Release notes¶
Here you can find the release notes for Kanzi AI Assistant.
1.5.0¶
Kanzi AI Assistant now works with Kanzi 4.1.1.
Kanzi AI Assistant can now see the rendered result of its work. The
capture_viewporttool captures the Kanzi Studio preview as an image — rendered offscreen, so no live preview window is required — andverify_renderhas the configured model describe what is actually on screen. Together they give Kanzi AI Assistant a run-and-see loop that confirms visual results such as shadows, materials, colors, and layout, instead of assuming a successful tool call produced the right pixels.Added plan mode with an approval gate. For complex requests, Kanzi AI Assistant explores the project, designs a complete step-by-step solution, and submits it for your approval through
exit_plan_modebefore making any changes — so you review the intended steps first and nothing is modified until you approve.Redesigned the Settings window. Settings are now organized into a category tree — AI Provider, Knowledge Base, Web Search, MCP Server, and General — with a quick-search box that filters settings as you type.
Added web search (opt-in). The
web_searchtool lets Kanzi AI Assistant look up current information on the live web for questions beyond the curated knowledge base, such as recent release notes or library changes. Web search is off by default; enable it and add a web-search API key under Settings > Web Search.Added a Blender-to-Kanzi import pipeline (opt-in).
blender_export_scenelaunches Blender headlessly to bake an HDRI environment, render a background, and extract lights and Principled-BSDF materials from a.blendfile into a bundle, andimport_blender_exportbrings that bundle — geometry, materials, lights, camera, environment, and background — into your project. It requires a local Blender installation, which Kanzi AI Assistant detects automatically from yourPATHor a standard install location.Added image import. The
import_imagetool imports a standard image (PNG, JPEG, TGA, BMP) as a texture, or an HDR image (.hdr,.exr) as image-based-lighting cubemaps, optionally including an environment background for image-based lighting.Improved the built-in knowledge base. Retrieval now uses semantic search with source citations and session-aware querying, and the corpus gained new render-effect recipes.
move_nodeandmove_resourcegained ordering control. In addition to reparenting, Kanzi AI Assistant can now reorder an item within its parent — to the top or bottom, up or down by a number of steps, before or after a named sibling, or at an explicit index. This also lets Kanzi AI Assistant reorder render passes within a render pass prefab.Added the
sync_material_uniformstool, which synchronizes a material type’s properties with the uniforms declared in its shaders — the equivalent of the Kanzi Studio Sync with Uniforms command. Use it to repair property/uniform drift, for example after a preprocessor macro exposes a new uniform.Added project validation tools (
get_item_validation,list_invalid_items,validate_render_pipeline) that surface Kanzi’s per-item validity, so Kanzi AI Assistant can detect invalid nodes and resources, read why each item is invalid, list every invalid node in the scene (optionally scoped to a subtree), and validate a render-pass pipeline by viewport or prefab — to diagnose and explain problems.Added comparison and render-pass inspection tools.
compare_nodesandcompare_render_pass_prefabsdiff two items and report what exists only in one, only in the other, or in both, andinspect_render_pass_pipelineexamines a project’s render-pass pipeline so Kanzi AI Assistant extends the existing pipeline instead of rebuilding it.Fixed Kanzi AI Assistant being unable to work with State Objects — the per-node children of a State (State Manager → State Group → State → State Object) that hold its property overrides. State Object is now a first-class resource type:
create_resourceanddelete_resourcehandle it, and the tools that resolve a resource by type (move_resource,duplicate_resource, the property tools,find_resource_usages, and validation) work with it. Previously it could not be created and those tools rejected it as an unknown type.
1.4.0¶
Kanzi AI Assistant now works with Kanzi 4.1.0.
Added Shader Graph authoring. With the experimental Shader Graph Editor enabled, Kanzi AI Assistant can create and edit shader graphs, including custom GLSL Code nodes, with built-in recipes for emissive, fresnel, UV scrolling, glass, vignette, subsurface scattering, FBM noise, and Beckmann effects.
Shader-graph write rejections now return per-node/per-edge error codes and remediation suggestions, so Kanzi AI Assistant can self-correct without bouncing the failure to the user.
Node-path errors now include nearest-match suggestions so Kanzi AI Assistant can self-correct typos and stale paths.
Property tools now reject unknown property names pre-flight, so Kanzi AI Assistant can retry with the correct name without wasting a turn.
MCP tool definitions now carry behaviour hints (
readOnlyHint,destructiveHint,idempotentHint) so external clients can gate auto-approval natively. Batchability and indicative cost are exposed under_metawith therightware.com/vendor prefix.Breaking change for external MCP clients: every
tools/callresponse is now a JSON envelope{ok, data, error, meta}. Clients reading raw text fromcontent[0].textneed to parse it as JSON and readdataon success orerrorwhenokis false.Consolidated 32 property tools into six unified ones (
get_property,set_property,add_property,remove_property,list_propertieswith atarget_kindargument;create_propertywith atypeargument) to give Kanzi AI Assistant fewer near-synonyms to pick between. The legacy per-kind tools remain functional but are flaggeddeprecatedin the tool catalog.Added batch tools (
batch_set_properties,batch_remove_properties,batch_create_nodes) so Kanzi AI Assistant can change several properties on one target, or create several children under one parent, in a single tool call.
1.3.0¶
Kanzi AI Assistant now works with Kanzi 4.1.0-alpha2.
Added smart 3D asset import. Drag a
.glb,.gltf, or.fbxfile onto the chat window to import it. Kanzi AI Assistant auto-creates the scene placeholder, adds a default light and camera if missing, and can upgrade imported Phong materials to physically based materials with named presets through theupgrade_materials_to_pbrandset_material_appearancetools.Added move and duplicate tools for nodes and resources (
move_node,duplicate_node,move_resource,duplicate_resource), preserving properties, bindings, and child nodes.Added reverse resource usage lookup through the
find_resource_usagestool, listing every reference to a given resource across scene nodes, prefab nodes, and resource libraries.Added a curated knowledge base with task recipes, domain concepts, patterns, common errors, and best practices, queried by Kanzi AI Assistant through the
query_knowledgetool.Added ONNX-based semantic retrieval (BGE-small-en-v1.5 embedding model). This is now the default knowledge retriever. The earlier BM25 keyword retriever is still selectable in Settings.
Added planned agent mode. For complex requests, Kanzi AI Assistant can submit a step-by-step plan through the
submit_plantool. The plan is shown as a checklist in chat and updated as Kanzi AI Assistant progresses.Added a tool result store with handle-based retrieval. Large tool results are offloaded and Kanzi AI Assistant retrieves them on demand using the
read_stored_resultandsearch_stored_resultmeta-tools.Grouped every Kanzi AI Assistant turn into a single undo entry, so undo reverts the whole turn at once.
Fixed MCP transformation parameters silently dropping to identity when the input omitted a rotation or scale component.
Updated bundled dependencies to address CVE-2026-26127.
1.2.0¶
Initial public release of Kanzi AI Assistant.
Kanzi AI Assistant works with Kanzi 4.0.0.