These seven samples use the published foundry-local-sdk 2.2.1 package. Except for
web-server, inference runs in-process without an HTTP service.
| Sample | What it demonstrates |
|---|---|
hello-world |
Manager, execution providers, catalog/model lifecycle, and native chat |
realtime-transcription |
Live microphone PCM input and native streaming transcription |
embeddings |
Native batched embeddings with EmbeddingsSession |
tool-calling |
Native chat tool definitions, calls, and results |
web-server |
Built-in server and its OpenAI-compatible endpoint |
vision |
Native chat with text and a local image |
byom |
Registering local model assets and running native chat |
Important
Every sample pins foundry-local-sdk from PyPI to 2.2.1.
Installation supplies the native runtime; no local SDK source build is required. See the
get-started documentation for versioned setup
guidance and additional examples.
- Python 3.11 or later.
- A supported Foundry Local platform.
- Microphone access for
realtime-transcription.
Each directory pins the released SDK in requirements.txt:
cd samples/python/hello-world
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
python -m pip install -r requirements.txt
python src/app.pyUse the same commands from any of the other sample directories. The first run may download execution providers and a model. Subsequent runs reuse the local cache.
All samples set the SDK application name to foundry_local_samples.