Coding agents · Project

SWE-agent

SWE-agent is a configurable research harness that uses language models to produce patches for software issues.

+ Follow

Builder facts

  • Configuration. YAML controls which agent and tools run. Source ↗
  • Local runtime. The project documents installation from source and Docker for local execution. Source ↗
  • Model access. Hosted models use provider credentials. The harness can also connect to local OpenAI-compatible endpoints. Source ↗
  • Batch evaluation. Batch mode runs multiple SWE-bench issues and saves generated patches as predictions. Source ↗
  • Usage reporting. The built-in cost calculator excludes custom models absent from the LiteLLM registry. Source ↗

Sources checked

Publisher update ·

SWE-agent 1.1.0 ↗

Adds multilingual and multimodal SWE-bench support and replaces the messages field with query in trajectory files.

Change timeline

0 stories

No verified releases in this timeline yet.