From VingeGPT to Vinley-1: Why We’re Building the Next Generation of AI Companions for Investors

VingeGPT started from a simple idea and a strong conviction: despite the abundance of financial tools and websites, most investors are still poorly supported in their actual investment journey. Existing solutions are often either too complex, too narrow in scope, or overly focused on raw data without helping investors interpret it nor giving the possibility to interact in natural language with an AI companion. For serious, long-term investors, information alone is not the problem, understanding is.

At the core of that understanding lies one fundamental question: valuation. Assessing whether an asset is cheap or expensive, risky or sound, and therefore assessing valuation risk is one of the most critical disciplines in investing. Yet in our opinion very few tools truly help investors reason through valuation in a structured, repeatable and intelligible way. VingeGPT was created to address exactly that gap, not as another data terminal or stock screener but as an AI companion capable of contextualizing financial information, explaining trade-offs and supporting better judgment.

Launching VingeGPT on the OpenAI GPT Store in May 2024 allowed us to test this conviction quickly. The response from our investing community made of more than 10.000 investors globally confirmed what we suspected: investors were not looking for faster answers but for clearer thinking. They wanted depth, structure and reasoning grounded in fundamentals, not just financial metrics, generic market opinions instead of facts.

Building VingeGPT taught us a lot. It proved there was real demand for a focused AI agent in investing and it helped us learn from user behavior and feedback. Furthermore OpenAI announced in mid-2024 that royalties would be introduced for custom GPT builders and monetization is obviously key for a Fintech startup. That announcement signaled a promising future for creators 

But as adoption grew so did the constraints. While the OpenAI GPT Store is a powerful distribution platform, it is ultimately a hosted environment with limited control. We began to feel those limits in areas that mattered most to us: how the agent reasons, how memory works, how workflows are orchestrated and how the user experience evolves over time. From a monetization strategy as time passed, nothing materialized. 

So overall from a OpenAI store & platform structure itself, it was not aligned with our long-term vision. Building a serious AI product, one that serious investors rely on daily, requires ownership over architecture, data flows, experience and iteration speed. Platform dependency introduces uncertainty, and uncertainty is not something we were willing to build on.

That realization led to a strategic decision: not to move away from OpenAI models but to move away from the OpenAI Store. Vinley-1 was born from that decision. It is a complete re-architecture of the AI agent, designed from the ground up over the last 9 months as a standalone product rather than a hosted extension. At its core,Vinley-1 still uses ChatGPT-5 as its underlying AI subprocessor but everything around it, logic, orchestration, memory, interface and knowledge, is now fully under our control.

We chose to build Vinley-1 on top of Microsoft Copilot Studio, leveraging .NET APIs, PowerPlatform, Azure from an infrastructure perspective and Azure DevOps to have clean & secured development processes. This shift fundamentally changed what we could do. Instead of a monolithic prompt-only based agent with VingeGPT, Vinley-1 is providing us more granular control over how the AI agent operates: how it routes questions, when it calls specific functions, how it combines reasoning with data, when we allow it to search the net. This level of control is essential when building an AI agent that needs to be precise, explainable, predictable and consistent.

One of the most powerful upgrades came from how we handle knowledge. With Copilot Studio, we were able to significantly extend the amount of curated knowledge available to the agent and, more importantly, organize it intelligently. Using topic-oriented file groups, Vinley-1 can now route queries to specific clusters of knowledge depending on the subject being requested by the user. This dramatically improves relevance and accuracy,and it allows us to continuously expand and refine the agent’s expertise without diluting its reasoning.

Moving away from the OpenAI Store also unlocked a much better user experience. Vinley-1 features a fully customized interface built with React, designed around how investors actually work. Users can now maintain a permanent portfolio and watchlist, giving the AI agent contextual memory over time. On top of that, the standalone architecture allows for far more API calls and advanced functions, enabling deeper analysis and richer workflows. Furthermore similar to famous apps like Gamma, Canva, we have integrated the AI agent into the user interface, allowing users by simply clicking on a specific UI interface ‘Ask Vinley’ and throwing a prompt at the AI agent. We believe this is a new user experience that has not been seen until now but let’s remain humble and time will tell us if users really adore Vinley.

Another major advantage of building Vinley-1 on Microsoft Copilot Studio is visibility and control. We now have significantly better insights into telemetry, including prompt usage, model responses, user analytics, and direct feedback loops. This allows us to understand not just what users ask but how the AI responds, where it performs well and where it needs improvement. These insights are critical for continuously refining the agent and ensuring it behaves consistently across similar use cases.

As an AI companion for investors, repeatability and predictability are non-negotiable. Financial analysis is not about being fast for the sake of speed, it is about being accurate, consistent and trustworthy. From the very beginning with VingeGPT and now carried forward even more rigorously in Vinley-1, we have been intentionally paranoid about curated data, predictability and accuracy. Reducing AI hallucinations has always been a core design principle. While still being in our early stages as a company, since February 2025 we started publishing a quarterly benchmark report, benchmarking us against other large-language models and specialised finance AI models. 

To achieve this, we systematically move all known calculations out of the AI engine and into our own backend systems. These computations are performed deterministically and returned to the AI as API responses, rather than being generated by the model itself. This approach ensures that numbers are not “imagined”, formulas are applied consistently and results are predictable and can be reproduced. A specialised AI’s role like VingeGPT or Vinley-1 is to interpret, explain and reason about the data, not to invent it.

This philosophy is central to how we think about trust. Vinley-1 is not designed to impress with clever language but to support sound decision-making. By combining curated data, controlled computation, and transparent reasoning, we aim to deliver analysis that users can rely on over time.

To be clear, this transition is not a rejection of OpenAI’s technology, quite the opposite. ChatGPT-5 remains a core component of Vinley-1. What has changed is the architecture around it. By combining world-class AI models with full control over product design, data integrity and intelligence orchestration, we can deliver a level of quality, reliability and depth that simply was not possible before.

For the VingeGPT community, this move represents continuity and progress. The mission remains the same but the foundation is now much stronger. Vinley-1 is the next step in building a serious AI platform for investors, one designed for long-term use, continuous improvement and real personalisation & decision support.

and this is only the beginning as we still have a lot on our roadmap..

With the deepest gratitude, Candi 

on behalf of the whole extended Vinley team together with our partners that helped us launch Vinley-1

Adriano, Stephanie, Kevin, Enrik, Maw, Deni, Mireia, Nicolas, Riya, Samridh, Ammar, Muhammad, Muzammil, Kevin G, Sven, Arvid

Thanking as well our global ambassadors and beta testers that provided very valuable insights