Run an AI coding agent on DOS
ClawDOS is an AI coding agent built to run on DOS, per the Show HN headline; no further details are given. ·
Done with a custom-built or unnamed agent
- What they did
- Claude wrote ClawDOS, a 176 KB DOS program that runs on an IBM PC XT (8088, 640K, CGA). It calls the Claude Messages API directly, through a TLS tunnel on a second machine, because the 8088 can't handle TLS. Claude gets 18 tools for files, running programs, screenshots and memory notes. It compiles with Turbo C, and safe mode asks the user to approve changes.
- How it went
- A single prompt produced, compiled and ran a primes program in 64 seconds. Claude also wrote a WarGames-style demo on the machine. The screenshots shown come from the 86Box emulator, with the demo also filmed on real hardware.
- Worth knowing
- You need a Mac or Windows PC with Python 3 and an Anthropic API key for the tunnel and build. Turbo C must come from your own floppy images.
Try it yourself with Claude Code
Help me build a small coding agent that runs on [retro or constrained platform, e.g. DOS, an old Raspberry Pi, a 1990s-era PC emulator]. Start by telling me what is realistic given its limits, then write the code in [language, e.g. C or Turbo Pascal] so it can take a typed request, send it to [AI model API or local model], and show the reply on screen. You are done when I can run it in [emulator or hardware] and get one working response; ask me before installing anything outside this project folder.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Discussion on HN · Nov 21, 2026
Related use cases
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Click through macOS app builds to catch visual bugs
A developer used Claude Code computer use to compile, launch, and click through macOS app screens to find layout bugs.
Route Linear tickets to PRs across multiple repos
A developer sends tasks via Telegram/OpenClaw into Linear, where Cyrus routes each to the right repo and model and opens PRs for review.
Build an AI app over ten days
The author spent about 10 days building an AI-based app with Replit Agent; it worked well but sometimes forgot instructions and reintroduced unwanted behavior.
Build a local knowledge app for agent outputs with Codex
The author built Mianotes using four Codex agents across 30+ sessions to centralize scattered files agents produced, and the agents documented the product inside it.
Fix a mobile web app by voice from a park bench
A developer used phone voice mode to give feedback to a coding agent, which started working up fixes to their mobile webapp while they were outdoors.
Custom agent
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.