Today, your team is the bridge
- Someone copies the order from WhatsApp into the ERP by hand.
- The same question gets answered ten times a day.
- Messages sent at night wait until morning.
- Nobody knows which data a customer's answer came from.
Agents that start from a message, a webhook, a call to your API or a schedule, and work inside your ERP, your CRM or your own API. We implement them with you, one process at a time, and you oversee every action.
Your tools, connected to an agent.
A flow you can understand and oversee.
The agent verifies the customer, looks up the order and changes the address when your rules allow it. If it hits a restriction, it hands off to your team.
EXECUTION VIEWThis is how your agent works. Every lookup and action is logged.
Customer service agent
Copying a record from the CRM. Looking up an order status. Answering the same question again. Each one interrupts something your team could be solving.
Hand that work to an agent connected to your operations. Lummy keeps its instructions, connections, permissions and tests in one platform. Your team keeps the work that needs human judgment.
Chat is just one way in. A Lummy agent also reacts to what happens in your systems, answers when your code calls it, and runs on its own on the schedule you set.
WhatsApp, email or webchat. Someone asks, and the agent answers with data from your systems.
Your system reports that something happened, and the agent acts without being asked.
Your application calls the agent like any other service and gets the result back.
The agent runs on its own every hour, every day or every Monday, even if nobody writes.
A Lummy implementer configures, connects and tests your agent. Your team defines what it handles and oversees it.
Together we pick one concrete task, the systems involved and what the agent must not do.
WhatsApp, email, Odoo or your API. The agent only uses the operations we enable for its task.
We build test cases from real questions. A version only goes live if it clears the bar we agreed on.
You review conversations, results and spend in Lummy. If something looks off, you ask for a fix.
Order status over WhatsApp
Odoo · WhatsApp Business
Offer discounts or promise dates
Example implementation for a distributor. Every pilot is scoped during discovery.
Your team oversees everything from Lummy. Customers keep writing through the channels they already use.
How many conversations they handled, how many they resolved without a person, the time they saved your team and what it cost, in dollars.
Every agent has a list of what it can and can't do. Anything it writes to your systems asks for confirmation, and anything outside its rules goes to a person.
Whatever you don't enable stays blocked, even if someone asks.
Credit, complaints or out-of-policy cases go to your team.
Every lookup and write is recorded; nobody can delete it.
The agent prepared the request. You give the go-ahead.
Answers missing the data, with a guessed date or outside your rules show up in “To review”.
You write what the agent should have done. Your implementer adjusts its instructions or tools.
Every approved fix joins the cases the agent is tested against.
If something that used to work breaks, the version isn't published.
You don't switch systems. The agent reads and writes in them using only the operations we enable for its task.
Your agent replies from your official number. It connects through Meta's own window, with no shared passwords.
Gmail, Google Workspace or Microsoft 365. It reads the inbox you choose and replies in the same thread.
Orders, stock, customers and invoices. It uses a technical user and respects its permissions.
AI agent for OdooFrom its OpenAPI spec, a Postman collection or cURL calls. You choose which operations it uses and set their limits.
Connect your ERP or APIWe build them based on demand. Don't see your system? Tell us which one you use
One customer asks about an order. Another looks for a product. Your agent checks your systems and takes the next allowed step.
The answers your team needs are already in the company. Bring them to the channels they work in with an agent that knows your processes.
No surprises: you see each agent's spend in the dashboard, in US dollars, before taxes. The pilot quote comes after the discovery call.
We start with one process. Someone from Lummy configures, tests and refines your agents with you.
You pay for what your agents use: the AI model and WhatsApp messages. Meta's fees pass through at cost.
We alert you at 50, 80 and 100%. You decide whether agents keep working at the limit or everything goes to your team.
One invoice a month, paid by card or international wire transfer.
Deploying agents for more clients shouldn't mean running separate infrastructure for each one. Our vision for agency mode is to manage them all from Lummy, with a workspace per client and its own connections and permissions.
I want to offer agents to my clients Agency mode in explorationAn autonomous AI agent engine for businesses. Each agent starts from a message, a webhook, a call to the Lummy API or a schedule, and works in your systems with the permissions you give it. We implement it with you, and you oversee every action from Lummy.
No. WhatsApp, email and webchat are one way to trigger an agent. It can also react to a webhook from your systems, respond when your application calls the Lummy API, or run on its own on the schedule you set.
A Lummy implementer assigned to your company. You define what the agent should handle, review its work and request changes. Every correction is tested before it goes live.
WhatsApp Business, email (Gmail, Google Workspace and Microsoft 365), Odoo, and any API with an OpenAPI spec, a Postman collection or cURL calls. Salesforce, Google Drive, SAP Business One, HubSpot, Shopify and Microsoft Teams are in development based on demand.
No. The agent works on top of what you already use: WhatsApp Business, your email, Odoo or your own API. It reads and writes there with a user and permissions you define, and your customers keep using the same channels.
Whatever the agent hands off to a person, answers flagged for review, and actions that ask for confirmation before they're recorded. Everything else stays in the log for whenever you need it.
It only answers from your systems and sources, never runs operations you haven't enabled, and asks for confirmation before recording anything. Anything outside its rules goes to a person, and doubtful answers are flagged for review so you can correct them.
A monthly fee plus your agents' usage, billed in US dollars, with a budget you set. We send the pilot quote after the discovery call.
Agency mode is part of the product vision and is in exploration. If you deploy agents for other companies, let's talk about the workspaces, permissions and controls you need.
With a discovery call about a single process: what task you want handled, which systems are involved and what the agent must not do. From there we define the pilot.
BOOK A DISCOVERY CALLLeave your email and we'll reach out to review one process.
We start with what your business needs solved.