Foundry Local

Foundry Local is Microsoft's runtime and set of software development kits for running AI models on the device itself rather than calling a cloud service. An application installs one package, selects a model by alias, downloads and caches it, loads it and then calls chat or audio clients from inside its own process. No Azure subscription is required, and because the runtime and the cached models sit on the machine, user workflows keep running without a network connection. It is built for developers shipping local AI inside desktop and client applications where latency, cost or privacy rule out a hosted endpoint. Native kits are published for Python, JavaScript, C# and Rust, so a prototype can start in a scripting language and ship in a compiled one. The kit chooses and registers execution providers automatically, letting an application target a neural processing unit, a graphics processor or the central processor without writing device specific plumbing. A command line tool and an optional local server that speaks the OpenAI API format are available for development workflows and for frameworks such as LangChain, though the vendor positions in process calls through the kits as the primary path. Prompts, audio and responses stay on the user's device. There is no pricing page and no paid plan. The project is published on GitHub under the MIT license with samples and releases, and the reference documentation sits on Microsoft Learn. Compute cost falls on whatever hardware the application runs on. The command line releases are labeled as preview, so interfaces and model catalog entries can still change between versions.