Endor is in private beta.

Train the model that makes your decisions.

Endor fine-tunes and serves small decision models. You write the training loop with a simple API and SDK; we run the GPUs and host the result.

project = client.projects.get_or_create("support-router")
run = project.runs.create(base_model="decider-2b")

for batch in data:
    run.forward_backward(batch)
    run.optim_step(learning_rate=1e-4)