looselips 0.1.3

Scan your LLM chat exports for personal information

looselips

Scan your ChatGPT and Claude chat exports for personal information you might not want sitting in the cloud.

Install

pip install looselips

Basic usage

1. Export your data (both services email you a download link):
   • ChatGPT: Settings -> Data controls -> Export.
   • Claude: Settings -> Privacy -> Export Data.
2. Create a looselips.toml config defining what to look for (see below).
3. Run:

looselips --config looselips.toml export.zip

The format (ChatGPT vs Claude) is auto-detected. Accepts .zip exports or raw conversations.json files from either service.

Config file

Define your matchers in a looselips.toml. See examples/example.toml for a full example with common patterns.

[[matcher]]
type = "regex"
category = "My Phone"
pattern = '212.?867.?5309'

[[matcher]]
type = "regex"
category = "Home Address"
pattern = '(?i)742\s+Evergreen\s+Terrace'

Patterns use the Python re module. Inline flags like (?i) for case-insensitive, (?s) for dotall, and (?x) for verbose mode (comments and ignored whitespace) work in the pattern string itself.

looselips --config looselips.toml export.zip

LLM matchers

For things regex can't catch, add LLM matchers to your config. Each one runs a separate inference pass per conversation chunk, so prefer a few focused matchers over many broad ones.

model = "ollama/qwen3:0.6b"

[[matcher]]
type = "llm"
name = "Employment & Financial"
prompt = "Find employment and financial information -- company names, job titles, salary figures, stock grants."

[[matcher]]
type = "llm"
name = "Medical & Health"
prompt = "Find medical and health information -- conditions, medications, doctor names."

You can override the model per-matcher with the model key.

Output

Default output is <input (without extension)>_report.html. Override with --output:

looselips --config looselips.toml --output=report.html export.zip

Everything runs locally -- no data leaves your machine (unless you use a cloud LLM model).