Posted patent here a couple days ago and got way more feedback than I expected so thanks for that, even the rough kind. First post got some fair criticism and some not so fair(!), either way this is me acting on the fair part. 0.2.0 has the two things people kept bringing up.
It still runs fully offline by default, ollama and the embeddings are both local, nothing phones home. the verdict just isn't locked to ollama anymore. you can point it at anything that speaks the openAI API now with --api-base and --api-key (or OPENAI_API_KEY), so LM Studio, vLLM, openRouter, wtv you run. and if you don't want an LLM anywhere near it, --fast skips that whole step and just hands you the ranked list.
patent "your idea" --api-base [https://openrouter.ai/api/v1](https://openrouter.ai/api/v1) --api-key sk-... --model qwen/qwen-2.5-7b-instruct
The other thing I fixed, it could pull up a near identical match and still act like nothing was there, the patent crate finding itself and then telling me to go build it was the best example of that. The open/crowded score always came from the embeddings, it was the headline text that could drift off it, so now a strong match floors the verdict and it can't say "open" over something that clearly already ships.
On the AI thing, I understand you lot, yeah it's partly AI assisted and Im not gonna pretend otherwise. It's open and it has tests (that do get approved by me, I try to test what I publish as much/thoroughly) as possible, so... yeah, hit me with more feedback.
The issues and PR s, Im reading all of them. Someone already sent a Nix flake one. After that? prebuilt binaries and an install script so you don't need the whole Rust toolchain just to try it. a couple people on Fedora hit missing build deps (openssl-devel, gcc-c++) so thats getting documented too. more sources after that, AUR and Nixpkgs came up a ton.
[link] [comments]