Hi Product Hunt, Years ago, at my previous company, many customer projects started with the same practical pattern: choose a dependable open-source publishing core, add a theme and proven extensions for the requirement, build our customer-specific layer, and hand over a system the customer could own. That ecosystem model stayed in my mind long after those projects. When AI agents became practical, I kept looking for the equivalent. There were strong model APIs and useful workflow tools, but the agent, its connectors, its execution record, and its extension ecosystem often lived in separate places. It was difficult for a customer to own the whole system and answer a basic question: “Is this exactly what ran?” That is what inspired Krelvan. The project began as reusable, self-hosted building blocks for agents. The turning point was deciding that the signed event ledger had to be the runtime itself, not an audit feature added afterward. Every screen - run output, graph, timeline, history, and memory- now reads from the same append-only record. From there, the product grew approval gates around six kinds of side effects, failure diagnosis and retry, support for multiple hosted and local models, and a public Git-backed registry that anyone can inspect, fork, or contribute to. This is version 0.1.2, a public beta. I would especially value feedback on the boundary between the runtime and the extension ecosystem: what would you want to install from an agent registry, and what would you insist on auditing first? — Sreenath

Krelvan - own your agents

Write a sentence. Get a working agent system you own

Write a sentence. Get a working agent system you own