Loading
pre.dev does two things. 1. The Long Horizon Coding Agent. pre.dev plans before it codes, remembers across sessions, and runs autonomously for hours, in the browser or your terminal. Every task lands as its own feature branch and pull request, verified against acceptance criteria, with a live preview. It works on existing codebases and connects to the tools your team already uses: GitHub, MCP servers and your own API keys. 2. RL coding tasks for frontier AI labs (pre.dev Labs). We build and sell verified RL coding tasks for model post-training. Each task is made from a real commit in a licensed private production codebase and ships as a runnable Harbor environment with a graded verifier, a hidden reference solution and solver trajectories. 1,000+ new tasks a month, from 10,000+ licensed private codebases. Also from pre.dev: Browser Agents (turn any website into structured data), the pre.dev AI Gateway (one API key for hundreds of models), and the Architect API and MCP server. Coding agent: pre.dev RL Tasks: pre.dev/rl-tasks Blog: pre.dev/blog Docs: docs.pre.dev