Storage & the local engine

Where JustWrite keeps your data

Everything JustWrite saves lives in one folder — your projects, images, the local AI engine and any downloaded models, and logs. By default that folder sits next to the app, so a portable install carries its data with it. If the app is installed somewhere the system won’t let it write (like Windows Program Files), it falls back to your user data folder instead.

You can see the current location under Settings → Storage.

Running the server yourself (see Headless access) follows the same rule — the data folder beside the app — and takes the same two overrides the desktop app respects: the JUSTWRITE_DATA_DIR environment variable, or --data-dir <path> on the command line. JustVoice behaves identically with its own JUSTVOICE_DATA_DIR, so the two apps never surprise you differently.

Moving your data to another folder

In Settings → Storage, click Change folder…, pick a location, and confirm. JustWrite makes a new folder for itself inside the one you picked, copies everything into it, and restarts on it. One move runs at a time, and it first checks that the drive has room.

Your work is never at risk: the move only takes effect once the copy has fully finished, and if anything goes wrong the app keeps using the old folder. Large model files make the move take a while — that’s expected.

The old folder keeps its copy until you remove it: Settings → Storage lists it as an Old copy with its size and Remove the old copy…, which deletes only what the move copied there. An old copy on a drive that isn’t connected shows as Not found: connect the drive to remove it, or Forget… it — the app stops listing it and nothing on any drive is touched. Move back to the default folder… returns the data to where it started. When the JUSTWRITE_DATA_DIR variable names the folder, the moves are off and the page says where the folder is set.

Reclaiming disk space

The Disk usage card under Settings → Storage shows where your data folder’s space goes: the downloaded AI models, the local engine, the server logs, the engine’s own start-up logs, and your project database — plus how much free space the disk has left.

Two of those can be cleared right from the card:

• Models cache — Clear… removes the model files JustWrite downloaded — the models in its catalog — and nothing else: other files in the folder, and a folder you linked elsewhere, are left alone. This is always safe: your models stay in the catalog and simply re-download the next time they’re used. If a model is currently loaded, JustWrite asks you to unload it first (on the AI page) before clearing. This is usually where most of the space goes — model files run to several gigabytes each.
• Engine logs — Clear… removes the local engine’s start-up logs. They’re only useful when diagnosing a model that won’t load, and they accumulate over time.

The other rows are managed where they live: the engine itself is installed, updated, and removed on the AI page; the server logs have their own retention controls in Settings → Logs; and the database is your actual work — it’s never cleared from here.

Log timestamps

On screen — in Settings → Logs and in the Server console under AI — log times are shown in your computer’s own date and time format, so a US machine reads 07/19/2026, 12:06:22 AM and a European one reads its local convention. The day list you pick a stored log from follows the same format. Nothing to configure: JustWrite follows whatever the machine it’s running on is set to.

Inside the saved log files — and in anything you Download or Copy out of the Logs panel — timestamps are written in the international 2026-07-19T00:06:22.169 form instead. That’s deliberate — it sorts correctly, it’s unambiguous no matter whose machine opens the file, and it keeps the millisecond precision that’s useful when you’re sending a log in for diagnosis. Both are the same moment on your local clock; only the presentation differs.

The local AI engine

The bundled local models run on a small llama.cpp engine. Installing it is its own one-time step, separate from downloading a model: open Settings → AI → Providers & models, open the built-in provider, and use Install engine under Local engine. Once it’s installed, downloading and loading a model just works — if a model won’t load with an “install engine first” note, the engine step is what’s missing.

If a model ever fails to start, View log on that panel shows the engine’s own output so you can see why (for example, a graphics-driver or out-of-memory error).