Skip to content

Dataset · SALT-NLP

Real-world AI coding sessions

SALT-NLP/SWE-chat

Developer dialogues with AI assistants: tool calls, reasoning traces, and code edits.

What's inside

SWE-chat collects real development sessions with AI coding assistants, including Claude Code, Codex, and Gemini CLI. Each session contains full conversation transcripts, tool calls, reasoning traces, code changes, and human vs. agent code attribution labels. Metadata languages: English, Japanese, Chinese, Korean, Russian, and Portuguese; size: 10M–100M records.

What you can do with it

  • Training coding agents on real-world sessions
  • Evaluating agents on SWE tasks
  • Studying human-agent interaction
Tasks
Text generation
Languages
English, Japanese, Chinese, Korean, Russian, Portuguese
Size
10M–100M rows
Downloads this month
18.2K
Likes
175
Access
on request on Hugging Face

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("SALT-NLP/SWE-chat")