novatrope

I built CodeDoc because I was tired of: - Setting up Claude API keys and billing - Getting surprise $300 bills when I forgot rate limits - Managing hosting separately from AI tools - Complex Docker/deployment for every project

So I bundled it: web hosting + Claude Code console + spend limits.

You get: • Actual server resources (RAM/CPU/disk - see plans) • Pre-configured Claude console access • Request-based limits (no surprise bills) • Human supervision of AI changes (safety net)

This entire site was built by Claude - not a single line written by a human. That's the proof of concept.

I'm here to answer questions! What would you build with this?

Try it: https://codedoc.us

Very much like my experience with Claude. First, gave it some simple tasks (Getting this error on my site, etc) The results were surprisingly good. Then I started giving Claude broader tasks, and learning how to write the prompts. Now I have come to the point where I haven't written a line of code in several weeks (quite a change for someone who learned to program on a Burroughs b5500 back in the 70's) So I guess I am a convert.

I have also been thinking about how to make claude more accessible to the less than super-adept web designer. I'd like to know how it is being received?

Keep it up. and keep us i formed here of your progress

At the moment we are concentrating on simply limiting maximum dollars spent. But like your thoughts. Gives me something to work on tomorrow during the holiday (US)

I'm literally living this right now with my service (managed AI coding + hosting). The shift from "buy software" to "request outcomes" is real. My customers don't want to "own Claude API access" - they want to "get code fixed" or "deploy a site." Biggest challenge I've found: pricing. Do you charge per request? Per outcome? Per time? Traditional SaaS pricing ($X/month) doesn't quite fit when AI does the work.

Interesting approach! Quick question about the browser execution:

  How are you handling security/sandboxing for the Python runtime?
  I'm working on a similar AI coding service and we've found that
  isolating code execution is critical.

  Are you using Pyodide or a custom WASM implementation?

  Looks promising - nice work!