Curie AI

The official Python SDK for Curie — inference infrastructure for scientific AI.

Run biology, chemistry, and physics AI models through one unified API.

Installation

pip install curieai

Quick start

import curie

client = curie.Client(api_key="sk-...")

# Or set CURIE_API_KEY environment variable and omit api_key

# Predict protein structure
result = client.fold("MKTIIALSYIFCLVFA...")
print(f"Confidence: {result.confidence:.1f}%")
result.save_pdb("my_protein.pdb")

# Generate embeddings
embedding = client.embed("MKTIIALSYIFCLVFA...")
vector = embedding.to_numpy()  # numpy array, shape (1280,)

# Design sequences for a backbone
design = client.design(result.pdb, num_sequences=3)
print(design.best.sequence)

Async support

import asyncio
import curie

async def main():
    async with curie.Client(api_key="sk-...") as client:
        result = await client.afold("MKTIIALSYIFCLVFA...")
        print(f"Confidence: {result.confidence:.1f}%")

asyncio.run(main())

Environment variable

export CURIE_API_KEY="sk-..."

import curie
client = curie.Client()  # reads from CURIE_API_KEY automatically

Models

Model	Method	Slug
ESMFold v1	client.fold()	esm/esmfold-v1
ESM-2 650M	client.embed()	meta/esm2-650m
ProteinMPNN	client.design()	bakerlab/proteinmpnn

Docs

Full API reference: curie.sh/dashboard/docs