Back to browse
GitHub Repository
5 starsGo

Desktop Automation with Codex

by nicbarth·Mar 6, 2026·1 point·0 comments

AI Analysis

MidBold Bet

LLM-driven desktop automation from screenshots, but unreliable and dangerous.

Strengths
  • Cross-platform (Windows, macOS, Linux) with pluggable LLM runners (Claude, Codex, Ollama).
  • Task markdown format makes workflows human-readable and version-controllable.
  • Global hotkey kill switch prevents runaway bots from destroying user data.
Weaknesses
  • Screenshot-to-action loop is fragile; author admits 'mostly works' and risks accidental actions.
  • No safeguards against destructive commands; high liability for unattended runs.
Category
Target Audience

Power users and automation engineers testing LLM-driven RPA on personal workflows

Similar To

UiPath · Blue Prism · Anthropic Computers.json research

Post Description

Tried using doing some desktop automation by sending codex screenshots and stepping through generated instructions. It's rough, but it (mostly) works. In the screenshot it accidentally presses 71336 haha.

Similar Projects

AI-Powered Web Automation APIs (Screenshot, Scrape, SEO, PDF)

Packages common web automation tasks — screenshots, scrapes, SEO checks and PDFs — into APIs, which is convenient but very crowded territory. The live share is broken (the page shows 'zrok share ... not found'), so you can't test reliability or AI value‑adds; unless it provides robust semantic SEO insights, evasion/anti-bot handling, or superior extraction accuracy, it's another Puppeteer/Playwright wrapper.

Ship ItNiche Gem
openclaw_ai
204mo ago
AI/MLMid

An Agentic Supercomputer

They've built a focused UI for launching goal-driven agent swarms and advertise three real pain points: integrations, stable decomposition, and long-running persistence — all the right battles to fight. The promise of spawning thousands of parallel agents and a harness that can persist multi-week runs is ambitious and useful if it actually works, but the landing page and sparse details leave key questions unanswered (cost controls, safety/guardrails, reproducibility, and evaluation metrics).

Bold BetShip It
andyprevalsky
203mo ago