I built this. I’ve been hacking on a small experiment called Gilda, a Chrome extension that explores a different way of using LLMs for coding.
Instead of relying on a single model, it runs several models in parallel, compares the outputs, and then merges them into a final response. It’s slower and more expensive than a single call, and definitely overkill for simple prompts, but in my own experiments it usually produces more careful and complete answers for non-trivial code, especially where edge cases matter.
The idea came from noticing how confident a single model can sound while still missing subtle details. I wanted to see if asking multiple models and reviewing them explicitly could reduce that, even if the whole thing feels a bit clunky at times.
There are no autonomous agents, no loops, and no hidden magic here. It’s just a deterministic pipeline. Nothing happens automatically; the user explicitly decides when to generate, compare, and merge.
It’s BYOK, but there are a few free runs so people can try the workflow without setting up keys.
Happy to answer questions or hear where this feels useful, unnecessary, or just a bad idea.
I built this. I’ve been hacking on a small experiment called Gilda, a Chrome extension that explores a different way of using LLMs for coding.
Instead of relying on a single model, it runs several models in parallel, compares the outputs, and then merges them into a final response. It’s slower and more expensive than a single call, and definitely overkill for simple prompts, but in my own experiments it usually produces more careful and complete answers for non-trivial code, especially where edge cases matter.
The idea came from noticing how confident a single model can sound while still missing subtle details. I wanted to see if asking multiple models and reviewing them explicitly could reduce that, even if the whole thing feels a bit clunky at times.
There are no autonomous agents, no loops, and no hidden magic here. It’s just a deterministic pipeline. Nothing happens automatically; the user explicitly decides when to generate, compare, and merge.
It’s BYOK, but there are a few free runs so people can try the workflow without setting up keys.
Happy to answer questions or hear where this feels useful, unnecessary, or just a bad idea.