Release History - samwega/obsidian-wordsmith GitHub Wiki

Release History

w͜s What's New in v3.0.0 - 400+ Models & Local AI Provider Support

This major release introduces a complete architectural overhaul brings Dynamic AI Model Providers to WordSmith. The hardcoded model list has been replaced with a dynamic, provider-agnostic framework. This unlocks access to over 400 models from top-tier services like OpenAI, Anthropic, Google, and OpenRouter, others you wish to add, and crucially, brings first-class support for running AI completely offline with local servers like Ollama and LM Studio. You now have ultimate control over your AI toolkit.

Key features and improvements include:

• Massive Model Expansion: Connect to any service with a standard chat completions API. The new system provides immediate access to hundreds of models, from the latest flagships to specialized open-source variants.
• First-Class Local AI Support: Run your transformations and generations privately and for free by adding your local Ollama or LM Studio endpoints as providers.
• Comprehensive Model Browser: A new "Browse All Models" modal displays every available model from all your configured providers in a single, searchable interface.
• Favorites System: Stared models are moved to the top of the list for quick selection.
• Future-Proof Foundation: WordSmith no longer requires manual updates to support new models. As providers add models to their APIs, they will automatically become available in your browser.
• Overhauled the prompt engineering system for significantly improved AI reliability and instruction-following. Now that we have access to thousands of models, each trained with different prompt and system instruction structures, we had to construct prompts in a way to maximize compliance by as many models as possible. This change enhances performance across all models, from top-tier APIs to local servers.
• Improved Settings Menu: Now with tabs, free from clutter.

w͜s What's New in v2.2.0 - Knowledge Graph Generation

This is a major feature release that introduces a powerful new way to visualize your ideas. With the new Knowledge Graph Generator, you can now transform your notes and research into beautiful, editable Obsidian Canvases.

Key features and improvements include:

• AI-Powered Knowledge Graph Generation—This is the centerpiece of the update!
  • A new command, Generate knowledge graph, analyzes your context to extract key entities and their relationships.
  • Intelligent & Context-Aware—Fully integrated with the Context Control Panel. Use your full note and/or provide detailed instructions and linked notes in the "Custom Context" field to guide the generation.
  • Spacious & Readable Layouts—Utilizes the d3-force engine to automatically arrange nodes in a clean, expanded layout with plenty of breathing room, avoiding cluttered clusters.
  • Automatic Node Sizing & Coloring Nodes perfectly expand to fit their entire text content—no more manual resizing! To improve readability, the most central "hub" nodes in your graph are automatically highlighted with a random color.
• Tested Successfully With Several Models—Not all models are capable of generating the precise JSON format required for the graph generation, but I have successfully tested with GPT 4.1 & 4o, Gemini 2.5 Flash, Grok 3, any of the Claude models, DeepSeek v3, Command A is a champ, all of the Llama models including the small 3.3 70B, and to my surprise, even small models like Qwen 3 32B succeeded! The size of the content to graph is a factor.
• All WordSmith Standard Functionality Works in Canvas Cards Seamlessly.

Minor Versions of v.2.2.x

• 2.2.1—Added a Max Output Tokens input box. Larger models were failing at Knowledge Graph generation because the output was too large and got truncated by legacy hardcoded max 2k tokens output.
• 2.2.2—Improved Name Format for Knowledge Graph, more flexible and human readable. Also changed WordSmith logo to W-shaped Lucide icon crown.

w͜s What's New in v2.1.0 - Context Panel Save State Across Sessions

• Feat: Persistent & Decoupled AI Context Control—Your preferred context settings (Dynamic, Full Note, Custom Context Toggle & Text) are now saved automatically across sessions, and the plugin can utilize them even when the Context Control Panel is closed.
• Bug Fix: was sending thinking budget to all "Flash" models causing Gemini 2.0 Flash-Lite to throw an error.

Minor Versions of v.2.1.x

• 2.1.1 🛠️ Under The Hood Improvements—The plugin's internal prompt construction logic has been centralized, leading to more robust and consistent AI interactions across all supported models.

w͜s What's New in v2.0.0 - OpenRouter Support - 28 New Models

Key features and improvements include:

• OpenRouter API Integration: A huge new addition! You can now use your OpenRouter API keys to access dozens of models from various providers.
• 28 New Models Added to previous existing 5 models list—Massively expanding your options, this release includes:
  • OpenAI: GPT-4o, GPT-o4-mini, GPT-4.5 & O3
  • Google: Gemini 2.0 Flash-Lite & Gemma 3 27B
  • Anthropic: Claude 3.5 Sonnet & Claude 3.7 Sonnet
  • Grok: Grok 3 beta
  • DeepSeek: DeepSeek Chat v3 & DeepSeek Chat R1
  • Qwen: Qwen 3 235B A22B & Qwen 3 32B
  • Llama: Llama 3.3 70B Instruct, Llama 3.1 405B Instruct, Nvidia/Llama 3.1 Nemotron Ultra 253B v1
  • Hermes: Hermes 3 70B, Hermes 3 405B
  • TheDrummer's Skyfall 36B V2, Valkyrie 49B v1 & Anubis Pro 105B v1
  • Cohere's Command A (111B)
  • Mistral Large 2411
  • And more: Goliath 120B, Magnum 72B, Lumimaid v0.2 70B
  • Note: 13 of these new models are flagged as "NSFW" due to their ability to comply with NSFW generation requests.
  • Most of the OpenAI & Google models are accessible both via their API's and OpenRouter's API.
• Per-Model Temperature Control: With so many models, each with different settings, the temperature slider needed to get smarter. It automatically adjusts to the specific minimum and maximum supported values for the selected model and resets to the model's default temperature when you switch, giving you fine-tuned control over creativity with sane defaults.
• Refactor that makes it much easier to add new models.
• Improved Model Information: The model selection dropdown in settings now provides more useful information about each model.
• Better Runtime Debug Logging: Enhanced logging makes it easier to diagnose issues.
• Bug Fixes: Fixed a smooth scrolling bug in styles. Fixed infinite loop when parsing wikilinks in custom context.
• Refined System Instructions: The underlying instructions sent to the models have been improved for better compliance and quality.

Minor Versions of v.2.0.x

• 2.0.1
  • Addressed a bug where newline removal suggestions (pilcrows) were not resolved by "Accept/Reject all in paragraph" commands and were not correctly identified for navigation, improving workflow consistency.
  • Added models Claude Sonnet 4 & Claude Opus 4 via OpenRouter API. Gave Opus for a 16k thinking token budget.

w͜s What's New in v1.11.0 - Temperature Sliders & Debug Logging

• Temperature Sliders—now included in both the Context Side Pane and the settings panel. Previously, the temperature was hard coded at 0.7. Now, it will defaut to 1.0 for enhanced creativity. I've seen pretty good compliance with temperatures as high as 1.8 (not thoroughly tested yet).
• Console Debug Logging—You now have a toggle which will log all WordSmith activity in the developer console. This will include the prompt sent to the model, so you can verify that it is sending the intended, properly constructed prompt. Temperature is also included.
• Upgrade Gemini models to latest versions:
  • Upgrades gemini-2.5-flash-preview-04-17 to gemini-2.5-flash-preview-05-20.
  • Upgrades gemini-2.5-pro-preview-05-06 to gemini-2.5-pro-preview-06-05.

w͜s What's New in v1.10.0 - Enhanced Context Awareness & Prompt Stability! w͜s

• Filename Inclusion for Improved Contextual Understanding:—When using "Full Note" or "Dynamic Context" options, WordSmith now automatically includes the name of the current file (e.g., Filename: My Note About AI.md) as part of the context sent to the AI. Previously, if you didn't include the title of the note as a header, it would not be incldued with the context.
• Reliable Preset Prompt Management:
  • The way preset (built-in) prompts are handled in your data.json settings file has been improved.
  • With every plugin reload, WordSmith now ensures that all preset prompts are correctly reset to their latest built-in definitions and maintained in their intended order.
  • Any user customizations made directly to preset prompts in data.json will be overwritten to ensure stability and access to the latest official prompt versions.
  • User-created prompts remain untouched and are preserved.
• Minor UI Style Tweaks: Subtle adjustments have been made to the user interface for a cleaner and more consistent visual experience.

Minor Versions of v.1.10.x

• 1.10.1—Complete codebase refactoring. Files were getting too big and unmanageable. Many improvements.
• 1.10.2—Fix: reintroduced the Pilcrow ¶ functionality accidentally removed during refactoring when we switched to ghost text CM6 inplementation. This time the ¶ symbol uses ghost text, rather than modify the user's actual text. Also completely revamped the style for both ¶ and ↵ symbols.
• 1.10.3—Improved AI compliance with any additional instructions the user might provite through the Custom Context input box, for the purposes of Wiki - Personalized Memory: Crafting Memory Notes

w͜s What's New in v1.9.0 - Streamlined Ghost Text Suggestions & Cleaner State Management! w͜s

This significant architectural update refines how WordSmith presents and manages "added" text suggestions, leading to a smoother, more robust, and performant experience.

• Native "Ghost Text" for Added Suggestions:

  • No More Direct Text Insertion (Initially): When the AI suggests adding text (the green highlights), this text is no longer immediately inserted into your note's content.
  • CodeMirror 6 Widget Power: Instead, "added" suggestions are now rendered as "ghost text" using CodeMirror 6's native widget system. This means the green text you see is a visual overlay, and your actual note content remains unchanged until you explicitly "accept" the suggestion.
  • Visual Consistency: The appearance of added suggestions (green background, active highlighting) remains identical, ensuring a familiar user experience.
  • Benefits: This approach is cleaner, reduces the chance of errors with text offsets, and improves performance by leveraging CM6's optimized decoration handling.

• Simplified Suggestion Resolution:

  • Accepting an "added" suggestion now cleanly inserts the ghost text into the document at the correct position.
  • Rejecting an "added" suggestion simply removes the visual ghost text, as no actual document change had occurred yet.
  • Rejecting all suggestions (Clear all active suggestions) is now more straightforward, as it primarily involves clearing visual decorations without needing to revert document changes for "added" text.

• Removal of File-Based Suggestion Persistence:

  • The previous experimental feature that saved active suggestion states (including CM6 decoration details) to a file (.obsidian/plugins/text-transformer/suggestions.json) has been removed.
  • Why? The new "ghost text" architecture for added suggestions means that the document itself isn't modified until acceptance. For "removed" suggestions, the original text remains until accepted. This simplifies state management significantly. While this means visual suggestions won't persist across full Obsidian restarts in their visual state, the underlying document integrity is better preserved, and re-triggering the transformation if needed is a cleaner workflow. This change prioritizes stability and a more native editor integration over the complexity of maintaining visual state persistence across sessions.
  • Note: This change streamlines the plugin's internal workings and reduces potential points of failure associated with syncing visual states with file content.

• Under-the-Hood Optimizations:

  • The core logic in suggestion-state.ts, textTransformer.ts, and suggestion-handler.ts has been extensively refactored to support the new ghost text model.
  • This results in a more maintainable codebase and aligns WordSmith more closely with CodeMirror 6 best practices.

Why this matters:

Version 1.9.0 marks a significant step forward in WordSmith's technical foundation. By adopting a more native CodeMirror 6 approach for "added" suggestions, the plugin becomes more robust, performant, and easier to maintain, all while keeping the user interface and core functionality consistent. The removal of file-based suggestion persistence simplifies the plugin's architecture and focuses on a stable, in-session suggestion experience.

Minor Versions of v.1.9.x

• 1.9.1—Critical Bug fix: improve suggestion handling for empty paragraphs and refine mark mapping logic (introduced in version 1.9.0 from the refactoring for "ghost text" suggestions).
• 1.9.2—The CM6 implementation of suggestions (ghost text) caused a lot more issues than expected. In particular: fixed aggressive view centering & the corect highlighting of the first of multi-line suggestions. I'm finally satisfied with the results. Finally a release that should be bug free!
• 1.9.3—Tweaks to Improve Preset Prompt; "Default Prompts" renamed "Preset Prompts" & "Custom Prompts" renamed "User Prompts".

w͜s What's New in v1.8.0 - Persistent Suggestions & Enhanced Prompts! w͜s

This release brings two major improvements to make your AI writing experience more seamless and powerful:

• Persistent Suggestions Between Reloads: Your active AI suggestions are now preserved when you reload Obsidian. Previously, if closing the application while suggestions were active, upon returning, you'd see old and new text mixed together and no way to accept/reject or undo it.
• Expanded Preset Prompts: A fresh set of carefully crafted default prompts has been added to the collection, giving you more options for text transformation right out of the box. The existing prompts have been recrafted for a more precise output.
• Moved all styling from JavaScript or HTML to styles.css - it required some pretty major refactoring of the UI.
• Replaced use of innerHTML (security risk) with safer DOM API
• Removed a line which was manually detaching leaves during the unload process - an antipattern
• Fixed a ton of bugs which were created in the refactoring process

Minor Versions of v.1.8.x

• 1.8.1—Improved built in prompts for improved context use. Created wiki and moved some of the README.md into it. Wiki contains other cool stuff!
• 1.8.2—Feature: Now you can automatically save your ad-hoc prompts (Generation at Cursor) to your clipboard if you prefer, so they are not lost if you need to use them again. Works best with a clipboard manager, so your existing clipboard content doesn't get overwritten, but simply added to.

Ready for Prime Time!: With this update, I'm finally getting ready to have my very first Plugin—WordSmith—added to the official Obsidian community plugin store. 🤞🏻

w͜s What's New in v1.7.0 - Enhanced Custom Context with Smart Note Linking! w͜s

This upgrade streamlines the process of providing the AI with the precise information needed for optimal results. By integrating seamless note linking into the Custom Context feature, WordSmith now leverages your interconnected notes, making context building faster, more intuitive, and more powerful.

Key Features of v1.7.0:

• Smarter Custom Context Input: Providing custom instructions or reference material to the AI is now more powerful and intuitive.
• Easy Note Linking with [[: You can now type [[ in the "Custom" context box of the WordSmith panel to quickly search for and link any note from your vault, seamlessly integrating with your Obsidian workflow.
• Full Note Content Embedding: When you link notes (e.g., [My Style Guide](/samwega/obsidian-wordsmith/wiki/My-Style-Guide)) in the custom context, WordSmith will now intelligently fetch their entire content and include it as part of the context sent to the AI. This means you can easily reference detailed style guides, extensive notes, or specific information without manual copy-pasting!

Minor Versions of v.1.7.x

• v1.7.1—Text Transformer Rebranded as WordSmith. Developers, note that most code still refers to TextTransformer.

w͜s What's New in v1.6.0 – Enhanced Suggestion Navigation! w͜s

This release supercharges your suggestion review workflow, making it faster and more intuitive to manage AI-generated edits and new content.

Key Features of v1.6.0:

• Suggestion Navigation Hotkeys:
  • Effortlessly cycle through all active AI suggestions in your document using new commands:
  • "WordSmith: Focus next suggestion" – Move your cursor to the start of the next suggestion, making it instantly active for review.
  • "WordSmith: Focus previous suggestion" – Jump back to the previous suggestion with a single keystroke.
• Improved Cursor Behavior:
  • Whenever new suggestions are generated—whether from a text transformation or an ad-hoc prompt—the cursor is now automatically positioned at the very first suggestion, immediately highlighting it for action. This streamlines the accept/reject workflow and eliminates manual cursor adjustment.
• Refined Keyboard-First Editing:
  • Combined with dynamic active suggestion highlighting (from v1.5.1), these navigation enhancements make it easier than ever to see, select, and act on the exact change you want—all without leaving the keyboard.
• Under-the-Hood Improvements:
  • Removed the legacy toggles in Settings to ask the AI not to modify text in quotation marks ("") and citations (lines starting with >). They were not working at all and required a much more involved implementation. Will add the feature in a future release if people ask for it.
  • Minor performance and stability tweaks for smoother suggestion handling.

Why this matters:

WordSmith now offers one of the fastest and most ergonomic AI suggestion review experiences in Obsidian—keeping you firmly in your creative flow.

Minor Versions of v.1.6.x

• v1.6.2—Granular Suggestions for Multi-line Generation: Text generated at the cursor that spans multiple lines is now intelligently split into individual suggestion segments (per line and per newline), allowing for more precise review and control.
• v1.6.3—UI & Workflow Refinements:
  • Context Panel: "Dynamic context lines" input now lives in the AI Context Control side panel, only showing when "Dynamic context" is on.
  • Removed Legacy Full Document Transform: This command was adding unnecessary complexity. You may simply Ctrl-A and Transform Selection.
• v1.6.4—Minimalistic Context Control Side Panel: All the explanations are now tucked away in a collapsible menu. The menu is as minimal as possible, allowing you to make it narrow and gain space. It is more aesthetically pleasing as well.

w͜s What's New in v1.5.0 - Context Aware Generation at Cursor! w͜s

This major release introduces a powerful new way to interact with AI: Ad-hoc Contextual Generation directly at your cursor, presented as a suggestion. Go beyond processing existing text – now you can generate new content, ideas, or continuations seamlessly within your writing flow.

Key Features of v1.5.0: Prompt Based Context Aware Generation at Cursor

• New Command: "Generate text with ad-hoc prompt (as suggestion)"
  • Trigger this editor command (via hotkey or command palette) to open the new "Context Aware Generator" modal.
  • Type your specific instruction (e.g., "brainstorm ideas for X," "write an intro paragraph about Y," "explain Z simply") into the modal.
• Contextual Awareness for Generation:
  • Leverages the settings in your AI Context Control Panel (Custom, Dynamic, or Entire Note).
  • When Dynamic or Entire Note context is active, a special marker (<<<GENERATION_TARGET_CURSOR_POSITION>>>) is inserted at your precise cursor location within the context sent to the AI. This tells the AI exactly where you intend the new text to be generated.
  • The AI uses this context and marker to produce more relevant and precisely placed generated content.
• Seamless Suggestion Workflow:
  • The AI-generated text is inserted directly at your cursor position as a single "added" suggestion.
  • You can then accept or reject this generated block just like any other WordSmith suggestion.
  • The cursor automatically moves to the end of the inserted text.
• Modal Enhancements:
  • The "Context Aware Generator" modal automatically focuses the prompt input area.
  • A notice appears if you try to submit an empty prompt.
• Streamlined Suggestion Navigation:
  • When using "Accept/Reject Next Suggestion," if there's only one suggestion in the entire document, it's now resolved immediately without requiring a second key press (the "scroll and press again" step is skipped).
  • For multiple suggestions, the two-step navigation (scroll to suggestion, then press again to act) remains, ensuring you can clearly target specific suggestions.

Why this is a game-changer:

This feature transforms WordSmith from primarily an editing/refinement tool into a versatile writing partner capable of both transforming existing text and generating new content with contextual understanding, all within the familiar suggestion-based workflow. Whether you're stuck, need a creative boost, or want to quickly draft a section, the Context Aware Generator is ready to assist.

Minor Versions of v.1.5.x

• v1.5.1—Dynamic Active Suggestion Highlighting: Implemented strong visual cues – the suggestion targeted by your cursor (when placed at its beginning) now stands out with significantly increased contrast and a glowing outline, making it unmistakably clear which suggestion you're about to accept or reject.

w͜s What's New in v1.4.0 - Precision Newline Tracking! w͜s

This update brings a crucial enhancement to how WordSmith handles changes involving newlines, making suggestions more accurate and intuitive, especially for structural edits and reformatting.

• Enhanced Newline Diffing:
  • Newlines are now consistently marked with green "↵" (added) and pink "¶" (removed).
  • Improved Accuracy for Structural Changes: The underlying diffing mechanism (jsdiff) has been switched to diffWordsWithSpace and combined with more granular internal processing. This significantly improves the plugin's ability to detect and represent newline additions and removals, even when they are part of larger text blocks or when AI rephrases sentences across line breaks.
  • Clearer Visuals for Reformatting: Whether the AI is breaking a long sentence into multiple lines, or merging multiple lines into one, these structural changes involving newlines are now clearly and correctly marked for your review.
• More Robust Suggestion Handling:
  • The logic for applying and resolving suggestions involving these new newline markers has been updated for consistency. Accepting an added "↵" correctly inserts a newline, and rejecting a removed "¶" correctly re-inserts the newline.
• Under-the-Hood Tweaks: Codebase cleaned with TypeScript typing refinements and removal of unused code.

Why this matters: This update solves inconsistent tracking of AI-suggested newlines additions and removals, ensuring you see and control proposed line changes for clarity, conciseness, or restructuring.

Minor Versions of v.1.4.x

• v1.4.1—Style no longer hard coded—you can edit the CSS now.
• v1.4.2—Adaptive Styling for dark/light themes.
• • v1.4.3—Disables spellcheck red squiggle during suggestion. Suggestion sticks the old and new word together and everything is flagged by spellcheck, so it needed to be disabled.

w͜s What's New in v1.3.0 - Revamped Suggestion Display! w͜s

This major update overhauls how AI suggestions are displayed and managed, moving to a more robust and visually integrated system within the Obsidian editor:

• New Suggestion Engine:
  • No More Markdown Markers: WordSmith no longer uses ==highlight== for additions or ~~strikethrough~~ for removals in the actual document text.
  • Direct CodeMirror Decorations: Suggestions are now rendered using inline-styled CodeMirror 6 decorations. This means:
  • Added text is typically shown with a light green background.
  • Removed text is shown with a light pink background and a line-through effect.
  • These styles are applied visually without altering the underlying text with special characters, providing a cleaner experience and better theme compatibility.
• Robust Internal Logic:
  • The internal state management for suggestions has been rewritten using a CodeMirror ViewPlugin for more direct and efficient handling of decorations.
  • Improved error handling and extensive logging have been added for easier debugging and more stable performance.
• Settings Enhancements:
  • Settings loading is now more robust, with better migration logic for prompts and ensures default values are correctly applied.
• Refined User Experience:
  • Command names have been simplified (e.g., "(CM6)" suffix removed).
  • Notices and messages have been clarified.

Minor Versions of v.1.3.x

v1.3.1—AI Model Selector in Side Pane: Quickly switch your preferred AI model directly from the AI Context Control Panel for even faster workflow adjustments. v1.3.2—Bug fixes.

w͜s What's New in v1.2.1 & v1.2.2 w͜s

These updates brought significant UI enhancements, a new dynamic translator feature, and a powerful new default prompt:

• Revamped Settings UI:
  • Implemented a two-column layout for Prompt Settings and Prompt Management for a more organized view.
  • Enhanced the overall settings UI with collapsible sections, improved styling, and unified API Key and Model settings for better clarity and ease of use.
  • Reordered settings sections for a more logical flow.
• Dynamic Translator with Language Input: The "Translate" prompt is now more dynamic, allowing users to specify the target language directly within the prompt flow, and it now includes an input field for the target language.
• New Preset Prompt - "Mind the Context!": This prompt instructs the model to do what the context says, and revises the text strictly based on the directives provided in the context. For instance, you can add any rule on the fly in the Custom Context box.
• Bug Fixes: Addressed issues related to the translator input box display.

w͜s What's New in v1.2.0 w͜s

This version significantly enhanced your control over the AI's context and streamlined prompt management:

• Take control of your AI's context with the new Context Control Panel:
  • Dynamic Context: Automatically include surrounding paragraphs for richer, more relevant AI suggestions.
  • Entire Note Context: Utilize the full content of your current note to inform the AI.
  • Custom Context: Provide your own specific text snippets through a dedicated input area for precise context.
  • Note: Dynamic and Entire Note context options are mutually exclusive to ensure clarity.
• New Setting - Dynamic Context Lines: Fine-tune the amount of surrounding text (number of paragraphs) used by the Dynamic Context feature.
• Prompt Management Update: Added a helpful note in the settings page explaining how to manually edit default prompts by modifying the data.json file (requires an Obsidian reload after changes).

w͜s What's New in v1.1.0 w͜s

Version 1.1.0 expanded your AI model horizons:

• Google Gemini Integration: Connect your Gemini API key to use Google's powerful models (including Gemini 2.5 Flash and Pro) for all text transformations. This offers greater flexibility in balancing cost, speed, and AI capabilities.
• Seamless Model Selection: The settings panel was updated to easily switch between your preferred OpenAI and Gemini models.

w͜s What's New in v1.0.0 w͜s

Version 1.0.0 marked the initial major transformation from the original Proofreader plugin, introducing core functionalities that define WordSmith today:

• Introducing the Prompt Palette: Moved beyond a single proofreading instruction to a versatile palette. Users can select from a variety of predefined prompts to transform text in multiple ways (e.g., Improve, Shorten, Lengthen, Fix Grammar).
• User Prompts Unleashed: Empowered users to create and save their own unique prompts. This allows for a highly personalized AI writing assistant tailored to specific needs and styles.
• Foundation for a Multi-Feature AI Assistant: This version laid the groundwork for WordSmith to become a comprehensive AI writing tool, moving beyond basic proofreading to a wider range of text manipulations.

)