obaid

Worth noting the "Unix-inspired" framing is the HN title, not the README — the project itself pitches "minimalistic" and "optimized for memory footprint." Curious what the author means by Unix-inspired specifically, since a single-binary TUI running a multi-tool agent loop doesn't immediately read as do-one-thing-well-and-compose.

Working on a few things:

Rundash (rundash.ai) - an easy way to create automated AI agents that can run tasks for you with over 1000+ integrations. Built this from my own needs to run better meetings, discover product insights etc..

Provision (provision.ai) - how I run a team of openclaw agents without burning money on Mac minis. Each agent is given a dedicated email inbox (powered by Mailboxkit) and a dedicated chrome browser that you can connect via browser to unblock if needed. Currently doing some pilots with a few startups. It's interesting to learn how teams want to use AI agents like OpenClaw.

[unnamed project] - a macOS menu-bar AI agent that drives your real apps similar to Perplexity Computer. Hit ⌘⌘, type a task (optionally @-mentioning apps like @slack), and an agent Claude Code or Codex; clicks, types, scrolls, switches apps, and reads the screen via accessibility APIs, with a visible cursor so you can watch it work. Everything runs locally on your own logged-in Mac (BYOK to Anthropic/OpenAI), so there’s no cloud VM or re-auth flow. It also snapshots the frontmost app for ambient context ("summarize this page" just works) and supports parallel tasks with persistent history/workspaces. Pre-release, but the core bet is that desktop agents should feel like a teammate living in your menu bar, not a browser tab or rented remote computer.

I did take break in those hours. Honestly it was tough to survive on 4 hours of sleep during weekdays. I usually slept in a bit on Sundays.

I would also recommend not to do it this way. It was brutal.