Built an AI Document Assistant That Runs Code in a Sandbox

Every AI document tool falls into one of two categories:

Chatbots that can talk about your files but can't download anything

Dumb converters that can merge PDFs but can't think

I built DocMake to bridge both worlds.

What DocMake Does

DocMake is an AI-powered document assistant at docmake.online. You upload a file, describe what you need in plain English, and AI picks the right tool to handle it.

The Cool Part: Sandboxed Code Execution

The AI Data Scientist feature is what I'm most proud of. Upload a spreadsheet and:

AI analyzes the data structure

Writes code to process it

Runs the code in a sandboxed subprocess (not in the main process)

Returns interactive charts + natural language insights

The key architectural decision: DataFrames never live in the main process. Every analysis runs in a forked subprocess with a timeout and memory limit. This prevents a single bad query from crashing the server.

The Analyst Swarm

Inspired by multi-agent systems, the Analyst Swarm assigns multiple AI personas to independently analyze your data. Each persona writes their own code, runs it, and reaches a consensus. It's like having a team of analysts working in parallel.

MCP Integration

DocMake is also an MCP (Model Context Protocol) server. This means AI assistants like Claude, Cursor, and VS Code can call DocMake's tools directly — not just talk about converting files, but actually do it.

Lessons Learned

Subprocess isolation is critical for sandboxed code execution

Free tier AI models are good enough for most users

MCP is becoming the standard for AI tool integration

Try it at https://docmake.online Free tier, no credit card.