Data storage & logs¶

Everything pAInapple Code writes lives in two directories — data under ~/.painapple-code/, credentials under ~/.config/painapple-code/ — never inside your project folders.

Data directory: `~/.painapple-code/`¶

Override with PAINAPPLE_CODE_HOME. The Docker image sets it to /data (a named volume by default).

~/.painapple-code/
├── config.json              # Global settings
├── serve.yaml               # Saved global serve defaults (`painapple setup`)
├── tab-state.json           # Open tabs + active tab (server-side, survives PWA quirks)
├── shortcuts.json           # Keyboard-shortcut overrides
├── favorites.json           # Favorited sessions
├── prompt-favorites.json    # Favorited prompts
├── drafts.json              # Saved prompt drafts
├── presets/                 # Prompt/config presets (one JSON per preset)
├── profiles/                # Named deployments — profiles/NAME/profile.yaml (host mode: also its data home)
├── instance-icons/          # Generated PWA icons (only with --instance-name / --accent)
├── shadow.duckdb            # Turn metadata database (DuckDB, + .wal sidecar)
├── logs/                    # Server logs (auto-rotating, see below)
└── projects/
    └── {hash}/              # 12-char SHA256 of the resolved project path
        ├── path             # Original project path (reverse lookup)
        ├── config.json      # Project-specific settings
        ├── sessions/
        │   └── {id}/
        │       ├── meta.json      # Session metadata (~350 bytes)
        │       ├── messages.jsonl # Parsed conversation + cost (append-only)
        │       ├── raw.jsonl      # Raw Claude I/O (debugging/audit)
        │       ├── tools/         # Large tool outputs (Read_xxx.txt, Bash_xxx.txt, …)
        │       ├── stash.json     # Comments-stash items
        │       └── uploads/       # Uploaded images/files
        └── shadow-git/      # Per-project shadow git repository (auto-journal)

Notes:

One project per path. The {hash} is derived from the absolute project path, so two clones of the same repo at different paths are separate projects.
UI-state files are per-tier. With --state-suffix dev, the top-level state files become tab-state-dev.json, presets-dev/, and so on; project and session history stays shared. See the Server CLI reference.
Uploads before a session exists land in uploads/tmp/ at the top level.

What the shadow journal captures¶

projects/{hash}/shadow-git/ is a private git repository that powers the timeline, undo, and per-file history. After each turn the server commits your entire working tree — including files git itself doesn't track — into that repo. Your project directory is never modified; the repo lives entirely under ~/.painapple-code/.

It skips a default exclude set:

.git/  node_modules/  .venv/  venv/  __pycache__/  .npm/  .cache/
dist/  build/  out/  *.egg-info/
.env  .env.*  *.local
*.log  logs/
.idea/  .vscode/  *.swp  .DS_Store  Thumbs.db
*.img  *.qcow2  *.vmdk  *.iso  *.dmg  (and similar disk images)

…plus anything larger than 50 MB (shadow_git.max_file_size_mb; 0 disables the cap).

Secrets outside the exclude list get copied

A credential file that doesn't match those patterns — credentials.json, secrets.yaml, a stray *.pem — is committed into the shadow repo and stays in its history, even if you later delete it from the project. Before pointing pAInapple Code at a directory holding production keys, either add those paths to projects/{hash}/shadow-git/info/exclude or turn the journal off for that project in Settings → Auto-journal.

Auto-journal also makes a small background model call after each turn (Haiku by default) to write the summaries and commit messages — real API usage, disabled in the same panel.

Config directory: `~/.config/painapple-code/`¶

Override with PAINAPPLE_CODE_CONFIG. Kept separate so credentials survive wiping the data directory.

Path	Purpose
`config.yaml`	Auth password — owner-only (mode `0600`, parent `0700`; an equivalent NTFS ACL on Windows)
`tokens/`	Optional token-profile files (alternate API credentials selectable per session)
`cert.pem`, `key.pem`	Auto-generated self-signed TLS cert/key (when TLS is enabled)

Log files¶

All in ~/.painapple-code/logs/ (override with --log-dir):

File	Content	Rotation
`server.log`	All events, INFO and up	10 MB, 5 backups
`access.log`	HTTP/WebSocket requests with timing	10 MB, 3 backups
`error.log`	Errors with tracebacks	5 MB, 10 backups
`crash.log`	Redirected stderr — uvicorn errors, unhandled exceptions	Append-only, no rotation
`server.pid`	PID of the running server (for post-mortem analysis)	Removed on clean exit

cat ~/.painapple-code/logs/error.log
tail -f ~/.painapple-code/logs/access.log

Docker¶

In the container, PAINAPPLE_CODE_HOME=/data and /data is a named volume (or a bind mount, depending on your DATA_VOLUME setting). To copy the volume out to the host:

painapple extract NAME ./painapple-data-export

See Profiles & container mode.

Backup, cleanup, and wipe¶

# Back up everything (sessions, journal, DB, settings)
tar czf painapple-backup.tgz -C ~ .painapple-code

# Find which project a hash belongs to
cat ~/.painapple-code/projects/{hash}/path

# Remove one project's history (sessions + shadow git)
rm -rf ~/.painapple-code/projects/{hash}

# Full wipe — auth config survives (it lives in ~/.config/painapple-code/)
rm -rf ~/.painapple-code

Stop the server before touching shadow.duckdb

DuckDB is single-writer and keeps a .wal sidecar. Back up or move the database only while the server is stopped, and copy the .wal file together with it.

Uninstalling the package removes neither directory — pip uninstall leaves ~/.painapple-code/ and ~/.config/painapple-code/ in place, so the two rm -rf lines above are the full cleanup. If you installed the optional helpers, install-helpers.sh --uninstall removes the two scripts from ~/.local/bin/ and the agent file from ~/.claude/agents/. Nothing else on the system is touched: pAInapple Code installs no service units, edits no shell rc files, and sends no telemetry — the only outbound request it makes on its own is the browser widget fetching a URL you asked it to open.

Resuming sessions in the plain CLI¶

Sessions created through pAInapple Code are regular sessions of whatever provider ran them — nothing proprietary is layered on top. The provider session ID is stored in each session's meta.json (provider_session_id), and you can pick any conversation up from a terminal.

The resume command follows the session's own provider. Each provider self-describes its template, so a Codex session is not resumable with claude:

claude -r <session-id>              # Claude providers (claude-sdk, claude)
codex exec resume <session-id>      # Codex provider (codex-app-server)

The in-app Continue in CLI quick action copies the right one for the session you're looking at, so you don't have to remember which provider it was bound to.