Coding agents worked well for individual engineers, but the surrounding workflow was a bit of a mess. We generally found that many engineers had different MCP connections and context for their agents, personal automations running that other people couldn’t access, and very little introspection for what a human’s input into the coding agent looked like.
So we built an internal system that converted coding agents into shared team infrastructure.
The system runs Codex, Claude Code, OpenCode, and other agent harnesses in gVisor sandboxes. You can mix and match coding agents to tune your cost and intelligence tradeoff (for example, our team has started to use GLM 5.2 for automation tasks and a combination of Codex/Claude Code for manual tasks). It connects to a bunch of tools like GitHub, Linear, Sentry, Slack, PagerDuty and makes previews easy to generate.
We were inspired by internal systems like Stripe Minions and Ramp Inspect, but wanted an open version other teams could inspect, run, and adapt so we made it MIT-licensed and self-hostable.
Hope you like it!
Website: https://143.dev