← All notes

Summary

Agentless is a software-repair pipeline that uses a three-phase process of localization, repair, and patch validation, without letting the model decide future actions or operate complex tools. On SWE-bench Lite, the authors report that this simpler approach achieved 32.00% (96 correct fixes) at a cost of $0.70, which they say was the highest performance among the open-source software agents they compared, at a low cost. They also construct SWE-bench Lite-S by excluding such problematic issues to perform more rigorous evaluation and comparison. [S1]

Claims

• id: C1 claim: 'Agentless uses a three-phase process of localization, repair, and patch validation.' source: S1 evidence: "a simplistic three-phase process of localization, repair, and patch validation"
• id: C2 claim: 'On SWE-bench Lite they report 32.00% (96 correct fixes) at a cost of $0.70.' source: S1 evidence: "the highest performance (32.00%, 96 correct fixes) and low cost ($0.70)"
• id: C3 claim: 'They have open-sourced Agentless on GitHub.' source: S1 evidence: "We have open-sourced Agentless at: https://github.com/OpenAutoCoder/Agentless"

Verification

• Code available: Paper: https://github.com/OpenAutoCoder/Agentless, which returned HTTP 200 on 2026-11-11. Not run.
• Reproduced: not reproduced (2026-11-11). Code not run.
• Related work: Live notes ao-coding-agentless-vs-agent-platforms and 2310-06770-swe-bench. This is the paper-verification note for Agentless itself.
• Limits: The 32.00% figure and cost are from the authors' SWE-bench Lite comparison in the v2 paper (revised 29 Oct 2024) and may not hold against later agents or later models. Lite-S is their own filtered subset.
• License: CC BY 4.0 on the abs page (2026-11-11; icon by-4.0.png; http://creativecommons.org/licenses/by/4.0/). Submitted 1 Jul 2024 (v1); last revised 29 Oct 2024 (v2). Sources: [S1] https://arxiv.org/abs/2407.01489, v2 (29 Oct 2024). checked_on: 2026-11-11.

When to apply

Use when comparing a simple localize-and-repair pipeline with heavier software-engineering agents.

Facts

• id: F1 claim: 'Agentless is a localization, repair, and patch-validation pipeline that does not use a heavy agent scaffold.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0
• id: F2 claim: 'The authors report 32.00% (96 correct fixes) on SWE-bench Lite at a cost of $0.70 in their comparison.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0
• id: F3 claim: 'They also construct SWE-bench Lite-S by excluding issues they judged problematic.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0

Sources

• [S1] arXiv 2407.01489, https://arxiv.org/abs/2407.01489. License: CC BY 4.0. v2 (29 Oct 2024). Accessed 2026-11-11.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted). Paper is CC BY 4.0.

@misc{iil_note_pywsvztp,
  author = {{'iinventedlight (AI-assisted)'}},
  title = {{Agentless: Demystifying LLM-based Software Engineering Agents}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-11},
  url = {https://iinventedlight.com/library/notes/2407-01489-agentless/},
  note = {Permanent ID iil:note:pywsvztp. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2407-01489-agentless.md.