An AI workspace I built with ChatGPT's help, before I started guiding agents.
Syvrona was a chat workspace for thinking and writing with AI: streaming chat with memory, prompt chaining, tagging and search, a choice of model providers, and subscription billing with usage limits. I built it in 2024 and 2025 with ChatGPT as a coding assistant, full stack: Python and Flask, MySQL, Docker, Stripe.
What it was for
Syvrona was the first thing Termondt Tech shipped, and it did two jobs. It was a product. It was also where I learned how to work with these models. Building it was AI assistance the ordinary way: a model answering questions and drafting pieces, me assembling and fixing, no rules file, no sign-off, no tests pinning what must never happen. What I took from that was that the prompt is the artifact worth reviewing, that context has to be written down or it is gone, and that a model's output needs a person between it and anything that matters. Those habits became the method behind Argus, built the opposite way: an AI agent wrote all of the code, and I wrote all of the rules it had to obey.
Where it stands
Syvrona is no longer hosted. I ran it as a subscription service with a week's free trial. A handful of people tried it, and nobody stayed to pay, which is the clearest answer a product can give. A general chat workspace was competing with the model makers' own apps, and the thing it did well, disciplined prompting, was a habit rather than a product.
The work moved to automated trading infrastructure and to reviewing AI-written code before it reaches production. What Syvrona taught me went with it.