Browser-local tool
Generate starter code for ESMFold2 examples without sending your sequence or token anywhere. Use the output as a scaffold and verify client imports against current official Biohub documentation before running.
Generate starter Python locally in your browser. Do not paste a private Biohub token here.
# !pip install <official Biohub ESMFold2 client package>
# Last verified: update against the current Biohub ESMFold2 docs before running.
# Official sources: https://biohub.ai/models/esmfold2 and https://github.com/Biohub/esm
import os
MODEL_ID = "esmfold2-fast-2026-05" # Verify model ID in official docs.
BIOHUB_TOKEN = os.environ["BIOHUB_TOKEN"]
# Replace these imports/client calls with the current official Biohub example.
from official_esmfold2_client import (
ESMFold2Client,
FoldingConfig,
StructurePredictionInput,
)
client = ESMFold2Client(
model=MODEL_ID,
token=BIOHUB_TOKEN,
)
prediction_input = StructurePredictionInput(
proteins=[{"id": "A", "sequence": "MKTAYIAKQRQISFVKSHFSRQDILDLWIYHTQGYFPDWQNYTPGPGIRYPLTFGWCFKLVPVE"}],
)
result = client.fold_all_atom(
prediction_input,
config=FoldingConfig(
num_loops=1,
num_sampling_steps=8,
include_pae=True,
),
)
with open("esmfold2_result.cif", "w") as handle:
handle.write(result.mmcif)
Choose the prediction type, output mode, and whether the scaffold should request PAE output. The tool then prepares a minimal Python example with a model placeholder, token placeholder, folding configuration, and mmCIF save path.
Use the result as a starting point, not as permanent API documentation. Before running it in Colab or locally, compare the imports and client setup with the current Biohub model page, GitHub repository, and official Colab.
The generator does not send your sequence to a backend, store API tokens, proxy Biohub, or create predicted coordinates. Track A is intentionally limited to browser-local preparation and explanation.
Hosted prediction is a separate Phase 2 product decision. It would require real ESMFold2 inference, sign-in, usage-based credits, job queues, result retention rules, and biosafety screening before launch.
The generated code uses environment variables and placeholders. Do not paste private Biohub tokens into browser tools.
Model IDs, package names, and client classes should be checked against the official model page or GitHub repository.
This free tool does not proxy Biohub APIs or run hosted predictions. Paid hosted prediction is a later compliance-gated phase.
No. It only generates starter code in the browser. It does not call Biohub APIs, run GPUs, or upload sequences.
No. Generated code uses placeholders and environment variables so private tokens stay in your own runtime.
No. Treat it as a scaffold and verify imports, model IDs, and package commands against current Biohub documentation.
No. It only mirrors safe starter patterns and should not be read as support for arbitrary small-molecule docking.
Written by ESMFold2 Tools Editorial Team
Fact-checked against Biohub official docs, GitHub examples, and linked primary sources.
Last updated May 29, 2026