Self-hosted, sandboxed AI agent autonomously develops and deploys applications

Autonomous Development Environment

A developer successfully implemented an "almost fully self-hosted, sandboxed, agentic software factory." This system allows an AI agent to receive a single prompt and then autonomously manage the entire software development lifecycle. This includes researching technology stacks, planning and writing code and tests, committing to Git, running CI pipelines, and deploying the application to a production server with databases, observability, and SSL.

Motivation and Challenge

The project was motivated by a desire to use AI for development tasks without granting it root access to a primary machine. The core challenge was to create a remote agentic development environment that structurally contains the LLM, rather than relying on trust. This setup aims to provide guardrails for AI agents during development.

Hardware and Stack

The system runs on dedicated hardware, specifically a 2021 10th-gen i7 machine with 32GB RAM, separate from the developer's existing homelab server. The core development stack is self-hosted through Coolify, while inference and integrations like Tailscale, Telegram, DNS, and ACME still connect externally. The only ongoing cost specific to this experiment is a £20 Codex subscription.

Sandboxing and Security

A key aspect of the setup is the sandboxed environment. By running the AI on its own dedicated hardware, any potential destructive actions, such as `rm -rf /`, would only affect the isolated machine. This physical isolation serves as the primary guardrail, minimizing the risk to other systems.