Foundry Local
Embed on-device model acquisition and inference into an application.
Documentation reviewed
Can I use this for my work?
Foundry Local embeds model acquisition and inference in an application. Its SDK is generally available; the optional CLI is separately labeled public preview in the current repository.
- Availability
- Mixed feature stages. Core SDK GA; optional CLI public preview. Azure Local or other related product stages must be checked separately.
- How to use it
- Library / CLI; API. Supported Windows, Apple Silicon macOS and Linux configurations; model/hardware compatibility varies.
- Processing and data
- On-device option. The documented inference path runs on-device after model acquisition. Usage telemetry may be sent to Microsoft; inspect the SDK privacy and application configuration.
- Costs and limits
- Local inference without provider token billing; device/model costs
- Outputs
- Application chat responses and supported audio transcription through SDKs or an optional local API server.
Use supported catalog models and hardware. The MIT SDK, Microsoft-licensed CLI and downloaded model terms differ. It is intended for device applications, not a general multi-user inference server.
Compare these facts with another tool →
Publisher documentation was reviewed. ASE has not independently tested this workflow, its outputs, billing, privacy controls or accessibility.
Application developers integrating supported local chat or transcription models on user devices.
You want a ready-made office assistant or a multi-user production inference server.
About Foundry Local
Microsoft Foundry Local is a runtime and SDK for shipping local model inference inside an application. The SDK is generally available and selects supported catalog models for compatible hardware. Current documentation separately labels its optional CLI public preview. SDK code, CLI software and individual models have different license terms. Local inference avoids provider token billing, while device resources and model requirements remain; it is not a general multi-user serving stack.