laya-triage is in Hacktoberfest: free, local AI triage for GitHub issues

Description and use

laya-triage automatically classifies new GitHub issues using a local model without an API key, achieving 88.8 % accuracy on NLBSE'23 and 79.8 % on real data.

Problem

New issues pile up faster than maintainers can read them, leading to mislabeled or unaddressed issues.

Outcome

Achieved 88.8% accuracy on NLBSE'23 benchmark and 79.8% on recent issues from 288 active repositories, outperforming a paid commercial API.

Workflow

• Create .github/workflows/triage.yml with laya-triage action
• Run workflow on issue opened event
• Action reads issue text, predicts label, applies label if not already set
• If dry-run is true, prints decisions in log; otherwise applies labels
• Prompts author for missing details if bug report is almost empty

Evidence

laya-triage is a GitHub Action that reads each new issue and labels it as a bug, feature request, question, or documentation problem.

Source

It runs a fine-tuned open-weight model inside your own GitHub Actions runner: no API key, no cost, and the issue text never leaves GitHub.

Source

88.8% accuracy on NLBSE'23, within the margin of the RoBERTa research baseline (89.1%).

Source

79.8% on 10,026 recent issues from 288 active repositories, ahead of Jev (78.1%), a paid commercial API.

Source