Run

New automation

Describe what you want and let the fleet draft it, or build the task graph by hand. Each task's output flows into the tasks that consume it.

minute hour day month weekday — empty for manual runs
IANA name — the schedule fires on this wall clock
A failed run revises itself through the drafting loop; the fix is applied only if it passes its test run
Add step

Add MCP tool

Register a tool server so automations can call it.

The server's streamable-HTTP endpoint URL.
Extra HTTP headers as JSON — reference the key as {{secret.NAME}} and store the value under Secrets, never paste it here.

Add secret

Stored server-side and never shown again. Tasks reference it by name.

Letters, digits and underscores. Referenced in a task config as {{secret.NAME}}.
Sent only to the source that needs it, never to a model.

Add inference endpoint

Point the fleet at an open-weights model on infrastructure you trust.

The exact model id your provider's API expects
An OpenAI-compatible chat completions endpoint
Stored server-side, never shown again

Add managed endpoint

Offered to every account on the Managed plan. Calls use the platform inference key.