![](data:image/svg+xml,<svg display="block" id="606156192" role="presentation" viewBox="0 0 8 10" xmlns="http://www.w3.org/2000/svg"><path d="M 7.143 10 C 7.143 8.816 6.183 7.857 5 7.857 L 2.143 7.857 C 0.959 7.857 0 8.816 0 10 L 0 0 C 0 1.183 0.959 2.143 2.143 2.143 L 5 2.143 C 6.183 2.143 7.143 1.183 7.143 0 Z" fill="rgba(238, 226, 227, 0.1)" height="10px" id="Xx4AxfBbq" transform="translate(0.43 0)" width="7.142857142857143px"/></svg>)
![](data:image/svg+xml,<svg display="block" id="1133099305" role="presentation" viewBox="0 0 24 12" xmlns="http://www.w3.org/2000/svg"><path d="M 1 2 L 23 2 C 23.552 2 24 1.552 24 1 C 24 0.448 23.552 0 23 0 L 1 0 C 0.448 0 0 0.448 0 1 C 0 1.552 0.448 2 1 2 Z M 1 11 L 23 11 C 23.552 11 24 10.552 24 10 C 24 9.448 23.552 9 23 9 L 1 9 C 0.448 9 0 9.448 0 10 C 0 10.552 0.448 11 1 11 Z" fill="rgb(249, 243, 242)" height="11px" id="h9hMSmemz" transform="translate(0 0.5)" width="24px"/></svg>)
WhatsApp AI Concierge
AI agent
Reference build
[
Overview
]
A WhatsApp concierge built around the customer.
So staff handle only the conversations that need them.
An ETHIOR reference build, in development on ETHIOR’s own infrastructure.
[
Overview
]
A WhatsApp concierge built around the customer.
So staff handle only the conversations that need them.
An ETHIOR reference build, in development on ETHIOR’s own infrastructure.
[
Overview
]
A WhatsApp concierge built around the customer.
So staff handle only the conversations that need them.
An ETHIOR reference build, in development on ETHIOR’s own infrastructure.
Target first reply (seconds)
Design target for the median time from a customer’s message to the first reply. To be replaced with a measured result.
Target answer accuracy
Design target for answers graded correct against a written rubric on the test set. To be replaced with a measured result.
Planned test conversations
Planned test set, including misspellings, mixed languages, abusive messages and prompt-injection attempts.
Target first reply (seconds)
Design target for the median time from a customer’s message to the first reply. To be replaced with a measured result.
Target answer accuracy
Design target for answers graded correct against a written rubric on the test set. To be replaced with a measured result.
Planned test conversations
Planned test set, including misspellings, mixed languages, abusive messages and prompt-injection attempts.
Target first reply (seconds)
Design target for the median time from a customer’s message to the first reply. To be replaced with a measured result.
Target answer accuracy
Design target for answers graded correct against a written rubric on the test set. To be replaced with a measured result.
Planned test conversations
Planned test set, including misspellings, mixed languages, abusive messages and prompt-injection attempts.
Appointment-based businesses lose hours to repeat WhatsApp messages. This build handles the routine ones and passes the rest to staff with full context.
[
Challenges
]
Prices, opening hours, bookings, changes and complaints arrive all day on WhatsApp. Replies are slow outside working hours, answers vary between staff, and urgent messages get buried among routine ones.
[
Solutions
]
An agent on the official WhatsApp Business Cloud API classifies each message, answers from the business’s own documents, books appointments and escalates complaints to a live inbox, where staff take over with an AI summary.
What the system does
Receives messages through the official WhatsApp Business Cloud API and classifies each one: booking, pricing, FAQ, change or cancellation, complaint, or a request for a person.
Answers from the business’s own documents, retrieved from a vector database, so replies stay grounded in approved information.
Remembers each customer’s history, preferences and open requests across conversations.
Checks availability, creates or moves bookings and sends a confirmation.
Hands over to staff when a message is sensitive, unclear or asks for a person, with an AI summary of the conversation.
Gives staff a live inbox to take over, reply with AI-drafted suggestions and return the conversation to the agent.
Handles opt-in, opt-out and the WhatsApp customer-service window, and masks personal data in logs.
Stack
WhatsApp Business Cloud API · n8n · Supabase (Postgres, pgvector, Realtime, Auth) · Next.js · language models selected per task · Docker on ETHIOR’s own server.
Status and results
Designed, built and tested by ETHIOR. This is a reference build, not a client engagement. It is currently in development. The figures above are design targets, not results. They will be replaced with measured results, with sample sizes and dates, once the system is running.
Limits
Answer quality depends on the knowledge supplied. WhatsApp policies control which messages a business can send and when. The public demo uses a simulated business unless stated otherwise.
Appointment-based businesses lose hours to repeat WhatsApp messages. This build handles the routine ones and passes the rest to staff with full context.
[
Challenges
]
Prices, opening hours, bookings, changes and complaints arrive all day on WhatsApp. Replies are slow outside working hours, answers vary between staff, and urgent messages get buried among routine ones.
[
Solutions
]
An agent on the official WhatsApp Business Cloud API classifies each message, answers from the business’s own documents, books appointments and escalates complaints to a live inbox, where staff take over with an AI summary.
What the system does
Receives messages through the official WhatsApp Business Cloud API and classifies each one: booking, pricing, FAQ, change or cancellation, complaint, or a request for a person.
Answers from the business’s own documents, retrieved from a vector database, so replies stay grounded in approved information.
Remembers each customer’s history, preferences and open requests across conversations.
Checks availability, creates or moves bookings and sends a confirmation.
Hands over to staff when a message is sensitive, unclear or asks for a person, with an AI summary of the conversation.
Gives staff a live inbox to take over, reply with AI-drafted suggestions and return the conversation to the agent.
Handles opt-in, opt-out and the WhatsApp customer-service window, and masks personal data in logs.
Stack
WhatsApp Business Cloud API · n8n · Supabase (Postgres, pgvector, Realtime, Auth) · Next.js · language models selected per task · Docker on ETHIOR’s own server.
Status and results
Designed, built and tested by ETHIOR. This is a reference build, not a client engagement. It is currently in development. The figures above are design targets, not results. They will be replaced with measured results, with sample sizes and dates, once the system is running.
Limits
Answer quality depends on the knowledge supplied. WhatsApp policies control which messages a business can send and when. The public demo uses a simulated business unless stated otherwise.
Appointment-based businesses lose hours to repeat WhatsApp messages. This build handles the routine ones and passes the rest to staff with full context.
[
Challenges
]
Prices, opening hours, bookings, changes and complaints arrive all day on WhatsApp. Replies are slow outside working hours, answers vary between staff, and urgent messages get buried among routine ones.
[
Solutions
]
An agent on the official WhatsApp Business Cloud API classifies each message, answers from the business’s own documents, books appointments and escalates complaints to a live inbox, where staff take over with an AI summary.
What the system does
Receives messages through the official WhatsApp Business Cloud API and classifies each one: booking, pricing, FAQ, change or cancellation, complaint, or a request for a person.
Answers from the business’s own documents, retrieved from a vector database, so replies stay grounded in approved information.
Remembers each customer’s history, preferences and open requests across conversations.
Checks availability, creates or moves bookings and sends a confirmation.
Hands over to staff when a message is sensitive, unclear or asks for a person, with an AI summary of the conversation.
Gives staff a live inbox to take over, reply with AI-drafted suggestions and return the conversation to the agent.
Handles opt-in, opt-out and the WhatsApp customer-service window, and masks personal data in logs.
Stack
WhatsApp Business Cloud API · n8n · Supabase (Postgres, pgvector, Realtime, Auth) · Next.js · language models selected per task · Docker on ETHIOR’s own server.
Status and results
Designed, built and tested by ETHIOR. This is a reference build, not a client engagement. It is currently in development. The figures above are design targets, not results. They will be replaced with measured results, with sample sizes and dates, once the system is running.
Limits
Answer quality depends on the knowledge supplied. WhatsApp policies control which messages a business can send and when. The public demo uses a simulated business unless stated otherwise.
[
Get Started
]
LESS BUSYWORK
AI agents built for the repetitive work across your teams, tools and operations.
![](data:image/svg+xml,<svg display="block" id="204425381" role="presentation" viewBox="0 0 24 10" xmlns="http://www.w3.org/2000/svg"><path d="M 16.791 10 C 17.223 9.088 17.64 8.291 18.043 7.607 C 18.475 6.923 18.892 6.353 19.295 5.897 L 0 5.897 L 0 4.103 L 19.295 4.103 C 18.892 3.618 18.475 3.034 18.043 2.35 C 17.64 1.667 17.223 0.883 16.791 0 L 18.302 0 C 20.115 2.08 22.014 3.618 24 4.615 L 24 5.385 C 22.014 6.353 20.115 7.892 18.302 10 Z" fill="rgb(249, 243, 242)" height="10px" id="rZ5oJoBRr" width="24px"/></svg>)
[
Get Started
]
LESS BUSYWORK
AI agents built for the repetitive work across your teams, tools and operations.
[
Get Started
]
LESS BUSYWORK
AI agents built for the repetitive work across your teams, tools and operations.
LET'S BUILD WORK
THAT FLOWS.
Subscribe Newsletter
Don’t miss out our new updates.
ETHIOR
LET'S BUILD WORK THAT FLOWS.
Subscribe Newsletter
Don’t miss out our new updates.
ETHIOR
LET'S BUILD WORK
THAT FLOWS.
Subscribe Newsletter
Don’t miss out our new updates.
ETHIOR