brooksf
bforsyth1234.github.io/bot-in…
- Joined
- Nov 20, 2026
Comments
[Weekly] What’s your unpopular opinion about AI?
I think AI will allow users to create one super app that is custom built for that specific user. No two users will have the same software.
[Weekly] What are you building this week?
nice, the fallback is the part most people forget. mine goes the other way: If you ignore a card it just times out and nothing happens. file stays where it was, draft never sends. didn’t feel right having it guess “yes” on someone’s mac
ignored cards do count against whatever triggered them though, so if you keep ignoring a certain kind it backs off and shows up less
how do you pick the default on yours? per task or one rule for everything?
[Weekly] What are you building this week?
It watches triggers like new files in Downloads, incoming mail and iMessages, upcoming meetings, and your active app. When something comes up, it checks with a model to see if there's a helpful action (like filing a PDF invoice that just landed) and surfaces a card you can approve or dismiss.
A few things I focused on:
Approval-first cards: File moves, reminders, and email drafts wait for a click; only read-only actions skip approval.
Local by default with MLX: Runs locally on Apple Silicon (macOS 14+), with optional cloud API keys (OpenAI, Anthropic, Gemini, Groq, OpenRouter) and a ledger in Settings showing anything that left the Mac that day.
Self-authored, sandboxed tools: When you repeat a manual workflow, it can propose a small tool—you inspect the full source first, and each tool runs in a short-lived sandbox-exec process backed by its own git repo for easy rollbacks.
It's free and MIT-licensed, and the source code is on GitHub. Still early, so very curious how others handle approval UX and local model ergonomics on long-running background helpers.https://bforsyth1234.github.io/bot-inator/