Skip to content
EN
English 简体中文 soon 日本語 soon

pre.dev

Long-running coding agent

Visit official site

What pre.dev is

pre.dev is presented as a long horizon coding agent. It plans before it codes, keeps state in persistent memory across sessions, and runs autonomously for hours, aiming to produce verified pull requests rather than suggestions.

It is described as model-agnostic, with a CLI that plans, builds and verifies from the terminal and a web option for building and publishing.

What you can do with it

  • Describe a build once and let the agent work
  • Run tasks that span many steps
  • Keep context between sessions
  • Work from the terminal or browser

Who it is for

  • Startup founders and engineering teams
  • Developers delegating long tasks
  • Teams wanting verified PR output

What to watch out for

  • Cost and performance claims such as matching larger models with smaller ones are vendor statements; test on your own repository
  • Autonomous runs need review before merging; verified means the agent's own checks passed, not yours
  • Repository access and credentials must be scoped
  • Confirm current plan limits and credit terms

Pros & cons

✓ What we like

  • Designed for long autonomous runs
  • Persistent session memory
  • Model choice retained

! What to watch out for

  • Vendor performance claims to verify
  • Output still needs review
  • Access scoping required

FAQ

What does it do?

Plans, builds and verifies software tasks autonomously over long sessions.

Is it tied to one model?

It is described as model-agnostic.

How do I run it?

Through a terminal CLI or a web interface for building and publishing.

Last reviewed: 2026-09-17

More AI coding tools tools

View all →

How we review