Helix Automations
Helix turns an outcome you describe into an automation that runs itself — reliably, over minutes or over weeks. You define the goal and the steps; Helix runs them as a pipeline, pausing for approvals, staying inside your budget, and respecting the guardrails you set. Access via Automations → Helix in the sidebar.
What Helix is
A Helix is a goal-driven automation. Instead of wiring up scripts, you describe the result you want — “screen new rental applicants and draft a reply”, “collect rent and chase late payments”, “triage incoming support tickets” — and Helix builds a pipeline of steps to get there.
What sets Helix apart from a one-off script or a single agent run:
| Property | What it means for you |
|---|---|
| Durable | A run can span seconds or weeks. Helix remembers where it is, survives waits, and picks up exactly where it left off — no lost progress. |
| Approvals built in | Insert human approval steps anywhere. The automation pauses, waits for a person to sign off, then continues. |
| Budget-aware | Every automation runs under a spend cap. When the cap is reached, Helix stops rather than surprising you with a bill. |
| Guardrailed | Rules you define can block or warn on risky actions before they happen. |
| Learns over time | Automations improve from past runs, so common cases get faster and more accurate. |
How an automation is shaped
Every Helix has four parts:
- A goal — the outcome you want, in plain language, plus what “done” means so every run can be judged against it. See Building an Automation.
- A stimulus — what starts a run: on demand, on a schedule, or from an incoming webhook. See Triggers.
- A pipeline of steps — the work itself, expressed as steps wired together with dependencies. There are five step kinds — agent, transform, action, wait, and sub-automation — and any step can be conditioned, retried, or gated behind a human approval. See Step Types.
- Controls — the autonomy settings, guardrails, and budget cap that keep the automation safe. See Controls & Policy.
Get started
- Build an automation — use the guided editor or just describe your intent, and set the goal it’s judged against.
- Choose your steps — agent, transform, action, wait, and sub-automation steps, wired together with dependencies, conditions, and human approvals.
- Pick a stimulus — run it on demand, on a schedule, or from a webhook.
- Set your controls — autonomy, guardrails, and a budget cap.
- Test and go live — dry-run first, then promote draft → trial → active.
Prefer to start from something proven? Browse the Templates gallery and adopt a ready-made automation in one click.