I got tired of drowning in books and PDFs and not trusting AI summaries. So I built my own solution. It turns your books and documents into a cited, interactive wiki that stays on your computer. Every claim links back to the exact source. A second AI checks the answers. You own the files. Free version, no email signup: Brainary.app If you’ve felt that same overload, try it and tell me what you think.

I got tired of drowning in books and PDFs and not trusting AI summaries. So I built my own solution. It turns your books and documents into a cited, interactive wiki that stays on your computer. Every claim links back to the exact source. A second AI checks the answers. You own the files. Free version, no email signup: Brainary.app If you’ve felt that same overload, try it and tell me what you think.

A lot of the criticism of vibe coding is fair. Too many projects look good in a demo then fall apart in production — weak security, missing edge cases, code nobody understands later. One of the biggest gaps is testing (or the total lack of it). I took both automated and manual testing seriously while building Brainary. Ended up with a testing-to-code ratio of about 1 to less than 3, plus a lot of hands-on checking. I think it turned out solid because of that. Brainary.app It’s free, no sign-up needed. If anyone wants to try it and see for themselves, I’d love the feedback (or just your honest take).

No matter what idea I throw at Claude or GPT, it acts like a cheerleader. “Great concept… strong potential… clear need…” So how do you actually know if it’s a good idea? I’ve stopped asking “is this a good idea?” Now I ask: “Evaluate this idea. List all the pros and cons and justify every point with a source.” Then I go read the sources myself. That’s the process that led to Brainary → Brainary.app Free, no sign-up. Point it at your docs and it builds a local wiki with citations. Would love any feedback if you try it.

Brainary now runs on Mac, Windows, and Linux. “Windows only?” — not anymore. Point it at your documents and it builds a cited, verified wiki. Run it fully local on your own models, or mix in the frontier ones. If you held off because it wasn’t on your OS — it’s there now: Brainary.app

Drowning in documents and wasting hours digging for answers + organizing everything? Brainary turns thousands of docs (PDFs, ebooks, even handwritten notes) into a neatly organized, cited wiki. Every answer is verified by a second AI. No hallucinations. Everything stays local — no cloud upload. Free demo (no signup): brainary.app/demo/ What documents are eating up the most of your time right now?

Shipped local-model support this week.Then discovered the thing no demo ever shows you.Cloud model → answer in seconds. Same question through Llama 3.1 on a 16GB laptop with no GPU → minutes.My 60-second timeout? Every local request failed.Raised it to 10 minutes. Made it fully user-configurable.The uncomfortable truth: Demos show the text appearing. They don’t show users waiting 2–5 minutes while nothing happens.If you’re building on local inference, budget for the wait — and set your timeouts before your users find them.What’s your local setup, and how slow does it actually feel in practice?

Shipped local-model support this week.Then discovered the thing no demo ever shows you.Cloud model → answer in seconds. Same question through Llama 3.1 on a 16GB laptop with no GPU → minutes.My 60-second timeout? Every local request failed.Raised it to 10 minutes. Made it fully user-configurable.The uncomfortable truth: Demos show the text appearing. They don’t show users waiting 2–5 minutes while nothing happens.If you’re building on local inference, budget for the wait — and set your timeouts before your users find them.What’s your local setup, and how slow does it actually feel in practice?

Shipped local-model support this week.Then discovered the thing no demo ever shows you.Cloud model → answer in seconds. Same question through Llama 3.1 on a 16GB laptop with no GPU → minutes.My 60-second timeout? Every local request failed.Raised it to 10 minutes. Made it fully user-configurable.The uncomfortable truth: Demos show the text appearing. They don’t show users waiting 2–5 minutes while nothing happens.If you’re building on local inference, budget for the wait — and set your timeouts before your users find them.What’s your local setup, and how slow does it actually feel in practice?