SWE-chat: Coding Agent Interactions From Real Users in the Wild
- Type
- dataset
- Venue
- Social And Language Technology Lab (SALT-NLP)
- Year
- 2026
- Source
- huggingface
- Access
- restricted
- Language
- English, multiple programming languages
- Added
- 2026-07-17T20:18:03.808710+00:00
- Verified
- 2026-07-17T20:18:03.808710+00:00
Summary
SWE-chat is a dataset of real-world AI coding sessions from developers using AI coding assistants (Claude Code, Codex, Gemini CLI, and others via the Entire.io CLI). It includes 2.69M conversation turns across 5,851 sessions, 14,459 commits, and 205 repositories, with full conversation transcripts, tool calls, thinking traces, code changes (diffs), and attribution of human vs. agent-authored code. The dataset enables research on human-AI collaboration in software engineering.
Keywords
code agent coding-agent traces human-ai-collaboration software-engineering claude-code codex gemini-cli
Topics
Code / Software Engineering
Research notes
- Requires agreeing to share contact information to access. Associated with arXiv:2604.20779 and website swe-chat.com. Includes PII redaction and deduplication. Six tables: conversations, sessions, session_logs, checkpoints, commits, repositories.