Your customers ask Claude and ChatGPT first. Be the answer.
An agent app puts your product inside the conversation, with your own interface, your data and your rules. One app, on the open MCP Apps standard both assistants support.
If we’re the wrong fit, we’ll say so.
Show up where the question is asked.
Customers use ChatGPT daily, and it shows in how they judge every other app. J.D. Power, on banking apps
- 01 Where the question is asked Your customer asks the assistant first. The answer can be your product, not a link to it.
- 02 Your brand, not a paragraph Your own interface in the conversation: your components, your data, your wording.
- 03 One step instead of five The customer books, orders or checks status right there, without leaving the chat to find your app.
- 04 A new channel you own Built once on an open standard, so the same app serves every assistant that supports it.
From your systems to the conversation, in four steps.
The assistant never reaches your data directly. It goes through a server you control, and the customer approves every action.
- 01 Discover Which questions your customers already take to an assistant, and which of them your product should answer.
- 02 Connect An MCP server in front of your systems and data, with your authentication and the limits of what the assistant may touch.
- 03 Design The interface from your design system, validated before anyone sees it, with an approval step on every action.
- 04 Launch One app, shipped to Claude, ChatGPT and your own product, then measured and extended.
Four kinds of agent app.
For customers or for your own team, in any industry with a question worth answering well.
- Customer support Order status, returns and account questions answered with the real record, in your interface.
- Sales assistant Configure, compare and quote inside the conversation, with the catalog and prices from your systems.
- Internal productivity Your team’s tools inside the assistant they already use. We ship our own that way, in Claude Code.
- Domain experts A specialist that knows your rules, your products and your limits, and shows its answer as a screen.
We build for the assistants we work in every day.
What you can open today, before the first call.
- How we built it Wingspan, in Claude Code Our engineering workflow, shipped as a Claude Code plugin from the Very Good marketplace. Wingspan
- Open source Very Good AI Flutter Plugin Flutter conventions an assistant can follow, published in the open. On GitHub
- Try it Same question, side by side A chatbot answer and a GenUI answer to one question, in the same app. The chatbot demo
Generative interfaces, engineered like production software.
The same team that builds GenUI and ships Flutter apps at scale, with the security review your team will ask for.
Before your app goes into the assistant.
Short answers. Anything else, bring it to the first call.
01 Which assistants can run it?
Claude and ChatGPT both implement the open MCP Apps standard, so one server and one interface serve both. Other hosts are adding support, and the same interface can live in your own app.
02 How is this different from a chatbot on our site?
It lives where your customer already asks, inside the assistant, and it answers with your interface instead of a paragraph.
03 Is it secure?
The interface runs sandboxed inside the assistant, every message between them can be audited, and the customer approves actions before they run. Your data stays behind your own server. VGV is SOC 2 Type II.
04 Do we need an app already?
No. You need the systems and data the agent app should reach. If you also run an app, the same interface can appear there.
05 How long does it take?
It depends on how many systems it reaches and what it may do in them. After the first call you get a written summary with an estimate.
06 Who owns what you build?
You do. The server, the interface and the IP are yours.
We’ll care about your success as much as you do.
Our unicorns read every message and reply, usually the same day.
Start the conversationA 30-minute call about what you’re building.