Data & cache
One data directory, one config directory, your OS keychain. Nothing is stored anywhere else, and backing it up is one copy while the app is closed.
On this page
The files
| File | Contents |
|---|---|
cache.db |
The SQLite store: every run’s history, snapshots, learned selectors. Shared by the app and the CLI |
records/ |
Per-run artifacts beside the store: <runId>.jsonl crawls, <runId>.scrape.json replays, .part resume legs |
config.yaml |
Core config: provider and model defaults, cost caps. Shared with the CLI |
desktop.json |
The app’s own preferences: theme, density, window geometry, network defaults, plan ledger |
schemas/ |
The schema library, beside config.yaml |
watches.json |
The watch registry |
sessions.json |
Captured logins, sealed — the OS keychain where one exists |
| OS keychain | Provider keys, webhook secrets, the licence record |
Settings → Data prints the exact store path on your machine, and the footer shows it on every screen. Open folder reveals the directory.
Where, per OS
| Config | Data (store + records) | |
|---|---|---|
| macOS | ~/.config/magpie/ |
~/Library/Application Support/magpie/ |
| Windows | %APPDATA%\magpie\ |
%LOCALAPPDATA%\magpie\ |
| Linux | ~/.config/magpie/ |
~/.local/share/magpie/ |
The store deliberately sits in the OS data dir, not the cache dir: cache cleaners empty cache folders, and a multi-gigabyte history is not a cache. An SSRF guard also stops the app from fetching internal network addresses — the data directory is the only place results go.
Backups
Quit the app (and any running CLI crawl), then copy cache.db and records/. That is the whole backup — restore them to the same paths and every run, snapshot and learned selector is back. WAL mode means a live copy can catch a half-written transaction, which is why the app should be closed.
Deleting is equally direct: remove a schema file and it leaves the library; clear the selector cache in Settings → Data and the next extraction re-learns. The store itself is meant to be kept — History’s per-run LLM ledger is the only cost record you have.
The legacy path
Installs from before v0.1.30 keep their store at ~/.cache/magpie/cache.db. It still works — it is never moved automatically, because a move races the CLI and can lose records on a cross-volume rename — but cleanup tools may empty it. Settings → Data shows the exact move steps when it applies:
- Quit Magpie and any
magpieCLI run. - Move the store to the data path shown.
- Start Magpie again.
Research runs additionally keep their stored page copies under the data directory; Forget on a recent run deletes that run’s copies and read trail — its spend stays in the budget history.
What the exports contain, field by field, is on Export formats.