substrate 220240617.1.7

Substrate Python SDK

Substrate Python SDK

Substrate is a powerful SDK for building with AI, with batteries included: language models, image generation, built-in vector storage, sandboxed code execution, and more. To use Substrate, you simply connect tasks, and then run the workflow. With this simple approach, we can create AI systems (from RAG, to agents, to multi-modal generative experiences) by simply describing the computation, with zero additional abstractions.

Substrate is also a workflow execution and inference engine, optimized for running compound AI workloads. Wiring together multiple inference APIs is inherently slow – whether you do it yourself, or use a framework like LangChain. Substrate lets you ditch the framework, write less code, and run compound AI fast.

Documentation

If you're just getting started, head to docs.substrate.run.

For a detailed API reference covering the nodes available on Substrate, see substrate.run/nodes.

Installation

# install from PyPI
pip install substrate

Usage

from substrate import Substrate, ComputeText, sb

Initialize the Substrate client.

substrate = Substrate(api_key=SUBSTRATE_API_KEY)

Generate a story using the ComputeText node.

story = ComputeText(prompt="tell me a story")

Summarize the output of the story node using another ComputeText node. Because story has not yet been run, we use sb.concat to work with its future output.

summary = ComputeText(prompt=sb.concat("summarize this story in one sentence: ", story.future.text))

Run the graph chaining story → summary by passing the terminal node to substrate.run.

response = substrate.run(story, summary)

(To run the graph asynchronously, simply use async_run and await.)

response = await substrate.async_run(story, summary)

Get the output of the summary node by passing it to response.get.

summary_out = response.get(summary)
print(summary_out.text)
# Princess Lily, a kind-hearted young princess, discovers a book of spells and uses it to grant her family and kingdom happiness.

Examples

To run the above example as a notebook, navigate to the examples/notebooks directory and run:

make ensure                         # install dependencies
poetry run marimo edit basic.py     # run the notebook

Many more examples are included in the /examples directory.