TYDAL 1.5.0

Added

• AITY suggestions at the vault write boundary. A gallery ingest with
  suggest: true runs AITY on the stored photograph — opt-in per photograph,
  because the consumer may hold the photographer's consent for some and not
  others. AITY only proposes; what the consumer sent is never changed, and
  AITY failing never fails the ingest. GET …/w/suggestions?resources=…
  returns the proposed title and description per ingested resource (off,
  pending, done, failed) to a key that may update; reading never
  starts AITY.
  @tydal/client gets suggestions(resources). See
  docs/architecture/VAULT_WRITE_METHODS.md §3.

Upgrading

No migrations and no new configuration. Suggestions run on the existing AITY
pipeline, so they need what AITY already needs: a configured LLM driver and a
queue worker. @tydal/client goes to 1.7.0 (suggestions);
@tydal/org-mcp and @tydal/vault-mcp keep their versions.

Full changes: v1.4.1...v1.5.0

TYDAL 1.4.1

Added

• tools/deploy/server/: the files of a working server setup, ready to
  copy — one infrastructure compose (MySQL, Elasticsearch, Redis with a
  password, Tika; ports on 127.0.0.1, data on a data disk) and, per
  installation, a compose file, a make-env.sh that writes the production
  backend/.env and creates the database, and an nginx site. With
  ELASTICSEARCH_INDEX_PREFIX and the per-installation Redis, cookie and
  database names, several installations share one server. DEPLOYMENT.md:
  "On a server, by hand".

Changed

• FullFrame, one word. The photo-exhibition product is FullFrame
  everywhere: the docs, the help of exhibitions:setup / exhibitions:create
  and the lines they print ("Paste into FullFrame → …", which FullFrame's
  user guide quotes), comments and seeders. The CLI guide's section anchor is
  now #photo-exhibitions-fullframe.

Upgrading

Nothing to do: documentation, example files and printed text only — no
migrations, no configuration, and @tydal/client, @tydal/org-mcp and
@tydal/vault-mcp keep their versions.

Full changes: v1.4.0...v1.4.1

TYDAL 1.4.0

Added

• Several installations on one server (#23): configure.sh <ai> <host|docker> --infra=shared --name=NAME (also accepted by install.sh) makes a checkout
  use another checkout's running MySQL, Elasticsearch, Kibana, Tika and Redis
  instead of starting its own. NAME derives what must differ: database and user
  tydal_NAME (test database tydal_NAME_test), index prefix NAME_, Redis
  prefix tydal_NAME_ plus its own Redis DB numbers, session cookie, the
  SPA's login-token cookie (VITE_AUTH_COOKIE, default JWT),
  containers tydal_NAME_* (Compose project tydal_NAME), and app/Vite ports
  offset by 100·slot (--slot=N, else kept from the last run, else the lowest
  free slot; a slot whose ports or Redis DBs another installation uses is
  refused). The infrastructure services carry a Compose profile that
  is empty (always on) by default and switched on only in shared mode; on the
  docker tier the app services join the infrastructure stack's network.
  New tools/deploy/provision-shared.sh creates the database, test database
  and user (idempotent, --dry-run); start.sh, test.sh,
  first_install.sh (resets only that installation's database, indices and
  Redis keys), install.sh and reindex.sh follow the configured names via
  tools/lib/stack.lib.sh. Container names and host ports in
  docker-compose.yml are now variables (TYDAL_STACK, TYDAL_HTTP_PORT, …)
  whose defaults are the old literals, so --infra=own (the default) is
  unchanged. Tests take a per-installation database and index prefix from
  TYDAL_TEST_DB_DATABASE / TYDAL_TEST_INDEX_PREFIX (tests/bootstrap.php).
  DEPLOYMENT.md: "Several installations on one server".

• --url=URL for configure.sh / install.sh: the public address of an
  installation (own or shared) — writes APP_URL, SANCTUM_STATEFUL_DOMAINS
  and the SPA's VITE_API_BASE_URL, and keeps it as TYDAL_PUBLIC_URL in
  backend/.env so a later re-run no longer resets it to localhost;
  --url= drops it. DEPLOYMENT.md: "The public URL".

• configure.sh --check, run by install.sh before its prompt: validates
  the flags, the slot and, in own mode, that no other checkout holds the
  tydal_* container names (it names that checkout's folder) — instead of
  failing halfway through docker compose up. --name=NAME alone now
  implies --infra=shared.

• install.sh ends with a summary of what it installed (installation, slot,
  app and Vite URLs, public URL, database and index prefix, container status)
  instead of letting it scroll away; composer and npm run quieter.

• ELASTICSEARCH_INDEX_PREFIX (#14): a per-installation prefix applied to
  every physical index name — collection indices, their _chunks
  companions and vault_<uuid> — through one helper,
  ElasticsearchService::physicalIndexName(). search_indexes.index_name
  stays the logical, unprefixed name. Destructive commands only reach this
  installation's prefix: search:wipe-indices deletes
  {prefix}tydal_*,{prefix}vault_* (and warns when the prefix is empty),
  search:setup-indices --recreate and search:reconcile --fix act on the
  prefixed names. A prefix with wildcards, or one starting with tydal_ /
  vault_, is refused. Empty by default, so existing installations are
  unchanged. In every backend env template, DEPLOYMENT.md
  and CLI.md, with the
  migration recipe for an existing installation. The test suite forces
  test_, so it never touches a dev installation's indices.

• Archived page (#25): archived resources left every listing, search and
  vault, and no page listed them, so once out of the basket they couldn't be
  found again in the web app. New GET /api/v1/resources/archived (MySQL, not
  Elasticsearch) lists the current organization's state = archived
  resources with the trash's rule — editors and viewers see their own,
  administrators and owners see all — sorting (date, name, ID) and
  pagination. The SPA's new Archived page (/archived, linked beside
  Deleted resources) is modelled on the trash. Each card has Set live
  and Move to trash icons; the page has Set all live (through
  POST /resources/bulk/state in batches of up to 200) and Delete all
  (moves them to the trash one by one, after a confirmation). Both report
  skipped resources like the basket and leave them listed, and each control
  shows only with its permission (resources.update / resources.delete).
  User guide chapters 5 and 7.

• When a resource was archived (resources.archived_at): stamped when a
  resource moves to archived (single update, bulk state, any save — the
  model's saving hook, called explicitly by the quiet bulk write), cleared
  when it leaves. Migration 2026_10_06_000000_add_archived_at_to_resources
  backfills existing archived rows from updated_at. GET /resources/archived
  sorts by archived_at by default (sort_by accepts archived_at, name,
  id, and still updated_at), and the Archived page's cards read
  "Archived ".

• aity:purge-batches: deletes AiTy Review batches reviewed (opened from
  the AiTy Review page) or whose auto-approve job finished more than
  AITY_BATCH_RETENTION_DAYS (default 30; 0 disables) days ago — the batch
  workspace only, through WorkspaceService::deleteWorkspace, never its
  resources. Batches still waiting, running or being analysed are kept.
  --days=N, --dry-run; scheduled daily at 04:00. docs/CLI.md.

• Required fields at the wizard's first step (#9). When the collection's
  scheme has required metadata fields, step 1 shows a Required by this
  collection panel and Next waits until they're filled; the values go into
  every created resource and pre-fill Review. If every upload fails, the
  Upload step shows the server's reason.

Changed

• Editors can use the wizard's Auto option (#10). Auto opens an AiTy
  Review batch (a workspace of purpose aity_review), which needed the
  admin-only workspaces.create; editors may now open those batches
  (resources.create suffices), while ordinary workspaces stay admin-only.
  The wizard no longer swallows an Auto failure: it stays open with the
  reason and reuses the batch already created on retry.

• Manage Workspaces leaves AiTy Review batches out. Every Auto upload
  creates one, and the dialog listed them all under Managed by the system;
  they have their own page (AiTy Review). Other system workspaces are still
  listed.

• Consistent wording (#8, #12, #22). The resource view's legacy
  Active/Visibility controls are gone and cards say Live / Draft / Archived;
  the delete confirmation no longer says "cannot be undone" (resources go to
  the trash); one short id (the tail of the UUID v7) on cards, trash and
  basket; the vault key help names ingest / update / withdraw; the AiTy
  Review empty state points to the wizard's Auto option; VAULT_SYSTEM.md §3
  matches who creates vaults; categories are documented as API-only.

• install.sh / configure.sh print the full help when the tiers are
  missing.

• Only administrators change what a vault shows. Adding resources to, or
  removing them from, a workspace attached to a vault publishes or unpublishes
  them outside TYDAL, so it now takes the new ability
  workspaces.manage-vault-resources (admins and owners, via workspaces.*)
  on top of workspaces.manage-resources. Enforced in
  WorkspacePolicy::manageResources, so it covers the single add/remove
  endpoints (resource editor chips, org-mcp tools) and the basket's
  bulk-attach/detach, which report every id skipped as vault_connected.
  Editors keep curating plain workspaces and AiTy Review batches; vault write
  ops are unaffected. GET /workspaces now returns vaults_count, and the
  basket's Workspace dialog and the resource editor's workspace chips show
  vault-connected workspaces disabled ("Shared through a vault — ask an
  administrator") to users without the ability. The basket's empty Workspace
  list now says how to get one (ask an administrator, or Manage workspaces).

Fixed

• Unbounded page size on the trash and Archived listings:
  GET /resources/trashed and GET /resources/archived now clamp limit to
  1..200, the bulk endpoints' cap.

• Viewers could run AiTy auto-approve (#19). POST /aity/auto-approve
  and /aity/auto-approve/stream ran for any member of the organization;
  both now require manageResources on a workspace of the current
  organization (403, or 404 for another organization's), the stream before
  it opens.

• The wizard's exit could fail silently or leave drafts (#20, #21). Keep
  and Delete all report per-resource failures with Retry or Leave anyway;
  they wait for the upload in flight ("Finishing uploads…") so a resource
  created after the click isn't left as a draft for the hourly purge. Back →
  Next with the same files retries only the failed rows; a resource whose
  file upload fails is deleted again, so the counts match what exists.

• Editors were offered Save as workspace in the basket (#11), which they
  can't do; it now shows only with workspaces.create.

• Resources without a description couldn't be edited (#24). The built-in
  schemes (multimedia, documents, general) required description, but only
  the edit form enforced it, so resources created by the wizard, FullFrame or
  AI ingest, or the API — all without one — couldn't be saved or archived
  from the form. description is now optional in those schemes (seeder and a
  migration for existing installations; the 10-character minimum still applies
  when one is ...

TYDAL 1.3.0

Added

• User guide in docs/user-guide/, for the
  people who use TYDAL rather than run it: viewers, editors, administrators,
  owners and platform admins. Nine chapters with screenshots: getting started,
  finding resources, the upload wizard and AiTy suggestions, tags, the basket
  and bulk actions, workspaces and vaults, the trash, organization settings
  and platform administration. Its Known issues section links to the open
  bug issues on GitHub.
• Agent guide, docs/AGENT_GUIDE.md, for AI agents
  that use a TYDAL installation. It covers the three rings (installation;
  organization, with AiTy and @tydal/org-mcp; vault, with
  @tydal/vault-mcp), which one to use, how to connect, the tools, and rules
  of conduct.
• AGENTS.md for AI coding agents working on this
  repository: layout, tier-aware commands, tests, and the codebase's rules.
• TRUSTED_PROXIES (backend/config/trustedproxy.php): the IPs or CIDRs
  of a TLS-terminating proxy to trust, or * for the direct peer. Behind a
  host nginx, TYDAL saw plain http and built http:// vault and file links;
  with the proxy trusted, they're https://. Empty by default, so
  artisan serve and the docker dev tier are unchanged. In every backend env
  template and in DEPLOYMENT.md.

Upgrading

1. Behind a TLS-terminating proxy (a host nginx, a load balancer): set
   TRUSTED_PROXIES in backend/.env to the proxy's IPs or CIDRs, or to *
   to trust the direct peer. * is safe only when the app's port is published
   on 127.0.0.1, so nothing but the proxy can reach it. The templates have
   the variable, with a comment. Without a proxy, leave it empty.
2. Nothing else: no migrations, and the MCP packages and @tydal/client keep
   their 1.2.0 versions.

Full changes: v1.2.0...v1.3.0

TYDAL 1.2.0

Added

• user:create and org:create. Create a user (--email, optionally
  --org=SLUG --role=… or --superadmin; the password is asked hidden, or
  generated) and an organization (--name, --slug, --owner=EMAIL; the
  owner must exist, and defaults to the oldest platform admin). See
  CLI.md, "Users and organizations".
• A vault reports its language. /meta carries language, the ingest
  target collection's language (null for a vault with no ingest target), so a
  product can default to it (Full Frame's exhibition language).
  @tydal/client 1.6.0 types it.
• vault:reslug-files [--vault=SLUG] [--dry-run] brings existing file-link
  slugs to the current rule. Run it after upgrading, or after changing
  TYDAL_EXPORT_VISIBLE_FILENAMES.
• TYDAL_IMAGE_MAX_BYTES for @tydal/vault-mcp (default 716800, about
  700 KB). read_image always asks for the ai-prepared rendition within this
  budget, so a tool result fits Claude Desktop's 1 MB limit.

Changed

• Uploaded filenames no longer leave TYDAL. A vault's file links, the
  public JSON filename and download names used to come from the uploaded
  file's name, which can carry private details (a person's name, a date).
  Now a file's slug is a 3-character code derived from the file and the
  vault's salt, and downloads are named after the resource:
  {resource-slug}-{code}.ext, {resource-slug}-{code}-{variant}.ext for a
  rendition, {resource-slug}-preview.ext for a preview. Set
  TYDAL_EXPORT_VISIBLE_FILENAMES=true to keep the old behaviour. The
  resource agent view and @tydal/org-mcp's get_resource no longer return
  the file slug. See VAULT_SYSTEM.md §4.3.
• mcp:token keeps an existing member's role. Re-issuing a token used to
  reset the machine user to editor. Now a new member gets --role (default
  editor), and an existing one keeps their role unless --role is given,
  which changes it and says so. Editors may only change resources they own: an
  agent that edits other people's resources needs --role=admin.
• @tydal/org-mcp's update_resource_metadata answers briefly: the
  resource id, the fields updated with their saved values, and updated_at,
  instead of the whole resource.
• exhibitions:create asks for a new curator's password with the same
  hidden, confirmed prompt as user:create.
• Brand. The SPA's default theme is now TYDAL, in the logo's blues
  (primary #1A5D7D), with the new logo, the "ty" isotype as favicon, an
  apple-touch-icon and a 512 px icon. The tagline is "the typed Digital Asset
  Layer" throughout. SVG versions of the logos are in
  frontend/src/assets/brand/.
• @tydal/vault-mcp and @tydal/org-mcp are now 1.1.0.

Fixed

• Vault link hashes could clash across vaults. A link's hash must be
  unique across the installation, but each vault encodes with its own salt, so
  two vaults could produce the same hash. Creating the link then failed. The
  hash now grows by a character until it's free.

Upgrading

1. Add TYDAL_EXPORT_VISIBLE_FILENAMES=false to backend/.env (the
   templates have it, with a comment). Leaving it out means the same.
2. Run php artisan vault:reslug-files. Human file addresses change: a
   /v/{org}/{vault}/{resource}/{file} address shared before the upgrade ends
   in the old filename-based slug and stops resolving. Machine addresses
   (/h/{vault}/{link}) and resource addresses are unchanged.
3. MCP clients: add TYDAL_IMAGE_MAX_BYTES to the vault MCP's environment only
   if you need a budget other than the default (see mcp.example.json).

Full changes: v1.1.0...v1.2.0

TYDAL 1.1.0

Added

• Purpose-agnostic vault write ops ingest / update / withdraw.
  Every purpose's ingest now lands in the vault's ingest target, and
  gallery vaults gain all three next to activate/open/close (curator
  uploads for Full Frame). The payload is one flat metadata document:
  name and description go onto the resource, every other key into its
  metadata. update and withdraw only accept resources that vault's
  ingest created; the write-info probe lists them (ingested) and reports
  max_upload_bytes. See VAULT_WRITE_METHODS.md §3.
• exhibitions:setup / exhibitions:create: provision Full Frame photo
  exhibitions. setup runs once per organization (photo scheme, index and
  Photos collection); create runs once per exhibition (workspace, private
  gallery vault, its read + write keys, and optionally a curator account).
• POST /api/v1/auth/identify: an identity check for products that use
  TYDAL as their identity provider (Full Frame curators sign in with their
  TYDAL account). No token or session is issued; rate-limited per email and
  per IP.
• Workspaces chosen by name when creating a vault. The vault form picks
  the workspaces it reads from (workspace_ids on vault create/update), and
  the workspaces dialog manages the same links from the other side.
• tools/clients/fullframe.sh setup|create: a tier-aware front for
  exhibitions:setup / exhibitions:create, so provisioning Full Frame
  exhibitions no longer needs docker exec -w /var/www/html ….
• exhibitions:setup --language=CODE: the Photos collection gets the
  language you pick (asked interactively, default en), and a re-run with
  the option corrects it. When run through fullframe.sh, the closing hint
  suggests fullframe.sh create instead of php artisan exhibitions:create.
• exhibitions:create --curator-password=…: a new curator's password
  can be chosen instead of changed later. Interactive runs ask for it
  (hidden, confirmed; empty = generated). It's validated before anything is
  created, and an existing account's password is never overwritten.
• Quick reference: every everyday script and
  artisan command on one line, in docker and native form, with preconditions.
• Tools index: what goes in each tools/ category.
• The scheduler now runs in dev. Before, nothing ran schedule:work, so
  resource:prune, files:purge-uncommitted and resources:purge-drafts never
  fired. The docker tier gets a scheduler compose service (tydal_scheduler,
  toggled with app/queue by configure.sh). On the host tier start.sh runs
  schedule:work beside the queue. start.sh prints where the daemon runs and
  warns if the container is down. DEPLOYMENT.md gains the production cron entry.

Changed

• ai vault ingest payload. name and source_hash now travel inside
  the metadata JSON field instead of as top-level form fields, which are no
  longer read. @tydal/client ≥ 1.5.0 and @tydal/vault-mcp already send the
  new shape; raw multipart callers must update.
• Login and registration are throttled per IP (20 and 10 a minute), on top
  of the existing per-email limit on failed logins.
• tools/ reorganized by role. deploy/ keeps the stack lifecycle.
  clients/ holds what sits at TYDAL's border (vault-apps.sh, formerly
  deploy/clients.sh, plus fullframe.sh). dev/ holds seeding and smoke
  (seed-vault.sh, loadtest.sh), and lib/ the sourced tier.lib.sh. The
  root wrappers ./clients.sh and ./seed-vault.sh still work.
• tools/deploy/test.sh no longer resets the dev database. It used to run
  migrate:fresh --seed on tydal although Pest runs on tydal_test. It now
  creates tydal_test if missing, runs the client/org-mcp/vault-mcp
  suites (it pointed at the removed mcp/ workspace), runs every suite even
  after a failure, and accepts --backend-only and -- PEST_ARGS.

Fixed

• Uploads over 50 MB on the docker tier. The app image's nginx capped
  every request body at 50 MB, below PHP's limits. Its body limit now matches
  PHP_POST_MAX_SIZE (310 MB).
• seed-vault.sh works against the current API again. Resources are created
  with state: live, and the vault with organization_id, state: public and
  workspace_ids. The removed visibility / is_published fields made every
  resource create fail and the vault come out private.
• vault-apps.sh shields org-mcp/ and vault-mcp/ node_modules
  (it listed the removed mcp/).

Removed

• tools/deploy/populate.sh, tools/deploy/query.sh and tools/bulk_uploader/.
  They had broken against the current API: visibility instead of the required
  state, and calls to the removed extract/autotag endpoints. The uploader's
  accept/tags logic already lives in AutoApprovalService. Use
  tools/dev/seed-vault.sh for dev data (see tools/README.md).

TYDAL 1.0.0

TYDAL 1.0.0

First public release of TYDAL — the typed Digital Asset Layer.

The semantic layer between your files and your AI.
Schema-driven semantic vaults for AI agents.

From files to typed semantic resources.

TYDAL is a multi-tenant, schema-driven semantic knowledge system that structures, enriches, and projects digital resources through contextual Vaults.

Highlights

• Schema-driven resources with dynamic field contracts, validation, mappings, facets, and MIME rules.
• Contextual Vaults projecting Workspaces for delivery, galleries, Obsidian, AI, and mixed use cases.
• Semantic search and embeddings at resource and chunk level.
• Knowledge graph based on semantic tags and embedding similarity.
• AI-native Vault interface with grounded, citation-bearing Q&A.
• Two MCP surfaces for internal management and read-only Vault consumption.
• TypeScript client SDK shared across the SPA, MCP servers, and applications.
• Multi-tenant organization model with role- and permission-based access control.
• Elasticsearch indexing derived from the canonical MySQL data model.
• Docker-first development environment and deployment tooling.

This release establishes the public baseline for TYDAL. Internally, it incorporates the project's v2 architecture; as a publicly released product, this is TYDAL 1.0.0.

See CHANGELOG.md for the complete feature and change history.