Skip to content

Agentless

Fix software issues in three agent-free stages

Visit official site

What Agentless is

The method is described as an agent-free way to solve software development issues through localisation, repair and patch validation. It narrows a fault to files, classes or functions and then samples candidate patches.

What you can do with it

  • Locate a fault at several levels of detail
  • Sample patches in a simple diff format

Results in the material

  • 82 fixes are reported on SWE-bench lite
  • A 27.3 percent resolve rate is reported
  • The average cost is given as 0.34 USD per issue

Pros & cons

✓ What we like

  • The approach avoids an agent loop
  • Training and patch scripts are open source

! What to watch out for

  • A Python 3.11 environment is set up before use
  • Patches are re-ranked from test results before one is chosen

FAQ

What results are reported?

The material reports 82 fixes on SWE-bench lite and a 27.3 percent resolve rate.

What is the average cost?

The material states an average of 0.34 USD per issue.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review