OntoBricks — Release Notes V0.8.0

Release window: August – September 2026

Test status: all changes shipped with the non-scenario suite green: 5860 passed, 304 skipped, 6 deselected, 23 warnings.

Highlights

Ontology-only domains — a domain can now use No Backend, move through review and publication with an ontology but no materialized graph, and expose describe_ontology through MCP while graph-dependent UI and tools stay unavailable.
Per-domain MCP policy — each domain controls which graph tools it publishes and whether datasets, bridges, actions, and virtual attributes are normal, preferred, or disabled context.
Virtual attributes — ontology entities can declare values computed on demand by Unity Catalog functions. Values appear in Graph Explorer, REST, and MCP without entering mappings, R2RML, SHACL, cohorts, or graph materialization.
Data-source and mapping integrity — metadata refresh now shows column-level changes before applying them, source deletion lists affected mappings, and schema drift is visible in diagnostics and on the Mapping canvas.
Lakehouse view-only materialization — Lakehouse domains can expose a live pass-through view instead of copying mapped triples into a Delta table; analytics automatically creates and removes a temporary snapshot when required.
Safer graph operations — Builders can refresh frozen IN-REVIEW and PUBLISHED graphs, inspect and purge generated reasoning/cohort triples without touching mapped source triples, and rely on bounded Graph Chat reads.
Configurable product branding — administrators can change the application title, primary color, logo, and default entity icon with server-first rendering and immediate accessible preview.
Unified application experience — navigation, sidebars, tabs, buttons, Designer panels, responsive layouts, and page rhythm now use one documented Clarity design system.

Ontology and Mapping

Stronger metadata and mapping integrity

Removing or clearing a referenced data source now opens a detailed confirmation listing affected entity and relationship mappings.
Generated R2RML and SQL are invalidated when a referenced source disappears, preventing a stale build artifact from surviving the removal.
Update from UC computes added, removed, type-changed, and unchanged columns and presents the diff before the refreshed metadata is saved.
Mapping Diagnostics and the Designer surface advisory schema-drift warnings, including missing columns and affected relationship sides.
Aliased SQL projections are excluded from physical-column drift checks to prevent false positives, and warehouse checks run in the background without blocking the Designer.

Virtual attributes

Entity classes can bind a Unity Catalog scalar or table function as a group of virtual attributes.
Function output columns become ontology-visible attribute declarations; name collisions are resolved safely.
Graph Explorer computes values on demand, caches them for the page session, and isolates failures by function.
External REST and MCP expose declarations separately from computed values, preserving the distinction between “not computed” and a computed null.
A dedicated MCP compute_virtual_attributes tool complements describe_entity and get_entity_context.
Virtual attributes intentionally remain outside mappings, R2RML, SPARQL, GraphQL, cohorts, SHACL, and Data Quality materialization.

Reliable Designer persistence

Ontology and Mapping Designer edits auto-save when leaving their Design section or navigating away.
Entity and relationship panels no longer require Apply/Cancel actions; closing or switching context flushes pending edits.
The ontology load path now waits for authoritative ontology state before rebuilding the canvas, preventing stale design-layout copies from overwriting saved changes.
Design views are reconciled with the ontology, and orphaned datatype-property mirrors are pruned so deleted attributes do not reappear.
Relationship direction updates immediately across the Ontology Designer, ontology map, and Mapping Designer.
R2RML exports now emit standards-compliant bare column names while retaining compatibility with older quoted exports.
The Manual Mapping panel correctly owns its shared detail pane and no longer leaves previews spinning indefinitely.

Auto-Mapping observability

The Auto-Map progress overlay shows the agent's live ordered activity without adding a separate streaming endpoint.
Cancel is cooperative: no new agent chunk starts after cancellation, while completed chunks remain saved.
Every completed, failed, crashed, or cancelled run records a durable execution report in the Domain Audit Trail.

Knowledge Graph and Backends

Ontology-only and per-domain backend choice

No Backend joins Lakehouse, Lakebase, and Neo4j as a first-class domain type.
Mapping and Knowledge Graph navigation is disabled for ontology-only domains, including direct-route protection.
New domains default to Lakehouse and explain all backend choices; backend selection remains editable under Domain → Information.
Backend selection is presented as accessible branded cards and is natively frozen for viewers, frozen versions, and edit-locked sessions.

Lakehouse view-only materialization

Lakehouse domains can choose:
- TABLE — materialized copy for a build-time snapshot.
- VIEW — no data copy for live reads over the mapping gateway.
The Build page, health summary, run records, and domain payloads report the active materialization mode.
Switching modes safely replaces the previous relation type.
Graph Analytics uses a disposable Delta snapshot for view-only domains and always removes it after the run.
Lakebase and Neo4j continue to resolve to table materialization for their UC analytics snapshot.

Generated inference management

Builders and administrators can inspect the combined count of materialized reasoning and cohort triples.
Purge Inferences is available from Lakehouse/Lakebase Build, Inference, and Cohorts.
Purging truncates only the generated companion; mapped source triples remain untouched.
Unsupported backends, including Neo4j, reject the operation safely.

Frozen-version refresh

Builders can rebuild graph data and materialize reasoning for IN-REVIEW and PUBLISHED versions without reopening ontology or mapping edits.
The UI skips domain-design persistence for frozen versions while retaining viewer and edit-lock enforcement.

Graph reliability

Lakebase and SQL Warehouse reads have scoped statement timeouts.
Lakebase authentication, Settings discovery, and provisioning follow every Postgres project-list page, so projects beyond the API's first page resolve correctly.
Graph Chat result size is bounded and reports blocking-pool pressure.
Analytics no longer displays historical charts as if they described a graph whose live objects were deleted; run history remains available under Management → Runs.

Registry, Workflow, and Permissions

DRAFT versions can enter review when they contain either a successful graph build or a valid ontology; truly empty versions remain blocked.
Published domain listings expose has_graph, enabling ontology-only MCP selection without advertising graph tools.
Databricks group grants now participate in application and domain role resolution using the current user's SCIM membership; the most privileged direct or group match wins.
Editors and Builders can discover Unity Catalog catalogs and schemas for Add Data Source without requiring application CAN_MANAGE; shared Settings writes remain administrator-only.
Registry configuration now prefers an explicit environment volume over the seeded session default.
Starting New Domain first resolves the currently loaded domain's save/discard/cancel flow, and the creation fields remain editable even if the previous domain was read-only.

External API and MCP

Domain-aware MCP publication

Domain Information adds an MCP tab with four always-on registry tools and configurable domain tools.
Tool visibility resets when selecting a domain and is enforced again at call time, preventing cached clients from invoking disabled tools.
Datasets, bridges, actions, and virtual attributes can be Preferred, Normal, or Disabled.
Cross-domain bridges include the target-domain description and direct the model to select_domain for a real hop; unavailable MCP targets are filtered from external context.

Ontology-only publication

describe_ontology returns ontology classes and OWL content without requiring a graph.
Graph tools are hidden and refused for a domain whose latest PUBLISHED version has no graph.
Per-domain policy can still disable describe_ontology.

REST, GraphQL, and Swagger

GET /api/v1/domains exposes normalized graph_backend, has_graph, and mcp_policy.
POST /api/v1/domain/info includes normalized backend information and typed file/statistics response models.
GraphQL domain discovery lists only materialized graph domains.
Settings → Developer → API now selects only API-visible domains and PUBLISHED versions, shows backend readiness, and disables graph-dependent Try-it actions when no graph exists.
Swagger/ReDoc preserve descriptive tags, support contact, license metadata, and accurate typed response schemas.
External endpoints expose virtual-attribute declarations and on-demand computation.

UI and Settings

Configurable UI branding

Settings → Configuration → UI manages application title, primary color, application logo, and default entity icon.
Branding is stored atomically in registry global configuration with compatibility for the legacy navbar logo.
Server-first rendering applies title, favicon, logo, and derived CSS tokens on the initial HTML response, avoiding a default-theme flash.
Settings provides immediate preview, validation, accessible swatches, Reset icon, Default, Save, and Discard actions with an unsaved-change guard.
GraphiQL, Help Center, Ontology Assistant, navbar state, and graph highlights use the configured brand.

Clarity design-system refresh

The sidebar is a framed card, user identity moved to the navbar, and level-two workspace navigation became a compact segmented control.
Page and panel tabs use a shared underline rail and density model; card-integrated tabs are now the default page pattern.
Ontology and Mapping Designers use persistent two-card split layouts, manual resize handles, placeholders, and optional session-persisted dotted grids.
Shared button focus, hierarchy, grouping, responsive wrapping, and tokenized Graph Chat controls were standardized.
Sidebar content roots, scrolling ownership, mobile natural flow, and bottom alignment were normalized across Domain, Ontology, Mapping, Knowledge Graph, and Settings.

Security and Reliability

Dependency floors were raised for MLflow, GitPython, aiohttp, and sqlparse, and the lockfile remains public-index deploy safe.
The remaining cryptography advisory cannot yet be raised to its patched major version because MLflow constrains cryptography<50; the constraint is documented for follow-up.
GraphiQL branding output is escaped and serialized safely to prevent stored-content injection.
Settings branding endpoints require the administrator role at both route and service layers.
Session persistence guards prevent concurrent unload writers and stale design-layout data from reverting ontology edits.
Read-only MCP policy controls are disabled by stable classes rather than tool-name-derived IDs.
Project rules now require English for comments, docstrings, changelogs, logs, audit-trail text, and traces.

Deployment and Documentation

Lakebase bootstrap installs or relocates pgcrypto in public, verifies digest(), and documents that extensions are database-scoped.
Deploy preflight detects missing or misplaced pgcrypto.
Optional Neo4j secrets are no longer forced into the DAB deployment; connections remain configured through Settings → Neo4j.
Instance-suffixed MCP application names are resolved consistently, and first-deploy Unity Catalog grants run before the Lakebase schema-existence guard.
The 0.7 → 0.8 Lakebase migration adds the persisted mcp_policy column with bootstrap and lazy self-healing support.
Documentation now covers UI branding, ontology-only domains, MCP domain policy, virtual attributes, graph limits, Lakehouse materialization, inference purging, and the Unity Catalog HTTP connection used to attach OntoBricks MCP to Genie.
Generated Sphinx HTML is no longer tracked.
The retired GitHub Pages copy was removed; https://ontobricks.org/ is the canonical product site.

Bug Fixes (selected)

Fixed table-comment and schema probes using unsupported %s placeholders instead of Databricks SQL qmark bindings.
Fixed missing MCP service-principal grants on first instance-suffixed deployment.
Fixed graph-limit fields loading blank and potentially clearing stored overrides on Settings save.
Fixed relationship direction changes producing duplicate-name errors or stale canvas arrows.
Fixed Mapping Designer schema-drift checks repeatedly querying the warehouse.
Fixed stale Analytics results appearing after the graph objects were deleted.
Fixed New Domain fields being disabled by the previously loaded domain's view-only state.
Fixed ontology attributes, relationships, inheritance, and other structural edits being restored from stale mirrors after navigation.
Fixed Lakehouse Build storage labels and a JavaScript scope error on populated graphs.
Fixed mobile Graph Chat, Data Quality, and Logs panes blocking natural page scrolling.
Fixed Lakebase Autoscaling endpoint resolution failing when the bound project appeared beyond the first Postgres API page.

September follow-up (still v0.8.0)

Work that landed after the original August notes and remains on the 0.8.0 line:

Analytics ranking chart — one exhaustive ordered metric series with server-side LTTB sampling above 5,000 nodes; distribution histograms default to a logarithmic count axis.
Lakehouse Health — Settings reports Unity Catalog schema permission readiness for the app principal without blocking on domain graph assets.
Mapping Designer — unchanged entity previews are reused in memory; closing an inspected-but-unedited panel does not save or rebuild the canvas.
MCP server packaging — src/mcp-server/server/ is split by topic (tools, resources, session, HTTP client). Import paths and the uvicorn entry point are unchanged.
Unity Catalog identifiers — catalog, schema, and table names may start with a digit (5_g_subscribers).
Ontology Designer — metadata-only saves refresh visible map elements without a full rebuild.
GET /api/v1/digitaltwin/triples — no longer raises NameError; CI enforces repository-wide Ruff F821.
Lakebase bootstrap — credential minting failures from lakebase-perms.sh are observable instead of exiting silently.
Deployment docs — explicit 0.7 → 0.8 upgrade path, including deploy.config.sh backup.

Lakehouse//RT query warehouses, the Build vs Query split, and CloudFetch control belong to v0.9.0, not this line.

Upgrade Notes

Registry migration

Run the standard Lakebase bootstrap or the idempotent scripts/migrations/upgrade_0.7_to_0.8.sql migration so the registry domains table contains mcp_policy jsonb NOT NULL DEFAULT '{}'::jsonb. Deploy preflight validates the migration asset and expected column.

Existing domains require no policy backfill: an empty policy preserves the pre-v0.8 MCP tool and context behavior.

Graph backend behavior

Existing domains with no stored backend continue to normalize to Lakebase.
Newly created domains default to Lakehouse.
Choose No Backend only for ontology-only publication; Mapping, Knowledge Graph, graph API operations, and graph MCP tools are intentionally unavailable.
Rebuild a Lakehouse domain after changing TABLE/VIEW materialization so the target relation matches the selected mode.

Lakebase `pgcrypto`

pgcrypto must be installed in every Lakebase database used by the graph store, not only the registry database, and should live in the public schema so digest() is resolvable. Run make bootstrap-lakebase or the documented bootstrap script with the correct graph database coordinates.

Neo4j configuration

The Asset Bundle no longer binds an optional neo4j-password secret. Configure named Neo4j connections in Settings → Neo4j using Databricks secret scope/key references.

MCP publication

Review each domain's new MCP tab after upgrade if you want a reduced tool set or preferred/disabled context. Ontology-only domains automatically force graph tools off. Re-selecting a domain refreshes the visible tool set for the MCP session.

Branding

Application branding is instance-wide and stored in registry global_config. Administrators should use Settings → Configuration → UI; Default restores factory branding, while Reset icon changes only the draft icon until Save.

v0.8.0