You can still use normal debuggers for the code path, but we found it really valuable to isolate and inspect the agent data stream itself: the exact prompts, model outputs, tool inputs/outputs, and how that impacts cost, time, and behavior over long runs. That visibility has been a big lever for improving overall product quality for some of the deeper agentic experiences we are building. Ability to modify and change models has been useful too.
Haha. One of our objectives is to allow for local debuggging but not just pure debugging; Also enabling users to fine tune a version of the model that performs better. We are working on that feature set and involves Lora :) Hence the name. I guess its a future vision ? :)
Hey, I’m one of the builders behind vLLora, so let me clarify the reasoning.
We split the project intentionally: everything embeddable (the Rust crate you ship inside your own product) is released separately under Apache 2.0. So if you’re embedding it, you’re not inheriting license-key / “licensing baggage” concerns in your codebase. (https://crates.io/crates/vllora_llm)
The parts under the fair-code license are the local debugging UI/tooling. Will always be free to use, we just don’t want it copied and resold.
Any paid, advanced observability lives in a separate cloud offering under a different name so there is no confusion whatsover.
We use it to build deeper agentic workflows. it’s been extremely useful for iterating and we want to share this free to use with everyone. Happy to share our experiences if you want to know more.
Re: "no moat, just a few prompts + Claude/Codex". I’ll be a bit cheeky you’re entitled to that view, but we’re in different camps. Some folks vibe code everything; We believe in having proper tools. You still want a screwdriver for screws.