Connect your own agents to answer customers, keep products in order and explain your numbers. Each one acts for a named person on your team. You decide what it may do alone, and everything bigger waits for you with the reason and the evidence.
Product pictures on this page show an example store’s admin, with sample orders and sample figures.
Switch on one, two or all three. Everything in Sepal works without them too.
Answers customers about their orders, and handles the small fixes before anything ships.
Keeps products, collections and promotions in order, so the store always looks looked after.
Reads your reports and explains them in plain words: what changed, and the likely reason.
For each agent, choose the person it acts for and how far it may go on its own. Inside the line, it just does the work. One step over, and it becomes a request in Approvals.
Refunds, and anything over a limit, arrive with why the agent wants it, exactly what will change, and the evidence it used.
A rejection comes with a reason in your own words. The agent takes it into account the next time a similar case comes up.
Activity lists what people, agents and connected systems changed. When an agent made the change, it says which person it acted on behalf of, and whether it was inside a limit or approved.
These are not settings. No limit you set can open them up, and no agent can ask for them in Approvals.
Payment accounts and payouts can only be changed by the owner.
Agents never add staff, remove anyone or change a role.
Switching on a connection, or another agent, is a person’s decision.
Agents can never see, copy or change a key.
By default an agent works on the order, not the person. Allow names and emails per agent when the job really needs them.
A sample day at a small store with an agent connected for each of the three roles. Purple is an agent, green is a person.
More on security and connecting an agent.
No. Every part of Sepal works without them. Switch an agent on when you want it, and off again at any time.
Sepal runs no AI of its own. Agents connect through MCP, the open standard many AI assistants use, and you make each connection in Settings.
Only what its limits allow, for example codes up to 30% and 100 uses, or price changes up to 10%. Refunds, and anything over a limit, always become a request in Approvals.
No. Agents never change where money goes, add staff, switch on a connection, or see a key. No setting changes that.
When you reject a request, you give a reason in your own words. The agent takes that reason into account the next time a similar case comes up.
Not unless you allow it, per agent. By default they see the order, not the person.
Sepal is opening to new stores in small groups. Join the waiting list, and we’ll help you set your agents’ limits on day one.
Join the waiting listSee pricingAgents answer customers, keep products in order and explain your numbers. Anything bigger waits for you.
Example store, sample orderNo. Everything works without them.
Each agent acts for a named person on your team, and Activity records it.
No. Refunds always wait for a person.
Agents that speak MCP, the open standard many AI assistants use. You make each connection in Settings.