What VIA does
VIA takes routine work away from a business so people can do their actual jobs without being interrupted by tasks a digital employee can handle.
A customer can call, write on WhatsApp or Instagram, or send a photo. VIA answers in the same channel, gets the missing details, checks the connected business systems, and completes the next step. That can be a booking, a service request, a follow-up, or another action the owner can immediately use.
This is not a voice bot that only talks. VIA can use the tools of the business and carry the conversation through to a real result.
Where it started
Before Build Week, VIA was already working in a real salon in Valencia. It handled calls and messages, knew the schedule, created appointments, and worked with the salon's existing systems.
A salon does not have to live inside one calendar. VIA can connect Booksy, Google Calendar, and other calendars at the same time, combine their availability, and use the information already stored there. During onboarding it can import employees, services, prices, and working hours, then build the agent's business context and system instructions automatically.
In about five minutes, the owner can connect the business data, get a working workspace, connect WhatsApp, Instagram, and voice, and receive a phone number for the agent. This is not a future idea. It is the approach already used by the beauty product.
During Build Week, I changed the concept: the same system should not be limited to salons.
Each new vertical can bring its own calendars, portals, CRMs, rules, and workflows into the same VIA core. The owner should not have to understand the technical side. They connect the systems they already use, and VIA turns them into one working agent.
What I built during Build Week
I moved VIA from one beauty vertical to a platform for different service businesses.
For the three-minute demo, I used a compact Technical Services flow. A customer sends a WhatsApp message and a photo of a broken cabinet hinge. VIA understands what is visible, asks for the address, urgency, access, and timing, then creates a ready service request for the owner. It does not invent a diagnosis or a price.
The repair example was chosen because its input and result are visible in seconds. It is a proof of portability, not VIA's target market.
The harder proof already exists in the salon. VIA has to keep multiple employees, services, prices, calendars, reschedules, calls, messages, and external booking systems coherent at the same time. That is much closer to aesthetics, clinics, and other complex appointment-based businesses.
Technical Services proves portability. The salon proves operational complexity.
Beauty & Wellness and Technical Services now use the same core. The prompts, onboarding, Portal, business objects, and owner actions change with the vertical instead of becoming separate products.
I did not pretend to finish every vertical in one week. The other business families are marked coming soon. What changed is that there is no longer a technical reason to rebuild VIA from scratch for each of them.
The final Build Week flow is:
onboarding -> voice, WhatsApp, or Instagram -> business data and customer context -> GPT-5.6 -> completed work -> owner action
What GPT-5.6 and Codex changed
The hardest part used to be making the agent behave coherently for a long conversation. It had to remember the business rules, use tools, collect the right information, and finish the task without being trapped inside a huge set of rigid rails.
GPT-5.5 already improved this. GPT-5.6 made the behavior stable enough that I could remove much of that scaffolding and trust the agent to continue until it reached the result. It also made image understanding practical inside the same workflow.
That changed the scale of the product. Once the agent could reliably handle the behavior, a new vertical stopped being a long prompt-engineering project. It became a question of connecting the business systems and defining what result the business needs.
I used separate Codex agents for the backend and Portal. They worked in parallel and exchanged tasks and evidence directly. Codex helped change the existing system across the backend, Portal, integrations, tests, and staging while the production salon continued to work.
How it works
VIA uses Python, FastAPI, PostgreSQL, and SQLAlchemy on Railway. LiveKit Agents and the OpenAI Realtime API handle low-latency voice. Telnyx provides SIP telephony and phone numbers. WhatsApp runs through a Baileys linked-device service. Instagram Direct is another connected messaging channel. The owner Portal is Next.js, React, and TypeScript on Vercel.
All channels and integrations lead to the same path:
conversation -> business context -> action -> result -> human control
What I learned
The biggest thing I learned recently is that technically we can now build almost anything. The main obstacle is no longer the code. The harder question is what to build when almost every idea has become possible.
That is what happened to VIA. Instead of spending months going deeper into one narrow vertical, I could take a working salon product and expand the concept across service businesses.
The next stage is less about fighting technical limitations and more about distribution: working with real owners, connecting the systems they already use, and finding the routines worth removing.
What's next
I will add verticals together with real businesses, not from a list of imagined features. Aesthetics and clinics are obvious next candidates because they combine staff, services, prices, intake, permissions, several calendars, reminders, and follow-up. VIA can handle the operational work while medical judgment stays with people. Other industries will bring their own portals, CRMs, and communication channels into the same core.
The goal stays simple: give a business a useful voice and messaging agent in minutes, then let VIA grow into a deeper digital employee as the business needs more.
Built With
- baileys
- clerk
- docker
- fastapi
- gpt-5.6-sol
- gpt-realtime-2.1
- livekit
- livekit-agents
- next.js
- openai-codex
- openai-realtime-api
- postgresql
- python
- railway
- react
- sip
- sqlalchemy
- tailwind-css
- telnyx
- typescript
- vercel
- webrtc
- websockets
Log in or sign up for Devpost to join the conversation.