, such as simulation or training; this toolchain is an integrated, tiered framework spanning capture through fleet deployment with production security and governance built in.
Clone the repo, run ./setup-dev.sh, and follow the Tier 0 Dev recipe to close the full loop on one laptop and one robot.
Curated AI agent skills for Microsoft and Azure, sourced from Microsoft Learn.
Scrape any website with a prompt: LLM-powered scraping pipelines in a few lines of Python
Google's cross-platform framework for on-device ML perception: vision, text and audio pipelines
NVIDIA NeMo Data Designer: build validated, multimodal synthetic datasets from scratch or from seed data.
A Node.js crawler that turns websites into LLM-ready data and extracts structured SERP results.
Google's Python library for extracting structured data from unstructured text with LLMs, with grounding
The most-starred other Dev Tools projects in the LibreAI catalogue — the realistic shortlist if physical-ai-toolchain is not the right fit.
The open source coding agent
The model-definition framework behind modern AI — thousands of pretrained transformers for text, vision, and audio.
Anthropic's agentic coding tool that lives in your terminal and works across your whole codebase
Cross-platform desktop all-in-one manager for Claude Code, Codex, OpenCode and other coding agents
OpenAI's lightweight coding agent that runs locally in your terminal
Free and open — grab the source, star the repo, and start creating.