Bagger — Data skill for Claude Code

What bagger does (MVP)

One focused loop:

1. Scan — import local AI conversation transcripts into a local SQLite database.
2. Search — full-text search across every conversation (SQLite FTS5 + BM25, CJK-aware).
3. View — open any session and read the full conversation: messages, tool calls, and results.

That's the whole product. No accounts, no cloud, no telemetry.

Quick start

Requires Python 3.12 or newer.

# Install the CLI and API.
pip install -e ".[web]"

# Initialize the local database and scan existing transcripts.
bagger init
bagger scan

# Search your history, or replay a full session in the terminal.
bagger search "token expiration"
bagger replay 

The default database is `~/.bagger/bagger.db`. A normal scan imports every registered source; use `--source claude` or `--source codex` to limit one.

Optional: desktop app

Prerequisites: Python 3.12+, Node.js 22+, and Rust.

pip install -e ".[dev,web]"
cd ui
npm install
npm run tauri dev

In development, Tauri starts the Python API for you.

Data sources

Source	Transcript location	Format
Claude Code	~/.claude/projects/	JSONL session files
Codex	$CODEX_HOME/sessions/	Rollout JSONL files

When `CODEX_HOME` is unset, bagger uses `~/.codex/sessions/`. The parser registry auto-discovers concrete parsers in `bagger/parsers/`, and database identities are scoped by `(source, id)` so sessions from different tools never collide.

CLI

Command	Purpose
bagger init	Create the data directory and initialize SQLite
bagger scan [--full] [--source …]