How it works
An agent connects a trigger (an event that starts it) to a sequence of steps:- A trigger fires: a commerce event from the Colossal app, a webhook from a third-party app, or a schedule
- Steps execute in order, with support for branching, loops, parallel execution, and AI reasoning
- Each step can reference data from the trigger and previous steps using data access variables
Triggers
| Type | Description |
|---|---|
| Colossal triggers | Fire on Colossal commerce events — orders, products, payments, invoices, customers, notifications |
| Storefront triggers | Fire on cart and checkout events |
| App triggers | Fire when an external app sends an event to your project |
| Scheduled triggers | Fire on a cron schedule or at recurring intervals |
Step types
| Step | Description |
|---|---|
| App action | Execute an action from any installed app |
| Prompt | LLM-powered analysis with access to read and write tools, gated by per-tool approval |
| Approval | Pause for a human to approve or deny before continuing |
| Choice | Branch based on conditions |
| Delay | Wait for a duration |
| Parallel | Run branches concurrently |
| Map | Loop over an array |
| Wait for event | Pause until an event arrives from any app |
| Transform | Reshape data into a typed output object |
| End | Stop the agent and record the outcome |
Apps
Apps provide triggers (events that start agents), actions (operations agents can execute), and tools (the LLM-callable form of those actions). The Agent Builder composes them from your installed apps into agents. The built-in Colossal app ships with every project. It provides:- Triggers: 40+ commerce events (
ORDER_CREATED,CHECKOUT_COMPLETED,PAYMENT_FAILED, etc.) plus scheduled triggers (cron and interval). See the Colossal and Storefront triggers references. - Actions: customers, products, orders, inventory, invoices, email, payments, and HTTP operations, plus cart and checkout. See the Colossal and Storefront actions references.
- Tools: a curated set of LLM-callable functions for the AI Reasoning step and the chat agent. See the tools reference.
Third-party apps
Install apps from the Colossal dashboard to connect external services. Each app adds its own actions to the Agent Builder, and every action is automatically available as a tool to the LLM inside an AI Reasoning step or the chat agent.| Category | Apps |
|---|---|
| Payment | Stripe, PayPal |
| Marketing | Klaviyo, Mailchimp, Loops |
| Notifications | Slack |
| Tax | TaxJar |
| Reviews | YOTPO |
Creating agents
Open the Colossal dashboard and describe your automation in the chat:- “When an order is confirmed, send a Slack notification to #orders”
- “Every morning at 9am, check inventory levels and email me a report”
- “When a customer places an order over $500, flag it as high-value and notify the team”
Execution
Each agent run tracks:- Status:
PENDING,RUNNING,COMPLETED,FAILED,CANCELLED,PAUSED - Step-level status:
RUNNING,COMPLETED,FAILED,SKIPPED - Trigger data: the event or schedule that started the run
- Step outputs: results from each completed step