Move or remove an installation

Keep the old installation available until you’ve checked the new one. Moving LQ.AI means moving its data, settings and the keys needed to read them. This page assumes you’ve already read Backup and restore — it doesn’t repeat that procedure, only what’s specific to moving or removing rather than recovering from an incident.

Before moving

Include the database, uploaded files, the working gateway.yaml and the gateway master key. Some settings and secrets live in .env on the host computer, outside Docker’s data volumes.

1. On the old host, run Backup and restore in full: a Postgres dump, an object-store (RustFS) mirror, and a copy of the gateway-config volume.
2. Carry your .env secrets over unchanged — POSTGRES_PASSWORD, OBJECT_STORE_SECRET_KEY (or the legacy MINIO_ROOT_PASSWORD), LQ_AI_GATEWAY_KEY, JWT_SECRET, and LQ_AI_GATEWAY_MASTER_KEY if you use encrypted-at-rest provider keys. None of these live in a volume — they exist only where you put them, in your secrets vault or the .env file itself. The new host needs the same LQ_AI_GATEWAY_MASTER_KEY as the old one: without it, the gateway-config volume you moved decrypts into nothing.
3. Bring the stack up on the new host with the restored volumes and the same secrets, then run through Backup and restore’s verification step before you decommission the old host.

Before deleting anything

Check which Compose project and volumes you’re removing. Keep recovery copies until you’ve decided what data must be retained.

Professional duty

Decommissioning a deployment destroys client matter records and the audit trail that evidences what left the deployment, so record-retention and client-file obligations bear on whether a full export must be preserved (and for how long) before the volumes go — a call for whoever is responsible for those matters, not an infrastructure decision. No jurisdiction’s retention period is universal; this page states none.

Stopping is different from deleting

Stopping containers keeps their named data volumes: docker compose down stops and removes the containers but leaves pgdata, miniodata, gateway-config and the rest in place — your data survives a down and comes back on the next up. Deleting the volumes too is a separate, explicit flag:

docker compose down -v

Before you run this

down -v is not a confirmation prompt — it deletes every named volume in the compose file the moment it runs, and there is no undo. CLAUDE.md carries this as a hard rule for the project’s own dev environment for exactly this reason: “NEVER docker compose down -v — it wipes volumes including expensive-to-recreate acceptance data. Rebuild a single service instead.” That rule is written for contributors working against a dev stack, but the mechanism it warns about is identical on a production deployment: pgdata and miniodata are gone the instant the command completes, with no confirmation step in between.

Run -v only once you’ve completed and verified a restore-tested backup per Backup and restore — verified means you’ve actually restored it somewhere and confirmed a chat loads, not only that the dump file exists on disk.

Removing a desktop application does not necessarily remove its Docker data or host configuration — inventory both before you uninstall.

The macOS desktop app. The one documented data-wipe path is the launcher’s own Reset… button (two clicks to confirm), which docs/INSTALL-MAC.md describes as erasing all LQ.AI data on the Mac and re-running first-time setup. Removing the app is more than dragging the bundle to the Trash: quit LQ.AI first; use Reset… (or take a backup, per Backup and restore) to drop the -p lq-ai-desktop Compose volumes; then delete ~/Library/Application Support/lq-ai-desktop/ — “an encrypted config.enc plus a chmod-600 .env” per docs/INSTALL-MAC.md — since it holds the deployment’s secrets and survives deleting the app; only then remove LQ.AI.app from Applications.

Keeping a record instead of a full restore

If you want a record rather than a running restore — closing out a deployment, or handing off to someone who only needs the data, not the infrastructure — the same three artifacts from Backup and restore (a Postgres dump, an object-store mirror, and a gateway-config snapshot) are the complete export.

A narrower option also exists: a single user can export just their own data via POST /api/v1/users/me/export (poll GET /api/v1/users/me/export/{job_id}), with POST /api/v1/users/me/delete as the GDPR Article 17 deletion path. These are the export and deletion workers named in docs/HONEST-STATE.md §1 (api/app/workers/user_export.py, api/app/workers/user_deletion.py, api/app/api/users.py) — not a substitute for a full-deployment export.

Next

• Back up and restore
• Update LQ.AI
• Replace a leaked API key