Platform / Agent Factory
Agent Factory
Build automated workflows that run on a schedule, in response to events, or on demand. Define multi-step workflows, add approval checkpoints for sensitive actions, and get notified at every step.
Automation Layer
What is an automation?
An automation is a workflow built on top of a configuration. It defines when to run, what steps to follow, and who to notify -- but it does not define its own permissions. Permissions come from the configuration it references.
Quick Start
Getting started
Creating an automation takes five steps. You can start from scratch or customize a pre-built template.
Decide what starts the automation: a webhook event, a recurring schedule, or a manual run from the dashboard.
Break the workflow into sequential steps. Each step gets its own instructions and optional tool constraints.
Place approval steps between workflow steps or on specific high-impact actions.
Choose who gets alerted on start, pause, completion, and error. Supports Slack, email, and in-app.
The configuration determines the automation's tools, integrations, and data access. No separate permissions needed.
Automations inherit all permissions from their configuration. They do not have their own permission policies. An automation defines when and how to act, while the configuration defines what it can access.
This separation means you can reuse the same permission setup across multiple automations, or swap an automation's capabilities entirely by pointing it to a different configuration. One configuration could power a code review automation during the week and a security audit automation on weekends -- same tools, different workflows.
When to Run
Triggers
Triggers define when an automation starts. Each automation has exactly one trigger type.
Runs when an external event fires. The event data is passed to the automation as context.
Runs on a recurring schedule. Use for tasks that don't need an external event to start.
Triggered by a user action. Useful for on-demand tasks or testing.
Workflow
Workflow steps
Steps define the workflow. Each step has a name, a set of instructions, and optional tool constraints that limit which tools the automation can use during that step. Steps execute in order -- the output of one step feeds into the next.
Breaking work into steps gives you fine-grained control. You can place approval checkpoints between specific steps, restrict tool access per step, and see exactly where a workflow spent its time.
Query monitoring tools, gather logs, identify affected services.
Correlate findings, determine root cause, assess blast radius.
Create incident ticket, post summary to Slack, page on-call if critical.
Human Oversight
Approval steps
Approval steps pause the workflow at specific points and wait for a human to review and approve before the automation continues. This gives you control over high-impact actions without sacrificing automation for everything else.
Alerts
Notifications
Configure who gets notified and when. Notifications keep your team informed about automation activity without requiring them to watch a dashboard.
When the automation begins running
When an approval checkpoint is reached
When the automation finishes successfully
When the automation encounters a failure
Supported channels: Slack / Email / In-app. Each notification event can be routed to different channels and recipients.
Progress Tracking
Workflow progress
An automation runs through its steps, performing actions along the way. Approval checkpoints can pause the workflow for human review. Every run is tracked with full visibility into each step and action.
Event received, workflow queued
Actively executing steps and performing actions
Paused for human approval at a checkpoint
All steps finished successfully
Workflow stopped due to an error
Track progress and results in the Automations dashboard. Each run shows a timeline of steps, actions taken, approval decisions, and duration. Drill into individual actions for full detail.
Visibility
Audit and logging
Automation runs are tracked at two levels. The workflow log captures the full picture. Individual actions are logged separately with a link back to the workflow run.
Per-run detail
Per-action detail
Templates
Pre-built automation templates
Automation templates give you a working starting point. Each template includes suggested steps, triggers, and approval checkpoints. Customize any template to match your workflow -- they are starting points, not constraints.
Investigates alerts, correlates monitoring data, creates tickets, and pages on-call when needed.
Reviews pull requests for bugs, style issues, and security concerns. Posts inline comments and a summary.
Scans repositories for vulnerabilities, checks dependency versions, and tracks compliance posture.
Gathers yesterday's activity across tools and compiles a team digest with blockers and highlights.
Provisions access, creates welcome tickets, and guides new team members through setup checklists.
Checks open tickets against SLA thresholds, escalates breaches, and sends periodic compliance reports.
Examples
Use cases
Real-world workflows that show how triggers, steps, and approval checkpoints work together.
Guidance
Best practices
Tips for getting the most out of automations.
Any step that creates Jira tickets, posts Slack messages, or modifies production systems should have an approval checkpoint. Let the automation gather and analyze freely, but require approval before it acts on the outside world.
Each step should have one concern. "Investigate" and "Respond" are better as two steps than one. Focused steps are easier to debug, faster to approve, and produce clearer logs.
Pre-built templates are designed around common workflows. Start with a template that matches your use case, run it a few times, then customize the steps and instructions based on what you learn.
Check the Automations dashboard to see how the automation interprets your step instructions. If a step consistently takes too long or makes unnecessary actions, refine the instructions to be more specific.
Limit which tools are available in each step. An "Investigate" step that can only read from Datadog and GitHub will be faster and more predictable than one with access to every integration.
Configure error notifications from the start. Knowing immediately when an automation fails is far more valuable than discovering it missed three days of scheduled runs.