Lelemon ES

An AI copilot for SaaS, inside your product

Your users ask for the work in their own words and the agent does it in your system. You decide what it can do and what needs approval.

From US$700 a month · 1,000 conversations

Used by Haulmer and TUU · Plataforma Contable

LAgent · inside your app

User · inside your app

Build tomorrow’s route with the 12 deliveries in Providencia.

find_deliveries12 deliveries for tomorrow
create_routeDraft route · 12 stops
Approval requiredpublish_route · 12 stopsApproved by the operations lead

Tomorrow’s route is published.

Illustrative example

An AI copilot for SaaS that carries out tasks

An AI copilot for SaaS is an in-app AI agent. The user types what they need, for example move tomorrow’s shipments to Thursday, and the agent runs it with the actions your system already has.

Your users stop hunting for the right screen. The interface stays for anyone who prefers it, and the copilot becomes a second way to use the product.

How you integrate it: API and MCP

Your backend talks to the agent through the Lelemon API: you open one conversation per user and send their messages.

Your product’s actions reach the agent through your API, your MCP server or signed webhooks. No data migration. API keys are per organization with limited permissions, and everything the panel does can also be done in code.

Register actions and grant them one by one

It works like this. You register an action with its name, description and parameters. Then you grant it to a specific agent. Other agents in your organization do not see it.

A sensitive action, such as deleting data or changing a plan, can be set to wait for approval from the person you choose before it runs. Every action lands in the case history with the data used and who approved it.

  • RegisterDefine the action and the schema of its parameters.
  • GrantGive it to one agent and no other.
  • ApproveSensitive steps wait for a person’s yes.
  • ReviewThe history shows what was done, where and with what data.

The same agent also answers on WhatsApp

You do not need a different agent for each channel. The one inside your product also answers on your WhatsApp Business number, with the same actions and the same rules.

Your team sees everything in one panel where each conversation is a case, with teams, routing and an inbox. If the agent hands off, the case arrives with a summary. And when a job takes a while, the agent can keep working in the background and tell you when it is done.

Models and pricing

You choose the model for each agent: Claude Haiku for quick replies and Claude Sonnet for work that needs reasoning, on AWS Bedrock. Each organization sees only its own data, and when an action sends data to your systems, the purpose and destination country are recorded, as Chile’s Law 21.719 requires.

Pricing starts at US$700 a month with 1,000 conversations, and there is a 5,000-conversation plan at US$2,500 a month. Conversations cost the same with any model.

Questions

What is an in-app AI agent?

It is an agent your users reach from inside your software. They ask for the work in their own words and the agent does it by calling the actions you granted it.

Do I have to rebuild my product?

No. The agent connects through your API, your MCP server or signed webhooks. You register the actions that already exist and grant them to the agent.

What happens if it gets something wrong?

It can only use what you granted. Before a sensitive action it waits for approval from the person you choose, and your team can take over the conversation. Everything is logged.

How much does it cost?

From US$700 a month with 1,000 conversations. A conversation is up to 10 exchanges within 24 hours. If you add WhatsApp, Meta bills that channel separately, directly to you.

How do I get started?

You start with a paid 60-day pilot.

Try a copilot with your own actions

Ask the agent how it would work in your product. If it fits, it books you a call to build a case around a real action.