Tagvico AI v3.2.0

Tagvico AI v3.2.0

Tagvico 3.2 replaces the parallel dark and light surfaces with one coherent
Paper & Pine interface across the application, landing page, and bundled
documentation.

Highlights

• Home, Documents, Ask Tagvico, Organize tags, Activity, and Settings now map
  directly to the jobs users perform.
• Ask Tagvico shows its document and tag research, groups configured models by
  provider, and keeps every document or tag mutation behind an explicit
  approval.
• Content requests such as read, inspect, open, or review now read the bounded
  Paperless search results, while unrelated substring matches do not start
  research.
• GitHub Copilot now receives the saved reasoning effort, and mobile users can
  reopen conversations without losing the answer or research trail.
• Approval cards show the actual proposed document values, while tag deletions
  recheck the approved name and linked-document count immediately before the
  write.
• Duplicate-tag cleanup has a dedicated many-to-one review flow instead of a
  dense Settings control.
• Page-shaped skeletons preserve the expected layout while Home, Documents,
  Activity, and provider catalogs load.
• Document tags, provider controls, model metadata, buttons, and switches have
  stronger contrast throughout the light interface.
• The public landing page and versioned v3 documentation now use the same
  design, terminology, provider icons, and current product screenshots.
• Owner-only tag organization is hidden for other household roles, Ollama
  catalog inspection has one bounded deadline, and model-specific reasoning
  choices stay visually synchronized after saving.

Upgrade

Back up the tagvico_ai_data volume, pin
ghcr.io/arturict/tagvico-ai:3.2.0, 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 update. It does not introduce a v4 migration or a parallel beta
theme.

Tagvico AI v3.1.2

Tagvico AI v3.1.2

Tagvico 3.1.2 is a focused Ask Tagvico model-picker hotfix.

What changed

• Embedding-only models from live provider catalogs are no longer offered as
  chat models, including colon-delimited Ollama IDs such as
  qwen3-embedding:4b.
• The filter also covers common embed model names and provider IDs using
  slash-delimited segments.

Upgrade

Back up tagvico_ai_data, pin
ghcr.io/arturict/tagvico-ai:3.1.2, and recreate only the Tagvico container.
No settings or data migration is required.

Tagvico AI v3.1.1

Tagvico AI v3.1.1

Tagvico 3.1.1 completes the reliability and product-quality work started in
3.1.0. It keeps Paperless-ngx as the source of truth while making automation,
provider configuration, recovery, and Ask Tagvico coherent enough for daily
use.

What changed

• Ask Tagvico now has durable conversations, new-chat, search, rename, guarded
  deletion, retry, copy, stop, model selection, and privacy-safe tool activity.
• Paperless research is intent-aware. Greetings do not query the archive;
  counts, recent-document lists, searches, and bounded document reads use
  purpose-specific tools.
• Settings exposes eight supported runtimes with live model catalogs,
  write-only API-key fields, connection probes, and first-class ChatGPT and
  GitHub Copilot authentication. Direct Anthropic and Azure OpenAI setup is no
  longer shown.
• Trigger tags are optional. With no trigger tags configured, every eligible
  new document is scanned, and manual scans report exact eligible, processed,
  staged, skipped, and failed counts.
• Four remains the default maximum tag count, with stronger prompts for the
  smallest useful tag set and an archive-specific custom prompt plus an
  advanced system-prompt editor.
• Activity includes metadata diffs, custom fields, token usage, original state,
  bulk rescan, exact restore, orphan validation, and deliberate cleanup.
• AI and OCR failures retry up to three times. Permanently failed and ignored
  documents have dedicated recovery queues and navigation counts.
• Duplicate-tag proposals are presented as explicit many-to-one groups and
  retain per-group approval plus separate move and delete phases.
• App, versioned docs, landing page, package icons, and favicons use one
  transparent Tagvico AI mark with no inherited Paperless AI artwork.

Upgrade

Back up tagvico_ai_data, pin
ghcr.io/arturict/tagvico-ai:3.1.1, and recreate only the Tagvico container.
No destructive settings migration is required. Existing environment keys remain
compatible, and the persistent data directory continues to hold configuration,
history, queues, sessions, and restore snapshots.

After upgrading, open /docs/, test the Paperless connection, refresh the live
model catalog for each configured provider, run a manual scan, and verify one
non-sensitive Ask Tagvico research request before enabling Automatic write
mode.

Tagvico AI v3.1.0

Tagvico AI v3.1.0

Tagvico 3.1 is a reliability and product-coherence release. It does not change
the core promise of v3: Paperless remains the document system of record, model
reads are permission-aware, and every Companion write waits for approval.

What changed

• One green React interface now covers Actions, Ask Tagvico, Automation,
  Review queue, Activity, Recovery, Manual processing, Setup, and Settings.
• Navigation names describe user jobs instead of internal operating concepts.
• Ask Tagvico shows redacted research/tool activity and offers only configured,
  live-discovered models. The document-automation model remains the default.
• Paperless instance discovery is back in Settings.
• Tag library can propose duplicate-tag merges. Every merge is approved
  separately and executed as an explicit document-move phase followed by a
  verified source-tag deletion phase.
• The Docker image includes the matching versioned docs at /docs/;
  /documentation points to the same local copy.
• Loading paths have timeouts, partial results, retry actions, and stable error
  states instead of permanent skeletons.
• App, docs, and landing use the Tagvico icon and shared green visual system.

Upgrade

Back up tagvico_ai_data, pin the immutable 3.1.0 image after the release is
published, and start in Review first mode. Open /docs/, test the Paperless
connection, load the configured provider's live model catalog, and process one
non-sensitive document before enabling Automatic mode.

No destructive settings migration is required. Existing .env keys remain the
compatibility persistence layer while the typed Settings API writes validated
partial updates.

Tagvico AI v3.0.0

Tagvico AI v3.0.0

Tagvico v3 turns Paperless-ngx documents into accountable household work while
keeping the established reviewable filing workflow intact.

Highlights

• Create one Action Case per Paperless document with an owner, priority, due
  date, state, audit trail, and up to 100 checklist steps.
• Ask the Companion about permitted documents and current actions. Read tools
  execute directly; every AI-proposed write becomes a durable proposal that an
  owner or adult must explicitly approve.
• Give household members roles and encrypted, member-specific Paperless tokens
  instead of sharing one privileged credential.
• Mirror top-level case state to reserved Paperless custom fields and the
  tagvico/action tag while keeping full checklists local to Tagvico.
• Use the new Next.js console for actions, approvals, Companion sessions, and
  household settings alongside the existing filing and recovery operations.
• Optionally list actions and approve or reject proposals through Telegram,
  still using each allowlisted person's own Paperless token.

Upgrade from v2

1. Stop Tagvico and back up the complete tagvico_ai_data volume.
2. Pin ghcr.io/arturict/tagvico-ai:3.0.0 in Compose.
3. Start the container and allow the schema-v5 migration to finish. It adds
   households, members, cases, steps, Companion sessions, and approvals.
4. Verify /health and /api/health, sign in, and test the Paperless and model
   provider connections.
5. Process a non-sensitive document in Review first, create one Action Case,
   and approve one synthetic Companion proposal before using real household work.

The public application stays on port 3000. The scanner runs internally on port
3001 and must not be exposed. Do not run v2 and v3 against the same data volume.
Restore the backup before rolling back if v3 migrated the database.

Stable boundaries

• Paperless-ngx remains the document system of record.
• Companion tools never receive host shell or filesystem access.
• AI-proposed writes cannot bypass the durable approval gate.
• Provider model catalogs, prices, quotas, and subscription entitlements remain
  provider-controlled.
• Telegram is optional, private-chat-only, and not end-to-end encrypted.

Full documentation: https://tagvico.arturf.ch/docs/

Tagvico AI v2.0.1

Security\n\n- Prevent unauthenticated replacement of configuration and the local admin account through POST /setup after setup has completed.\n\n## Documentation\n\n- Pin v2 installation examples to the immutable 2.0.1 image.\n- Clarify one-time remote setup, the 409 recovery path, and exact rollback steps.\n\n## Validation\n\n- CI quality gate passed.\n- Node 22: 84 unit tests, type/lint/branding/EJS/browser checks, and versioned docs build passed.

Tagvico AI v2.0.0

Tagvico AI v2.0.0

Tagvico AI v2.0.0 is the first stable release of the self-hosted, reviewable
AI filing workflow for Paperless-ngx.

Highlights

• Queue durable metadata suggestions for review or enable validated automatic
  writes per installation.
• Apply or reject structured diffs and restore the first metadata snapshot,
  including exact tag, correspondent, document-type, date, language, custom
  field, and owner values.
• Constrain generated tags with controlled groups, per-document caps, and an
  exception queue.
• Use Ollama, Ollama Cloud, OpenAI, Anthropic, OpenRouter, OpenCode Go, Azure
  OpenAI, GitHub Copilot, OpenAI-compatible gateways, or experimental ChatGPT
  subscription access.
• Recover OCR-poor documents, interrupted work, retries, and terminal failures
  from durable queues.
• Inspect processing history, token usage, estimated cost, runner state, and
  provider health from the web interface.
• Optionally share a coarse installation heartbeat. Analytics are off by
  default, the exact payload is locally previewable, identifiers rotate, and
  no official collector endpoint is embedded.

Upgrade from v1 or the v2 alpha

1. Stop Tagvico and back up the complete tagvico_ai_data volume.
2. Replace deprecated pre-Tagvico environment variables with their documented
   TAGVICO_* equivalents.
3. Pin ghcr.io/arturict/tagvico-ai:2.0.0 in Compose.
4. Start the container and verify both /health and /api/health.
5. Sign in, test the Paperless and model-provider connections, and process a
   representative non-sensitive document in Review first.
6. Enable Automatic mode only after reviewing the resulting metadata.

Do not run two Tagvico versions against the same data volume. Restore the
volume backup before rolling back if a newer version migrated the database.

Stable boundaries

• The v2 Docker, persistent-volume, canonical environment-variable, migration,
  setup, review, history, and restore contracts are stable.
• ChatGPT subscription access remains experimental and account-specific.
• Model catalogs, pricing, quotas, and entitlements remain provider-controlled.
• Compatible and subscription-backed adapters may report provider health as
  unknown; use the explicit connection tests in Settings before processing.

Full documentation: https://tagvico.arturf.ch/docs/

Tagvico AI v2.0.0-alpha.1

Tagvico AI v2.0.0-alpha.1

This is the first public v2 prerelease. It is intended for representative
testing with an existing Paperless-ngx installation before stable 2.0.0.

Highlights

• Review suggestions before writing them, or opt into Automatic mode per
  installation.
• Inspect structured metadata diffs, apply or reject durable suggestions, and
  restore the first captured metadata snapshot.
• Keep tag vocabulary controlled with groups, caps, and an exception queue.
• Use local Ollama, direct APIs, compatible gateways, OpenCode Go, Ollama
  Cloud, GitHub Copilot, or experimental ChatGPT subscription access.
• Recover OCR-poor documents, interrupted work, retries, and terminal failures
  from the Operations UI.
• See token use, estimated model cost, history, runner state, and provider
  health from the web interface.
• Optionally share coarse installation analytics. Sharing is off by default,
  uses rotating identifiers, exposes the exact payload locally, and requires an
  explicitly configured HTTPS receiver.

Upgrade from v1

1. Stop Tagvico and back up the complete tagvico_ai_data volume.
2. Replace deprecated pre-Tagvico environment variables with their documented
   TAGVICO_* equivalents.
3. Pin ghcr.io/arturict/tagvico-ai:2.0.0-alpha.1; do not use latest for this
   prerelease.
4. Start the container, inspect its logs, and verify /health and
   /api/health.
5. Process synthetic or non-sensitive representative documents in Review
   first before enabling Automatic writes.

Do not run v1 and v2 against the same data volume simultaneously. Restore the
volume backup before rolling back if v2 has migrated its database.

Known prerelease boundaries

• ChatGPT subscription access is experimental and is not an API SLA.
• Model catalogs, entitlements, pricing, and quotas remain provider-controlled.
• The REST/configuration/database contracts may still change before stable v2.
• The reference analytics receiver must be deployed separately; no official
  collector endpoint is embedded in this prerelease.

Full documentation: https://tagvico.arturf.ch/docs/

Help validate v2

Test with synthetic documents, then share a redacted installation, upgrade, or workflow result. Reproducible bugs belong in the linked bug-report form; never include credentials or personal document data.

v1.3.0

v1.3.0 - 2026-07-05

• Project Rebranding: Renamed the project from Archivista AI to Tagvico AI due to trademark requests.
• Controlled Tagging & Tag Groups: Added dynamic configuration for Tag Groups (presets like Finance, Health, etc., plus custom groups) to restrict LLM suggestions to a precise vocabulary.
• Exceptions Review & Approval Queue: Created a SQLite-backed workflow where unknown LLM-suggested tags are held for manual review. Users can approve them into specific tag groups or reject them.
• Provider Modal & UI Polish: Replaced the bulky provider grid on the configuration page with a searchable provider/model modal featuring SVGL icons.
• Integration Test Stack: Added Postgres 18, Redis 8, and Paperless-ngx services to \docker-compose.yml\ for simplified local integration testing.
• Paperless Username Autodetect: Cleaned up credential lookup to fallback to the first active user when \PAPERLESS_USERNAME\ is blank.
• Fixed constrained scrolling in the model picker on mobile and desktop.
• Read the displayed application version dynamically from \package.json\ instead of a stale hard-coded value.

Tagvico AI v1.2.0

Highlights

• Official Codex app-server login for ChatGPT subscriptions
• Review queue with audit diffs, restore/rescan and live SSE progress
• OCR rescue and reconciliation operations
• Provider health checks and OpenAI-compatible setup documentation
• MFA, rate limiting and container hardening

See CHANGELOG.md for the full list.

Verification: npm test passes (4/4 tests), EJS templates compile, and the Codex app-server handshake was exercised successfully.