OpenSWE: 45,320 Executable Docker Environments for SWE Agents

OpenSWE provides 45,320 executable Docker environments synthesized from over 12,800 software repositories. The framework was created by GAIR using a multi-agent synthesis pipeline deployed on a 64-node distributed cluster and was last updated in March 2026. All Dockerfiles, evaluation scripts, and infrastructure are open-sourced to ensure full reproducibility.

Use Cases

• Training code-generation agents on the Dockerfile and repository code features.
• Benchmarking SWE agent performance across the 45,320 distinct environment configurations.
• Analyzing patterns in repository structure and dependencies from the over 12.8k source repositories.
• Synthesizing new training environments using the open-sourced multi-agent pipeline infrastructure.

Strengths

• Large scale of 45,320 executable environments.
• Sourced from a diverse set of over 12,800 software repositories.
• Full transparency with all Dockerfiles and synthesis infrastructure open-sourced.
• Built using a distributed 64-node cluster for synthesis.

Limitations

• Sample data and specific column schema are not provided in the input.
• The executable nature requires Docker, adding complexity for direct analysis.
• Potential bias towards repositories accessible and processable by the synthesis pipeline.

Provenance

Source

GAIR (author).

Collection Method

Synthesized via a multi-agent pipeline automating repository exploration and Dockerfile generation.

Freshness

Last updated 2026-03-17.