Loops let teams turn a repeatable business process into a governed AI workflow. A Loop is the reusable design; each Run is one execution with its own inputs, step history, approvals, outcome evidence, and report trail.
In this section
- Understand the difference between a Loop design and a Run.
- Build manual, AI-assisted, agent, and approval steps into one workflow.
- Start Loops manually or through schedules, webhooks, and email triggers.
- Use safety limits, outcome checks, and reports to keep work auditable.
What's inside
- 01
Loop library
Browse, search, filter, duplicate, import, export, and organize reusable Loops. Folders help keep the library tidy without changing who can run or manage a Loop.
- 02
Loop builder
Create a Loop manually or describe the process in plain language and let AI draft a starting structure. Teams can then review the steps, connect agents, add approvals, and publish only when the design is ready.
- 03
Step types
Agent steps ask a selected raia Agent to complete a bounded task. Loop AI steps summarize, transform, score, or decide from prior results. Approval steps pause the workflow until the required people approve, send back, or stop the Run.
- 04
Deterministic routing
Each step can move to the next step, jump to another step when a condition matches, finish, stop, or fail. Routes are rule-based, so the workflow path is predictable instead of being guessed by a model.
- 05
Triggers
Runs can start manually, on a schedule, from a webhook, or from accepted inbound email. Automatic triggers use the published Loop, while draft tests let owners and admins validate changes before publishing.
- 06
Runs and approvals
The Runs view shows queued, running, waiting, completed, failed, and cancelled work. A Run detail page shows the journey, input, step outputs, approval decisions, external actions, usage, events, and outcome.
- 07
Outcomes and reports
The Outcome tab defines the evidence a successful Run should produce. When a Run finishes, reports can include the objective, trigger, timing, inputs, usage, outcome verdict, step journey, approvals, and actions.
At a glance
| Capability | What it does | Best for |
|---|---|---|
| Agent steps | Send one bounded task to a selected raia Agent and capture a normalized result. | Delegating specialist work |
| Loop AI steps | Summarize, transform, score, or decide using earlier step results. | Keeping multi-step work moving |
| Approval steps | Pause a Run for human approve, send-back, or stop decisions. | Compliance and high-impact actions |
| Schedules | Start a published active Loop on a recurring cadence. | Daily or weekly operational checks |
| Webhooks | Start a Loop from an external system with a signed delivery. | Event-driven automations |
| Email intake | Start a Loop from accepted inbound email and optional attachments. | Shared inbox and request workflows |
| Safety limits | Set maximum steps, token budget, elapsed time, and external-action limits per Run. | Bounded, predictable automation |
| Completion reports | Send a structured summary by email or webhook after a Run finishes. | Audit trails and downstream systems |
Best practice
Publish only after testing the draft with realistic inputs. Treat approval steps and outcome checks as part of the workflow design, not as afterthoughts.