Use the platform

Select the organization and agent. Settings holds API keys and secrets.

Tab	What you do there
Specs	Inspect tasks and their runtimes. Bring your spec.
Evaluation	Read evaluation scores. Evaluate.
Training	Start a training run on a trainable model or pick a checkpoint to start at. Train.
Deployments	Test and promote a model to serve inference after training. Deploy.

Open a result from the SDK

The SDK returns IDs. Use organization_key to construct links back to the platform.

org_key = client.organizations.retrieve_current()["organization_key"]
print(f"https://platform.trajectory.ai/{org_key}/evaluate/runs/{run.eval_run_id}/trajectories")

The default API uses https://platform.trajectory.ai. If the client sets up a different TRAJECTORY_BASE_URL, use the associated platform host.

Record	Path
Spec version	`/<org-key>/agents/<agent-id>/benchmarks/<bench-id>`
Training run	`/<org-key>/models/<training-run-id>`
Evaluation trajectories	`/<org-key>/evaluate/runs/<eval-run-id>/trajectories`
One evaluation trajectory	`/<org-key>/evaluate/runs/<eval-run-id>/trajectories?tid=<trajectory-id>`

Training trajectories have no page of their own; link the training run.

Get started

Build and improve

Production learning

Reference

Open a result from the SDK

​Open a result from the SDK

Open a result from the SDK