Free Tool That Runs AI on Your Own Laptop Just Fixed a Login Bug

There's a piece of free, open-source software called llama.cpp that lets you run AI chatbots — the kind that write emails and answer questions — entirely on your own computer. No monthly subscription, no sending your private notes to a giant data center. It's popular with privacy-minded people, tinkerers, and companies that can't ship sensitive information to the cloud. This week it got a small update, labeled b11077, that fixes a login problem.

Here's what went wrong. llama.cpp has a "router" feature that lets one program juggle several different AI models at once. You give it a password once, and it's supposed to handle the door-keeping for everything behind it. But the software was accidentally handing your password file down to each individual model as well, and those models would check it differently — then reject you. The result: people using the straightforward password option got locked out with error messages that said "401," which is tech-speak for "you're not allowed in."

The fix is narrow but real. The update stops the password file from being forwarded to the smaller helper programs, so the router stays the only bouncer at the door. That means every password the router accepts now works all the way through, and — importantly — your password never ends up sitting in the list of settings that other programs or logs might expose. Internal calls between the router and its models also stop failing silently.

The bigger picture: this is housekeeping, not a headline feature. There's no new model, no speed boost, no new capability. But it's the kind of unglamorous maintenance that keeps free, do-it-yourself AI trustworthy enough for the people who rely on it. If you've ever been curious about running AI without the cloud, projects like this are how that stays possible — and why the boring bug fixes matter as much as the flashy ones.