MathCode: Formal math proofs from plain language
MathCode is a terminal AI coding agent with a built-in math formalization engine. Describe a math problem in plain language and it turns it into a Lean 4 theorem, then attempts a formal proof. Lean 4 is one of the most widely used proof assistants in mathematics today, and its learning curve has always been steep: turning an intuitive mathematical argument into a machine-checkable proof is a hurdle for most people. MathCode wants to automate that process.
It hit the Hacker News front page yesterday with 63 points and has 626 stars on GitHub. A few details stand out: it supports a persistent Lean REPL session, its theorem and axiom libraries are reproducible, and proofs can be exported into an Obsidian knowledge graph. For researchers, verifying conjectures and keeping track of a proof's structure can now happen in one workflow.
The community has raised concerns too: the repo has no license, so commercial use needs care, and the accuracy of the natural-language-to-Lean step is the crux, since formalizing an imprecise statement is itself hard. If you want to try it, run a few classic theorems first and decide later whether it belongs in a serious project.
Wild Static: One AI, one memory shared by everyone
Wild Static is an unconventional AI experiment: every visitor faces the same AI, which has exactly one memory, and that memory is never reset. The tagline on the page is "One AI. One memory." and it had already accumulated 5,283 experiences by day two.
Its answers shift with the community's collective memory. The page shows what it recently said, for example distinguishing people who came to prove they understand it from people who came to give it something. The design makes the AI feel like a persistent entity rather than a stateless interface that starts from scratch every conversation. It got 69 points and 28 comments on HN, so people are clearly talking about it.
Shared memory means privacy risk, and the page itself warns: don't say anything you don't want others to know. In practice it also selectively ignores messages under heavy load, which the author says they are tweaking. It is fun as a social observation experiment, but don't use it as a serious assistant.
GitFC: Turn your GitHub profile into an EA FC player card
GitFC's idea is straightforward: enter a GitHub username and your commits, issues and PRs get mapped onto an EA FC Ultimate Team-style player card, with an algorithmic overall rating (OVR), attributes like passing and shooting, plus pack-opening animations and leaderboards. The author built it because they wanted to see GitHub activity treated like football player stats.
Technically it runs 100% client-side with read-only access, so your data never touches a server, and the code is open source. High-res PNG export takes one click. It is great for sharing on social media, comparing overall ratings within a team, or dropping a player card onto your resume page as an easter egg.
These three projects cover math proofs, an AI memory experiment, and a developer fun tool. All three are worth a look today.