v1.3.3

OWL 2 DL reasoning via WebAssembly

• Konclude is now the default reasoning backend. Konclude runs entirely client-side as a WebAssembly worker — full OWL 2 DL tableau: equivalentClass axioms, someValuesFrom restrictions, ABox individual classification. N3 rule-based reasoning remains available as an alternative backend.
• Fixed single-pass realization sequence in the Konclude worker (previously a mid-sequence buffer drain caused realization to return empty results).
• Fixed SharedArrayBuffer availability in cross-origin recording/automation contexts (--enable-features=SharedArrayBuffer).
• Blank-node skolemization display fixes; blank nodes now render correctly on canvas.

MCP tool improvements

• addNode now accepts typeIris (array) — assign multiple rdf:type values atomically in a single batch, e.g. typeIris: ["owl:NamedIndividual", "ex:SalamiTopping"]. The scalar typeIri is still supported.
• Schema validation wrapper — missing required parameters now return a helpful error with rename hints (e.g. "you passed startIri — rename to fromIri").
• suggestOntologiesForTask returns full pack details on no-match to help the model recover.
• Added BFO 2020 atemporal profile (bfo2020-nt) to the engineering ontology pack.
• Removed deprecated subjectIri alias from addNode.

Hardened AI relay

• Relay injection now uses execCommand('insertText') — the only correct path for TipTap/ProseMirror-based inputs. Fixes UUID leaks that occurred with setContent and paste approaches.
• ChatGPT / plain ProseMirror injection fixed.
• Injection delay configurable via live slider (0–10 s) in relay.html, persisted to localStorage.
• Relay tested and working across three chat frontends: OpenWebUI, ChatGPT, and one internal deployment.
• isAiStreaming now detects streaming state via the OWUI stop-button; prevents premature idle detection during model chain-of-thought.

Canvas & UI

• Labels now prefer rdfs:label over prefixed IRI across all canvas surfaces (node cards, type chips, link labels, autocomplete).
• Scroll overflow fix on the loaded ontologies list in the top bar.
• Type chips wrap correctly on expanded nodes.
• Clear Data now clears both TBox and ABox views.

Full Changelog: https://github.com/ThHanke/ontosphere/commits/v1.3.3

Relay rewrite + Reasoning improvements

Relay — OpenWebUI

UUID leak fix
Injecting tool results into OpenWebUI no longer causes the model to echo
the OWUI message UUID ([RESPONSE] <uuid>) in its response. Root cause:
pasteText/setContent bypassed the browser's DOM event pipeline entirely;
OWUI's reactive layer detected the programmatic change and leaked internal
message metadata into the outgoing request. Fixed by switching to
document.execCommand('insertText') — the same native event chain as real
user typing. Relay now works flawlessly in both OpenWebUI.

Result format
Injected messages restored to structured form:

Full Changelog: v1.3.1...v1.3.2

Fixed: human-readable labels across all canvas surfaces

• Node type chips — show rdfs:label when available (e.g. "Material", "portion of silicon"),
  falling back to prefixed IRI, then full IRI. Hover shows prefixed <full-iri>.
• Edge labels — show prefixed IRI (e.g. pmdco:PMD_0000077) instead of full IRI.
  Labels are patched after ontologies finish loading so namespace resolution is correct.
• Edge hover title — prefixed <full-iri> for full context on demand.
• Dialog inputs — EntityAutoComplete (RelationEditor, NodeEditor) displays
  prefixed · label when both are available.

Full Changelog: v1.3.0...v1.3.1

DL reasoning and relay hardening

What's new in v1.3

OWL 2 DL reasoning via Konclude

Ontosphere now ships with Konclude as the default reasoner — to our
knowledge the only OWL 2 DL-capable reasoner available in the browser.
Konclude runs entirely client-side via WebAssembly and handles the full
OWL 2 DL profile: transitive class hierarchies, property chains, qualified
cardinality restrictions, and ABox individual classification — beyond what
OWL-RL rule sets can express.

The previous N3-rules backend remains available for custom rule authoring.
Switch between them in Settings → Reasoning.

Under the hood the integration moved to the v0.2.0 zero-copy binary wire
protocol (rdf-reasoner-konclude), eliminating the NTriples string
round-trip and the associated overhead.

Hardened AI relay (OWUI)

The OpenWebUI relay that drives Ontosphere from a chat model received
a major stability overhaul:

• Fire-on-idle dispatch — replaced the fragile MutationObserver with
  a 500 ms idle poll; tool calls are extracted and dispatched only once
  the model has fully stopped streaming
• Single injection path — one deterministic code path for injecting
  results back into the chat (TipTap setContent + transaction event);
  removed the fallback chain that caused UUID-prefix and double-submit bugs
• Multi-burst flushing — relay waits for a 45 s stability window before
  firing the next turn, preventing sends while Konclude classification is
  still in progress
• Streaming detection — content-length growth + relay queue state
  instead of the unreliable send-button signal

v1.2.0

New features

• Manchester Pizza Tutorial demo — full OWL pizza ontology built step-by-step via MCP tools: classes, disjointWith axioms, subclass hierarchies, object properties with domain/range, inverseOf, FunctionalProperty, ABox individuals, and OWL-RL reasoning. Available as both a seed-driven doc (docs/mcp-demo/pizza-tutorial.md) and a side-by-side AI tutor chat video (pizza-tutorial-chat).
• Side-by-side chat demo format — new demo-stage.html layout with mock AI chat on the left and live Ontosphere canvas on the right. Scripted via addChatMessage() / callToolOnStage() — no relay bookmarklet needed.
• Demo video pipeline — npm run demo:video records all demos at 1920×1080 (H.264 High/CRF 20) using xvfb-run; npm run demo:all regenerates all seed-driven markdown docs and auto-updates the README demo table with the latest SVG snapshot.
• MCP server renamed — visgraphMcpServer → ontosphereMcpServer; all internal and external references updated.

Improvements

• Caption overlay in demo recordings now appears after reasoning results are visible on canvas (captionAfter option in runSeedTurn).
• Chat stream scrolls on every word during AI message streaming — last message no longer hidden behind input bar.
• Anchor tag color uses --primary token instead of browser-default blue, consistent across light and dark mode.
• demo:all auto-patches README demo table with the last SVG produced by each seed run.

Fixes

• Video recording resolution was 800×450 (Playwright default) — fixed to explicit 1920×1080.
• demo-bootstrap.mjs pointed at renamed MCP server file.
• All VisGraph/visgraph branding references replaced with Ontosphere across source, scripts, and docs.

Metadata

• CITATION.cff updated: title, version, repository URL, and concept DOI (10.5281/zenodo.19605270).
• package.json name updated to ontosphere.

Full Changelog: v1.1.0...v1.2.0

AI Integration

What's new in v1.1.0

MCP Tool Surface

VisGraph now exposes a full Model Context Protocol tool surface, letting AI agents control the graph directly — add/remove nodes and edges, run SPARQL queries, validate with SHACL, trigger OWL-RL reasoning, cluster and layout nodes, and inspect the canvas state. No backend required.

AI Relay Bridge

A new bookmarklet-based relay bridge connects any chat UI (ChatGPT, Open WebUI, Gemini, …) to VisGraph via JSON-RPC 2.0 over BroadcastChannel. The AI issues tool calls as JSON-RPC messages; results are injected back into the chat automatically.

Demo Seeds

New built-in demo seeds showcase real-world usage: FOAF social graph, employment reasoning, and scene-ontology with BFO/RO.

Docs

• README rewritten with role-based entry points, live demo previews, and dynamic sidebar TOC
• AGENTS.md for AI integrators and MCP clients
• Public mcp.json manifest at /.well-known/mcp.json

Full Changelog: v1.0.0...v1.1.0

first release

What's Changed

• Convert/oklch theme by @ThHanke in #1
• Convert/oklch theme by @ThHanke in #2
• fixed dependencys by @ThHanke in #3
• Reactodia include by @ThHanke in #4

New Contributors

• @ThHanke made their first contribution in #1

Full Changelog: https://github.com/ThHanke/visgraph/commits/v1.0.0