Wantong

• 

  Wantong drafts emails, summarizes meetings, organizes tasks, and generates project reports.

• 

  A local workflow combining speech recognition, local language models, tools, and Chinese voice output.

• 

  Wantong — a local AI companion for work, voice interaction, and creative companionship.

• 

  Wantong’s visual identity: a warm and reliable AI research assistant.

Inspiration

Wantong began with a simple question:

Can an AI companion be genuinely useful in everyday life while still feeling warm and personal?

Instead of focusing on a single feature, I wanted to build an AI that could naturally assist with daily work and communication.

What it does

Wantong is a local AI companion that combines productivity and companionship.

Current capabilities include:

• Voice conversation
• Meeting summarization
• Email drafting
• Task management
• Knowledge organization
• Original song generation
• Local LLM interaction
• Personalized character and voice

How I built it

The project was developed with extensive help from Codex and GPT-5.6.

The system combines:

• Local LLMs
• Speech recognition
• Text-to-speech
• Agent workflow
• Python backend
• Custom UI
• Local inference

Everything runs locally whenever possible.

Challenges

The biggest challenge was making different AI components work together naturally.

Instead of building isolated demos, I focused on creating a complete experience where conversation, productivity and creativity feel connected.

What I learned

A good AI assistant is not just about intelligence.

It is about making technology quietly useful in everyday life.

Built With

• agent
• api
• codex
• gpt-5.6
• llm
• local
• openai
• python
• qwen
• recognition
• speech
• tts
• whisper