SmartHopper 1.2.2-alpha: Enhanced Performance and Stability [Patch]

• Component Stability: Promoted StatefulComponentBaseV2 to default stateful base, ensuring more reliable component behavior
• Chat Performance: Optimized DOM updates with keyed queuing, template caching, and LRU diffing to reduce redundant work on large chats
• Streaming Visuals: Removed heavy animations and switched to lighter wipe-in effects for better responsiveness during message streaming
• UI Freeze Prevention: Reduced WebChat dialog freezes while dragging/resizing by throttling DOM updates and processing in smaller batches
• Token Optimization: Simplified CanvasButton assistant system prompt to reference instruction tools instead of embedding long guidelines
• Selection Context: Enhanced selection counting with Rhino UI thread reading and robust Attributes.Selected fallback
• Tool Instructions: Added instruction_get tool to provide detailed operational guidance on demand
• Memory Efficiency: Reduced idempotency cache size from 1000 to 100 entries to lower memory footprint
• DeepSeek Fixes: Resolved deepseek-reasoner model failures and duplicated reasoning display issues
• Message Display: Fixed user messages not appearing in chat UI and improved turn-based metrics aggregation

---

Full Changelog: 1.2.1-alpha...1.2.2-alpha

SmartHopper 1.2.1-alpha: Runtime data snapshots & script canvas wrappers [Patch]

This patch release focuses on richer inspection of Grasshopper definitions (including runtime data snapshots), smoother script component workflows, and small but important chat and UI fixes.

Detailed list of changes

• Added new script wrapper tool script_generate_and_place_on_canvas that generates a Grasshopper script component and places it on the canvas in a single call, simplifying chat workflows and reducing token usage.
• Updated script tools so the low-level script_generate / script_edit tools are hidden from chat (still available via AIScript components) while wrapper tools handle full generate-and-place / edit-and-replace flows.
• Changed script generation behavior so script_generate no longer returns a pre-placement instanceGuid; instead, wrapper tools now report the real instanceGuid assigned by Grasshopper after placement.
• Enhanced gh_get to optionally include snapshots of runtime/volatile data (branch counts, item counts, and sample values) for inputs and outputs, enabling richer debugging and inspection.
• Added specialized gh_get variants such as gh_get_selected_with_data, gh_get_by_guid_with_data, and gh_get_errors_with_data to retrieve focused subsets of the document together with their runtime data.
• Improved gh_put tool results by returning an instanceGuids array with the actual GUIDs of components placed on the canvas, making follow-up queries and wiring much easier.
• Relaxed GhJSON validation so documents without any connections are treated as valid (missing connections is now considered an empty list instead of an error).
• Fixed a chat UI bug where sending the same user message twice could cause one message to be collapsed; each user message now gets a unique turn ID and shows correctly in the conversation.
• Improved conversation session turn handling so tool results inherit the correct turn ID from their tool calls, leading to accurate per-turn metrics aggregation in the UI.
• Refined dialog positioning so StyledMessageDialog windows open centered on the Rhino/Eto main window for a more predictable and polished user experience.

---

What's Changed

• feat(tools): add runtime data getters and script canvas placement wrappers by @marc-romu in #359

Full Changelog: 1.2.0-alpha...1.2.1-alpha

SmartHopper 1.2.0-alpha: Script Components Unleashed

This alpha release focuses on powerful AI-assisted scripting workflows, GhJSON merge/edit tools, and clearer on-canvas guidance—so you can create, review, and refine Grasshopper scripts in seconds.

🧠 AI-powered scripting workflows

Create powerful scripts in seconds. Let AI write, review, and refine your code following your instructions.

• New GhJSON-based script tools:
  • script_generate and script_edit now operate directly on GhJSON for Grasshopper script components.
  • script_edit_and_replace_on_canvas combines editing and placement in a single tool call, reducing token usage and simplifying workflows.
• Rich parameter and modifier support:
  • Full control over dataMapping (Flatten/Graft), reverse, simplify, invert, isPrincipal, required, and expression for inputs.
  • Output parameters gain matching support for dataMapping, reverse, simplify, and invert.
• Smarter script validation:
  • ScriptCodeValidator flags non-RhinoCommon geometry libraries (e.g. System.Numerics, UnityEngine, numpy, shapely) and helps the AI auto-correct them.
  • script_generate and script_edit include a validation retry loop that detects invalid geometry patterns and re-prompts the AI with correction instructions.
  • System prompts are enriched with RhinoCommon-specific and language-specific (Python/IronPython/C#/VB.NET) guidance, imports, and type mappings.
• Updated script components:
  • AIScriptGeneratorComponent now orchestrates script_generate / script_edit with gh_get / gh_put, exposing GhJSON, Guid, Summary, and Message outputs for better inspection.
  • AIScriptGeneratorComponent and AIScriptReviewComponent no longer require a Guid input—the target component is selected with the on-canvas button.
  • Both components now support processing multiple inputs in parallel.
  • The legacy script_fix tool has been renamed to script_review to better reflect its review-first behavior.

🧩 GhJSON merge & document tooling

• New GhJSON helpers and merge workflows:
  • GhJsonHelpers simplifies applying pivots and restoring InstanceGuids on deserialized components.
  • GhJsonMerger merges two GhJSON GrasshopperDocument instances with automatic ID remapping and conflict resolution.
  • New gh_merge AI tool returns merged GhJSON plus merge statistics (components / connections / groups added or deduplicated).
  • New GhMergeComponents Grasshopper component (“Merge GhJSON”) lets you merge two GhJSON documents directly on canvas.
• Smarter document extraction:
  • gh_get gains a categoryFilter parameter so you can focus AI workflows on specific component or object categories.

🎛️ Canvas UX & component UI improvements

• Dialog-to-canvas link visualization:
  • Dialogs such as script editors can now visually link to a Grasshopper component using a bezier curve and anchor dot, mirroring the native script editor anchor experience.
  • StyledMessageDialog (ShowInfo, ShowWarning, ShowError, ShowConfirmation) accepts optional linkedInstanceGuid and linkLineColor parameters to enable this linking.
• Enhanced component badges and selection UI:
  • AI-selecting stateful components now show a combined UI: the “Select” button plus provider/model badges in a single, clear block.
  • Selecting components use the dialog link line color for hover highlights and draw a connector line from the selection center to the “Select” button.
• Canvas and visual polish:
  • The Canvas button prompt has been improved to guide users toward in-viewport scripting workflows and avoid unnecessary external code blocks or testing patterns.
  • Component icons have been refreshed, including outlined variants for McNeel forum and script components, and a new ghmerge icon with updated ghget / ghput icons aligned to the merge workflows.

🧬 Model guidance & provider updates

• Model compatibility badges:
  • New “Not Recommended” badge (orange octagon with exclamation mark) appears when a chosen model is discouraged for the tools used by a component.
  • AIModelCapabilities now exposes DiscouragedForTools, and AIStatefulAsyncComponentBase exposes UsingAiTools and EffectiveRequiredCapability to drive badge logic.
  • The “Not Recommended” badge suppresses the “Verified” badge so you always see the most important status first.
• Provider improvements:
  • Anthropic provider adds the Claude Opus 4.5 model to the registry.
  • OpenRouter gains structured-output support for JsonOutput requests and now populates finish_reason and model in metrics for chat completions.

🐞 Stability & quality-of-life fixes

• Chat and UI reliability:
  • Chat UI now displays accurate, aggregated token metrics per turn (from user message to next user message), including tool calls and tool results.
  • ConversationSession exposes GetTurnMetrics(turnId) to aggregate metrics across interactions in a turn.
  • Tool results now inherit the correct TurnId, ensuring proper grouping and reporting.
  • Web chat dialogs are created as owned tool windows of the Rhino main Eto window, stay on top of Rhino while active, and behave correctly when confirmation dialogs appear.
• GhJSON and scripting robustness:
  • gh_put no longer enters infinite loops when replacement mode is active and no longer hits “object expired during solution” errors; replacement now uses IsolateObject() for safe cleanup and undo support.
  • GhJSON-based script tools have updated schemas requiring nickname, preventing crashes with OpenAI structured output mode.
• Additional minor fixes and refinements:
  • Numerous edge cases in script and GhJSON workflows have been smoothed out to make multi-step AI flows more predictable and resilient.

🛠️ Technical Requirements

• Rhino 8.19 or above is required
• Windows 10/11 (macOS has not been tested)
• Valid API keys for MistralAI, OpenAI or DeepSeek

⚠️ Important Notes

• This is an alpha release with some features still unstable or subject to change
• API keys are required, and usage costs apply based on your provider
• Documentation is currently under development

🤝 We Value Your Feedback!

Help shape SmartHopper's future by:

• Sharing your experiences with the new scripting workflows and GhJSON tools
• Suggesting improvements via our discussion
• Telling us what AI capabilities would help your workflow most

We hope you enjoy these new features and improvements!

Happy designing! 🎨

---

What's Changed

• feat(script-tools): add GhJSON-based script tools, merge workflows, and message-boxes to canvas linking by @marc-romu in #354
• feat(scripting): enhance script tools, model compatibility, and docs by @marc-romu in #355

Full Changelog: 1.1.1-alpha...1.2.0-alpha

SmartHopper 1.1.1-alpha: Provider model updates [Patch]

This patch focuses on AI provider registry and capability alignment, bringing clearer model identifiers and safer defaults across Anthropic and OpenAI.

Detailed list of changes:

• Anthropic:

  • Added explicit dated model identifiers for Claude 3.x/4.x while keeping shorthand names.
  • Switched default text/tool/json model to claude-haiku-4-5 for faster, cost‑effective defaults.
  • Enabled structured‑output (beta) for Sonnet 4.5 and Opus 4.1, and restricted Text2Json capability to those models for reliability.

• OpenAI:

  • Updated model registry to include the GPT‑5.1 series.

Full Changelog: 1.1.0-alpha...1.1.1-alpha

SmartHopper 1.1.0-alpha: Smarter GhJSON & knowledge tools

This alpha release focuses on a modern GhJSON/property system and powerful knowledge tools for McNeel forum and web—plus foundational scripting improvements that prepare SmartHopper for richer script workflows in future releases.

🌐 Knowledge tools for McNeel forum, web & Rhino files

• McNeel forum AI tools:
  • mcneel_forum_search with configurable result limits and raw JSON post payloads.
  • mcneel_forum_post_get (renamed from web_rhino_forum_read_post) to fetch full posts by ID.
  • mcneel_forum_post/topic_summarize to generate AI‑powered summaries of forum threads.
• web_generic_page_read now returns clean, context‑aware text for:
  • Wikipedia/Wikimedia, Discourse forums, GitHub/GitLab file URLs, and Stack Exchange questions via official APIs.

📦 Modern GhJSON & Property Management V2

• New PropertyManagerV2 architecture with:
  • Clear separation of filtering, extraction, and application.
  • PropertyFilter, PropertyHandlers, PropertyFilterConfig, SerializationContext, and ComponentCategory for precise, scenario‑aware property control.
• GhJSON optimization:
  • Groups only include members that are part of the current selection.
  • Runtime‑only properties (VolatileData, IsValid, ReferenceID, IsGeometryLoaded, etc.) are removed to keep JSON compact.
  • Fixed contradictory whitelist/omit rules and simplified ComponentProperty (removed Type and HumanReadable).
• Enhanced GhJSON schema:
  • Rich parameter metadata: parameterName, dataMapping, simplify, reverse, invert, unitize, expression, variableName, isPrincipal, locked.
  • Unified value property for component state across sliders, panels, scribbles, scripts, and value lists.
  • Extended component schema with Id, Params, InputSettings, OutputSettings, and ComponentState.
  • Removed redundant/implicit fields (expressionContent, access, description, optional, type/object metadata).
  • BREAKING: The enhanced schema is now the default for all gh_get and gh_put operations, with legacy Pivot preserved for positioning compatibility.

🧠 Preparing advanced scripting workflows

• Lays the groundwork for future scripting-focused releases with robust VB Script and script component handling:
  • Full 3‑section VB Script serialization/deserialization (imports, script, additional) with a dedicated VBScriptCode model and vbCode support in ComponentState.
  • GhJSON now extracts and restores VB Script code via ScriptSource properties (UsingCode, ScriptCode, AdditionalCode) using reflection.
• Script components implement IGH_VariableParameterComponent for reliable dynamic parameter creation/removal:
  • Parameter settings applied via CreateParameter/DestroyParameter and VariableParameterMaintenance() with names, optional flags, and modifiers.
• All script parameters and code operations are wrapped in RhinoApp.InvokeOnUiThread() to keep the Rhino/Grasshopper UI responsive and stable.
• A new unified script_generator AI tool replaces legacy script_new and script_edit, creating or editing script components from natural language and optional component GUIDs—forming the base for richer scripting assistants in upcoming versions.

🧪 Reliability, workflows & test coverage

• Utility helpers for centralized Grasshopper/Rhino operations:
  • GHConnectionUtils, GHGenerateUtils, RhinoFileUtils, RhinoGeometryUtils.
• SelectingComponentBase now persists selected objects across file save/load, including restored selection counts and graceful handling of missing objects.
• New SmartHopper.Core.Grasshopper.Tests project:
  • 40+ AIResponseParserTests covering complex parsing and edge cases.
  • 30+ PropertyManagerTests validating type conversion and persistent data behavior.
• Hotfix workflow system:
  • Automated hotfix branch creation, version bumping, conflict‑safe milestone updates, and fully documented hotfix pipelines.
  • Extended CI validations and test workflows for hotfix/** and release/** branches.

🛠️ Technical Requirements

• Rhino 8.24 or above is required
• Windows 10/11 (macOS has not been tested)
• Valid API keys for at least one supported AI provider (MistralAI, OpenAI, DeepSeek, Anthropic, OpenRouter)

⚠️ Important Notes

• This is an alpha release with some features still unstable or subject to change.
• API keys are required, and usage costs apply based on your provider and model choices.
• Documentation is actively evolving to cover the new GhJSON schema, scripting tools, and knowledge workflows.

🤝 We Value Your Feedback!

Help shape SmartHopper’s future by:

• Sharing your experiences with the new GhJSON, scripting, and knowledge tools
• Suggesting improvements via our discussions
• Telling us what AI capabilities or workflows would help your Grasshopper/Rhino practice most

We hope you enjoy these new features and improvements!

Happy designing! 🎨

---

Full Changelog: 1.0.1-alpha...1.1.0-alpha

SmartHopper 1.0.1-alpha: Enhanced Model Support & Error Handling [Patch]

This patch release introduces significant improvements to model handling, error reporting, and tool usability, along with several important bug fixes.

Detailed list of changes:

• Enhanced Model Support:

  • Added support for any model name, even if not explicitly listed in the provider's registry
  • Expanded the provider's database with more predefined models

• Tooling Improvements:

  • Enhanced AI Tool descriptions with better guided instructions
  • Added specialized wrappers for common Grasshopper operations (selection, visibility, grouping, etc.)
  • Improved list_filter tool with better natural language support for filtering, sorting, and reordering

• Bug Fixes:

  • Fixed model badge display to correctly show "invalid model" status (#332, #329)
  • Improved error message visibility in WebChat UI, especially for HTTP 400 and token limit errors (#334)
  • Fixed list_filter tool's automatic sorting and deduplication issues (#335)

• Technical Improvements:

  • Centralized error handling in AIReturn and tool calls
  • Enhanced metric aggregation in Conversation Sessions
  • Improved error handling to ensure all error messages are properly surfaced to users

---

What's Changed

• fix(model): bypass capability validation for unregistered models by @marc-romu in #330
• fix(component): fix model badge display by @marc-romu in #336
• fix(core): improve error handling in ConversationSession and metrics aggregation by @marc-romu in #337
• fix(gh-get): specialized ai tools by @marc-romu in #339
• fix(list-filter): prevent automatic sorting and deduplication of result indices by @marc-romu in #340
• refactor(models): added more predefined models by @marc-romu in #341

Full Changelog: 1.0.0-alpha...1.0.1-alpha

SmartHopper 1.0.0-alpha: Canvas Button, Component Badges & New Providers

This release introduces many exciting features; that's why the version is bumped to 1.0!! The Canvas button for a live assistance chat, component badges for model compatibility, new providers, and enhanced stability and security.

🎨 Canvas Button

• New button on the top-left corner of your canvas to trigger an AI chat with SmartHopper Assistant.
• Designed to help with Grasshopper tasks and equipped with tools for valuable information and basic tasks.
• Can be hidden in settings if not needed.

🏷️ Component Badges

• New badges display model compatibility with components before running them.
• Green tick badge for verified models (gpt-5-mini from OpenAI and mistral-small-latest from MistralAI).
• Blue badge indicates automatic replacement of incompatible models with compatible defaults.
• Red badge indicating that no model was found compatible.

🤖 New Providers & Streaming Support

• Added support for Anthropic and OpenRouter, alongside MistralAI, OpenAI, and DeepSeek.
• Now all providers can stream live responses, if provided by the API

✨ Other Improvements

• Internal improvements for stability and expandability.
• Encrypted API keys with individualized keys for enhanced security.

🛠️ Technical Requirements

• Rhino 8.19 or above
• Windows 10/11 (macOS untested)
• Valid API keys for chosen providers (OpenAI, MistralAI, DeepSeek, Anthropic, OpenRouter)

⚠️ Important Notes

• Settings file will rewrite on the first run due to security enhancements. Save your API keys again if you encounter issues.
• Image generation currently supports only OpenAI DALL·E models.
• Automatic model fallback may override user-selected models to meet component requirements. You will see a blue badge on the component indicating that the model was replaced.
• API usage costs apply based on your provider plan.

🤝 We Value Your Feedback!

• Share your experiences and suggestions via the discussion forum
• Report issues or request features on the issue tracker
• Contribute to a better SmartHopper!

Happy designing! 🎨

---

What's Changed

• test(infrastructure): add comprehensive unit tests for ModelManager and AIContextManager by @marc-romu in #295
• feat(ui): add CanvasButton to trigger SmartHopper assistant dialog by @marc-romu in #297
• refactor(settings): reorganize settings dialog into tabs and improve API key encryption by @marc-romu in #299
• fix(core): prevent temp file leaks in ImageViewer and resolve wildcard model default issue by @marc-romu in #302
• refactor(infrastructure): unify AI request/return schema and provider APIs; migrate tools and WebChat by @marc-romu in #309
• refactor(model-manager): centralize model management and provider-scoped selection by @marc-romu in #312
• feat(components): add model badges with hover labels and provider strip by @marc-romu in #314
• feat(conversation): introduce ConversationSession; refactor AICall, streaming, validation, and model selection by @marc-romu in #315
• feat(core-io): introduce safe versioned persistence (V2) by @marc-romu in #317
• feat(providers,ui,infra): Anthropic/OpenRouter providers, WebChat WebView UI, policy pipeline + streaming by @marc-romu in #318
• refactor(webchat): Enhance conversation session and tool calling across providers by @marc-romu in #321
• fix(components): correct index parsing in AIListFilter component by @marc-romu in #324
• docs: update documentation for 1.0.0-alpha release by @marc-romu in #325

Full Changelog: 0.5.3-alpha...1.0.0-alpha

SmartHopper 0.5.3-alpha: Fix script_new json schema [Patch]

Fixes small json schema request issue with gpt-5 models in tool.

What's Changed

• fix(script_new): incompatibility with gpt-5 json schema requirements by @marc-romu in #306

Full Changelog: 0.5.2-alpha...0.5.3-alpha

SmartHopper 0.5.2-alpha: First Initialization Fix [Patch]

This patch release resolves an issue of loading settings when the SmartHopper.json file is not available. This happened to everyone who installed the plugin for the first time. 😢

Detailed list of changes

• fix(settings): invalid settings file initialization by @marc-romu in #303

Full Changelog: 0.5.1-alpha...0.5.2-alpha

SmartHopper 0.5.1: Chat Settings and Bug Fixes [Patch]

This patch release adds a new chat setting for AI-generated greetings and resolves several bugs related to greeting logic, toggle behavior, provider initialization, and list generation.

Detailed list of changes:

• Added a new setting (EnableAIGreeting) to enable or disable AI-generated greetings in chat.
• Fixed greeting generation to fall back on the provider’s default model when no model is specified in AIUtils.GetResponse.
• Fixed components using boolean toggles not reprocessing when the Run toggle remained persistently true.
• Fixed race conditions in AI Provider Settings by introducing lazy default values during initialization.
• Fixed list generation in the list_generate tool to correctly handle long requests and meet expected lengths (#277).

Full Changelog: 0.5.0-alpha...0.5.1-alpha