Agentless
github.com/openautocoder/agentless- Category
- Developer Tools
- Rank
- No. 1023Tools index
- Pricing
- Open Source
- Platform
- cli
- Type
- TOOL
- Builder
- openautocoder
- GitHub
- 2.1k stars
- Latest release
- v1.5.0
- Date
About
Agentless is a research framework that automatically fixes software issues without using an autonomous agent loop, instead following a fixed three-phase pipeline of fault localization, patch generation, and validation. It achieved 27.3% resolve rate on SWE-bench Lite at roughly $0.34 per issue, outperforming several agent-based baselines, and later reached 40.7-50.8% when paired with Claude 3.5 Sonnet.
What it can do
Automatically fix software issues via a fixed localization-repair-validation pipeline
Software issue/bug report and codebase → Patch that resolves the issue
Localize the fault in a codebase related to a reported issue
Codebase and issue description → Identified faulty code location
Generate a candidate patch for an identified fault
Localized fault information → Generated code patch
Validate generated patches
Candidate patch → Validation result (pass/fail)
Why it made the leaderboard
Shows that a three-stage pipeline — fault localization, patch sampling, then regression/reproduction-test re-ranking — can rival multi-step agent frameworks on SWE-bench for cents per issue. Useful as a cheap, debuggable baseline before reaching for a full agent loop.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.