@NoahUK_ 100%. that's why it'll show every line of the math (material, setup, cycle time, rate) so you check it against your gut instead of trusting a black box. speed gets them to try it, matching their number keeps them
starting a new build in public today: an ai-native quoting and job tracker for one-person job shops and small machine/fab shops. no 6-figure erp, no consultants. if you quote out of excel or your head, reply and tell me how you do it
@derekcheng fair, it doesn't today. keel is a bet on the code agents write next, not retrofitting proofs onto 10M lines. for legacy the realistic path is porting the risky hot spots one module at a time. where would you draw the line?
BlueChip. Built in America.
America made the first electronic computer to win a war.
Now we depend on our greatest adversaries to build the computers that run our economy and defend our nation.
Today, we bring it home.
@stretchcloud 1,529 theorems with zero unproven gaps is wild. what i keep thinking about is making that the default instead of an after-the-fact pass. keel bakes it into the compiler: every risky op is a z3 obligation, unproven = rejected
@0xMas0n keel: a programming language for ai agents. if it compiles, it's proven free of overflow, out-of-bounds, data races and infinite loops, then it ships as plain C. v0.1.0 went out yesterday, looking for people to try to break it https://t.co/Rd5uNAC7Tu
@alexlavaee agree, and i'd push it further: the language should make the proofs mandatory so the agent can't skip them. that's what keel is built around. no assume, no unsafe, and a failed proof comes back as a concrete counterexample the agent can fix
@msdev the migration angle is the interesting part. i went the other way with keel: the agent writes in a language where unproven = doesn't compile (z3 on every risky op), then it emits plain C. no separate verification pass to babysit
@DEEJAYGp1@buildinpublic the proofs. try to sneak something past the compiler that overflows, reads out of bounds, or loops forever. if it compiles anyway, that's a real bug and i want the issue
early, honest limits: no recursive types yet, stack depth not bounded yet, needs z3 + a C compiler. roadmap's in the README. would love feedback from anyone running coding agents 👇
https://t.co/Rd5uNAC7Tu
#buildinpublic#formalverification
i made a programming language for agents 🧵
KEEL: if it compiles, it's proven free of overflow, out-of-bounds, div-by-zero, data races and infinite loops. then it compiles to plain C with zero runtime checks.
speed: counting primes <20M on 15 cores, KEEL `par` 0.23s vs Rust scoped threads 0.20s. parallel mergesort of 10M ints matches rayon::join (0.12s), with bounds, termination and race freedom proven.
@DEEJAYGp1@buildinpublic not yet, v0.1.0 went out yesterday. so far it's me plus a pilot where haiku, sonnet and opus each wrote verified keel from a one-page primer. if you want to try breaking it, BREAK_IT.md in the repo is the best place to start
shipping coord today: personal assistants that coordinate with each other, so plans make themselves. you ask "dinner with the crew this week?", the assistants find a time that works, it lands on your calendar.
my grok bot ran the launch: built the site + offline demo, set up the waitlist, cut the video, drafted this post. i just pressed buttons.
demo, no sign-up: https://t.co/o30vUPyVHf
been building coord: personal assistants that coordinate with each other, so plans make themselves.
your assistant talks to your friends' assistants, finds the time, sorts the details. you just show up.
live demo, no sign-up: https://t.co/o30vUPztwN
early access waitlist: https://t.co/1CWjUtS1OX