Mo - Your Whatsapp Personal Assistant-

• ## Inspiration

I wanted a personal assistant I could reach as easily as a friend: through WhatsApp. Someone who could help with research, remember important details, prepare responses, and eventually communicate with others on my behalf—with my approval.

That idea became Mo: an AI assistant that turns everyday messages into useful work.

## What it does

You send Mo a WhatsApp message to ask a question, research a topic, create a report, or get help with a task.

Mo can produce deliverables such as competitor analyses, documents, and presentations. It can save explicitly requested facts and preferences, then retrieve them in future conversations.

We also built automatic artifact memory so generated work can be preserved in Cognee with its full stored content, sources, and references. Each enabled workspace has its own dataset.

## How we built it

We combined four main tools:

• Claude Code helped us develop and iterate on the application.
• Hermes powers the agent runtime and tool execution.
• Bright Data provides access to web data for research.
• Cognee stores and retrieves persistent memories and generated artifact content.

We connected these components through a backend, integrated WhatsApp as the conversational interface, and deployed the application to Azure Container Apps.

## Challenges we ran into

One challenge was making memory reliable. An assistant saying “I’ll remember” does not guarantee that the complete work was saved. Our first test preserved a short competitor summary but missed the full report, which led us to build automatic artifact synchronization. preserve agent state during deployments. Response speed remains an area we are actively improving.

## Accomplishments that we're proud of

We built a working WhatsApp interface for an assistant that can research and produce useful deliverables. We connected persistent memory, added automatic artifact synchronization with retry handling, and deployed the system to staging.

We also verified memory retrieval and tested workspace isolation, duplicate prevention, and recovery after interrupted processing.

## What we learned

A useful assistant needs more than a capable model. It needs dependable tools, persistent context, clear permissions, and honest confirmation of what it actually completed.

We learned to verify saved information through retrieval and to keep background memory processing separate from the conversation.

## What's next for WhatsApp Personal Assistant

We want to make responses faster, improve recall of previous work, and add more everyday integrations.

The next major step is helping users manage conversations with others: drafting replies in their preferred style and sending approved messages on their behalf.

Built With

• brightdata
• cognee