hey Product Hunt — I'm Malith, the solo dev behind Code Bench and BenchLabs. i started this because of a specific frustration: i was paying Anthropic for an API key, and paying Cursor a monthly sub to call that key on my behalf. two bills, one model. it felt wrong. the second thing that bothered me was that i didn't actually know what was happening to my code between my editor and the model. somewhere in there was a SaaS i hadn't thought much about. so i spent three months building what i wished existed. a native Mac app, no backend, no account. you paste your own key — or point it at a local Ollama — and the agent loop runs on your machine and calls your provider directly. the part i'm proudest of is git. i kept running into agents that would edit five files and then hand me a wall of diffs with no way to pick through them. Code Bench shows every edit in a per-file diff panel — accept or reject, file by file — and the git state updates inline. commit and create-PR dialogs are built in. it's MIT licensed and public beta. mac only for now — i'd rather ship one platform well than three half-working ones. a few things i'd love your honest take on today: - does the "stop paying twice" framing land, or is the BYOK story not interesting to this audience? - mac-only as a starting point — fair tradeoff or dealbreaker? - what's the next bench you'd want: Chat Bench, Dev Bench, or Design Bench? find me on X at @mkappworks — i'll be here all day answering everything. → [Code Bench](https://benchlabs.app/code-bench)

Code Bench

Code with AI on your terms. Local-first, BYO model

Code with AI on your terms. Local-first, BYO model