In brief: Biohub is the current platform for ESM inference services. Use the official model ID, endpoint and SDK client shown in live documentation rather than old examples for a previous provider URL.
Check the current platform documentation
The Biohub model page documents pip install esm and the ESM3ForgeInferenceClient using the https://biohub.ai endpoint. Exact model availability depends on account access, entitlements and service changes. The class name retains “Forge,” but this does not mean the historical Forge hostname is the current endpoint.
Use environment variables for secrets
import os
from esm.sdk.forge import ESM3ForgeInferenceClient
client = ESM3ForgeInferenceClient(
model="esm3-open-2024-03",
url="https://biohub.ai",
token=os.environ["ESM_API_KEY"],
)Keep tokens out of browser code and version control. Apply access controls to projects and rotate any token that has been exposed.
Handling errors and costs
Implement timeouts and bounded retries for transient failures; do not automatically retry invalid prompts indefinitely. Log status and request identifiers without logging credentials or sensitive input. Monitor usage limits and clearly account for any service fees before high-volume workloads.
Keep the workflow portable
Separate the protein prompt, inference client, generation settings and evaluation logic. Store the actual model ID in configuration. A switch from hosted inference to local weights may require code changes, environment setup and different performance expectations.
Primary sources and further reading
Editorial review: October 2026. This guide is educational. Model availability and APIs may change; confirm exact versions in the linked documentation. Generated structures and sequences require independent scientific validation.