Image: GitHub
Why it mattersA developer who has run six months of problem-solving conversations across Claude Code and Cursor can pull them into one file and search, back up or train a small model on the whole set.
A conversation about why a build is broken, written on a Monday with Claude Code and copy-pasted into Cursor on Tuesday, lives in two different files that no tool knows how to search together.
ai-data-extractor is an MIT-licensed command-line tool from GitHub user jatechan that reads chat history out of ten AI coding assistants and writes it into one standard format. The repository was created on 11 October and has 438 stars and 96 forks eight days later.
What it reads
The README lists ten supported tools: Claude Code, Codex CLI, Cursor, Windsurf, Trae, Continue, Gemini CLI, OpenCode, Cline and its Roo Code fork, and Aider. The project describes its approach as detecting the operating system, building "a list of plausible data roots," then reading each tool's own storage format, which the README says can be a JSONL file, a SQLite database, or a JSON file depending on the tool. It runs on macOS, Linux and Windows.
Each conversation becomes a single line of a timestamped JSONL file. The schema the README describes carries the messages, the roles of who said them, the code the assistant saw, the file paths it touched, the tool calls it made, timestamps, session identifiers and the project the session was for. The author lists three uses for the result: fine-tuning a smaller model, personal analytics, and keeping a backup.
The privacy section matters more than the install step
The README has a privacy section that is worth reading before pointing the tool at a work laptop. The author tells users to scan the extracted files for secrets with a tool such as detect-secrets, to look at the code_context and tool_use fields for proprietary code and API keys, to keep the extracted directory out of a public repository, and to encrypt the output when the work is sensitive. The author also writes that each tool's terms of service apply, so a reader on a corporate account has to check whether exporting the conversations is allowed in the first place.
Why the file matters
A chat with a coding assistant is a record of what a developer was trying to do, what the assistant was told, what it got right and what it got wrong. Each tool keeps that record in its own place and its own shape, which means a team that uses two of these tools has no way to answer "when did we last run into this error" across both. Pulling the records into a single schema turns them into a thing that can be searched, saved alongside the project, or run through an analysis script.
The use the README puts first is fine-tuning. That is a reasonable idea if the user owns the conversations and the model, and reads the licences of the tools they are extracting from. For most readers the first use is likely to be the second one on the list: a durable copy of the work, in a format that will still open in five years.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.