Tagvico AI v3.2.5
Tagvico 3.2.5 makes the path from a new installation to the first useful
question safer, shorter, and easier to recover.
Highlights
- Guided setup verifies Paperless access, provider authentication, the live
model catalog, and the exact selected model before saving. - Built-in endpoints are prefilled, authenticated Ollama endpoints are
supported, and local and cloud Ollama credentials remain separate. - Docker- and host-injected connection values are verified as the effective
runtime configuration, and public model catalogs are bounded by bytes, model
count, time, and concurrency. - Non-secret setup progress resumes in the same browser tab without storing
tokens, passwords, provider keys, or account credentials in the draft. - Abandoned ChatGPT device sign-ins expire server-side and release their Codex
process before a rate-limited later setup attempt can start cleanly. - Paperless permission checks run concurrently and verified setup is persisted
once, keeping slow but valid connections within the completion deadline. - New installations start in Review first mode with scheduled scans paused and
continue directly to Ask Tagvico after setup. Enabling a schedule later in
Settings takes effect without restarting the container. - Ask Tagvico source titles link to the permitted Paperless-backed document
view for direct verification. - Authentication, rate-limit, and transient Paperless failures remain
retryable instead of being shown as missing documents. - Loading, empty, retry, and mobile states are clearer across setup, Documents,
Ask Tagvico, Activity, and Settings. - First-party aggregate landing requests remain separate from explicitly
opted-in installation reports, without cookies, fingerprints, referrers,
stored IP addresses, or permanent installation identifiers. - Release acceptance exercises setup, synthetic document upload, search, cited
answers, tag changes, approval-first actions, cleanup, and recovery against
an isolated real Paperless-ngx 2.20.15 container.
Upgrade
Back up the tagvico_ai_data volume, pin
ghcr.io/arturict/tagvico-ai:3.2.5, and recreate only the Tagvico container.
Existing history, conversations, credentials, model choices, automation rules,
and recovery queues remain in the persistent volume.
This is a v3 patch and does not require a data migration.