Smart Connections v3

New Features

Smart Chat v1

• Effectively utilizes the Smart Environment architecture to facilitate deeper integration and new features.

Improved Smart Chat UI

• New context builder
  • makes managing conversation context easier
• Drag images and notes into the chat window to add as context
• Separate settings tab specifically for chat features

Improved Smart Chat compatibility with Local Models

• Note lookup (RAG) now compatible with models that don't support tool calling
  • Disable tool calling in the settings

Ollama embedding adapter

• use Ollama to create embeddings

Fixed

• renders content in connections results when all result items are expanded by default

Housekeeping

• Updated README
  • Improved Getting Started section
  • Removed extraneous details
• Improved version release process
• Smart Chat v0 (legacy)
  • Smart Chat v0 will continue to be available for a short time and will be removed in v3.1 unless unforeseen issues arise in which case it will be removed sooner.
  • Smart Chat v0 code was moved from brianpetro/jsbrains to the Smart Connections repo

patch v3.0.1

Improved Mobile UX and cleaned up extraneous code.

patch v3.0.3

Fixed issue where connections results would not render if expand-all results was toggled on.

patch v3.0.4

Prevented frontmatter blocks from being included in connections results. Fixed toggle-fold-all logic.

patch v3.0.5

Fixes Ollama Embedding model loading issue in the settings.

patch v3.0.6

Fixed release notes should only show once after update.

patch v3.0.7

Added "current/dynamic" option in bases connection score modal to add score based on current file. Fixed issue causing Ollama to seemingly embed at 0 tokens/sec. Fixed bases integration modal failing on new bases.

patch v3.0.8

• Improved bases integration UX
  • prevent throwing error on erroroneous input in cos_sim base function
  • gracefully handle when smart_env is not loaded yet
• Reduced max size of markdown file that will be imported from 1MB to 300KB (prevent long initial import)
  • advanced configuration available via smart_sources.obsidian_markdown_source_content_adapter.max_import_size in smart_env.json
• Removed deprecated Smart Search API registered to window since smart_env object is now globally accessible
• Fixed bug causing expanded connections results to render twice

patch v3.0.9

• Reworked the context builder UX in Smart Chat to prevent confusion
  • Context is now added to the chat regardless of how the context selector modal is closed
  • Removed "Back" button in favor of "Back" suggestion item
• Fixed using @ to open context selector in Smart Chat
  • "Done" button now appears in the context selector modal when it is opened from the keyboard

patch v3.0.10

Fixed Google Gemini integration in the new Smart Chat

patch v3.0.11

Fixes unexpected scroll issue when dragging file from connections view (issue #1073)

patch v3.0.12

Fixes pasted text: should paste lines in correct order (no longer reversed)

patch v3.0.13

• Prevents trying to process embed queue if embed model is not loaded
  • Particularly for Ollama which may not be turned on when Obsidian starts
  • Re-checks for Ollama server in intervals of a minute
  • Embed queue can be restarted by clicking "Reload sources" in the Smart Environment settings

patch v3.0.14

• Improved hover popover for blocks in connections results and context builder
• Refactored context_builder component to extract context_tree component and prevent passing UI components
  • these components are frequently re-used, the updated architecture should make it easier to maintain and extend
• Fixed: should not embed blocks with size less than min_chars
• Fixed: Smart Chat completion requests should have a properly ordered messages array

patch v3.0.15

• Fixed: some Ollama embedding models triggering re-embedding every restart

patch v3.0.16

• Fixed: no models available in Ollama should no longer cause issues in the settings

patch v3.0.17

• Improved embedding processing UX
  • show notification immediately to allow pausing sooner
  • show notification every 30 seconds in addition to every 100 embeddings
• Fixed: Smart Environment settings tab should be visible during "loading" state
  • prevents "Loading Obsidian Smart Environment…" message from appearing indefinitely in instances where the environment fails to load from errors related to specific embedding models

patch v3.0.18

• Fixed: Smart Connections view rendering on mobile
  • should render when opening the view from the sidebar
  • should update the results to the currently active file

patch v3.0.19

• Added: model info to Smart Chat view
  • shows before the first message and anytime the model changes since the last message
• Fixed: ChatGPT sign-in with Google account
  • should now work as expected
  • will require re-signing in to ChatGPT after update
• Fixed: Smart Chat thread adapter should better handle past completions to prevent unexpected behavior
  • prevented build_request from outputting certain request content unless the completion is the current completion
    • logic is specific to completion adapters (actions, actions_xml, thread)

patch v3.0.20

• Fixed: Smart Environment settings tab should be visible during "loading" and "loaded" states
• Fixed: Open URL externally should use window.open with "_external" if webviewer plugin is installed

patch v3.0.21

• Implemented Smart Completions fallback to Smart Chat configuration
  • WHY: enables use via global smart_env instance without requiring chat_model parameters in every request

patch v3.0.22

• Improved connections view event handling
  • prevent throwing error when no view container is present on iOS

patch v3.0.23

• Added Getting Started guide
  • opens automatically for new users
  • can be opened manually via command Show getting started
  • can be opened from the connections view "Help" icon
  • can be opened from the main settings "Open getting started guide" button

patch v3.0.24

Fix Lookup tab not displaying.

patch v3.0.25

Fixed connections view help button failing to open

patch v3.0.26

Temp disable bases integration since Obsidian changed how the integration works and there is currently no clear path to updating.

patch v3.0.27

• Added: Smart Chat lookup now supports folder-based filtering
  • mention a folder when requesting a lookup using self-referential pronoun (no special folder syntax required)
    • ex. "Summarize my thoughts on this topic based on notes in my Content folder"
• Added: Smart Chat system prompt now allows {{folder_tree}} variable
  • this variable will be replaced with the folder tree of the current vault
  • useful for providing context about the vault structure to the model
• Improved: Smart Chat system message UI
  • now collapses when longer than 10 lines

patch v3.0.28

Fixed: Getting Started slideshow UX on mobile.

patch v3.0.29

• Fixed: prevented regex special characters from throwing error when excluded file/folder contains them
• Fixed: Smart Chat should return lookup context results when Smart Blocks are disabled

patch v3.0.30

• Added: Drag multiple files into the Smart Chat window to add as context
• Fixed: Smart Connections results remain stable when dragging connection from bottom of the list

patch v3.0.31

• Added: Smart Chat: "Retrieve more" button in lookup results
  • allows retrieving more results from the lookup
  • includes retrieved context in subsequent lookup to provide more context to the model
• Improved: Smart Chat: prior message handling in subsequent completions

patch v3.0.32

• Added: Anthropic Claude Sonnet 4 & Opus 4 to Smart Chat
• Improved: Smart Chat new note button no longer automatically addes open notes as context
  • Added: "Add visible" and "Add open" notes options to Smart Context selector
  • Added: "Add context" button above chat input on new chat for quick access to context selector
• Fixed: Removing an item in the context selector updates the stats
• Fixed: Smart Chat system message should render no more than once per turn

patch v3.0.33

• Improved: Context Tree styles improved by samhiatt (PR #1091)
• Improved: Smart Chat message should be full width if container is less than 600px
• Fixed: Smart Chat model selection should handle when Ollama is available but no models are installed

patch v3.0.34

• Added: Multi-modal support (images as context) using Ollama models
  • requires Ollama models that support multi-modal input like gemma3:4b

patch v3.0.37

• Fixed: Ollama max_tokens parameter should accurately reflect the model's max tokens
• Fixed: Getting Started slideshow should only show automatically for new users

patch v3.0.38

• Fixed: Smart Chat LM Studio models handling of tool_choice parameter

patch v3.0.39

• Improved: Release notes user experience to use the same as the native Obsidian release notes
  • Now uses new tab instead of modal to display the release notes
• Fixed: Reduced vector length OpenAI embedding models should be selectable in the settings

patch v3.0.40

• Added: Smart Chat: Support for PDFs as context in compatible models
  • Currently works with Anthropic, Google Gemini, and OpenAI models
  • PDFs must be manually added to the chat context. The context lookup action will not surface the PDFs because they are not embedded.
• Improved: Smart Chat: LM Studio settings
  • Added: Instructions for setting up LM Studio (CORS)
  • Removed: Unecessary API key setting

patch v3.0.41

• Fix: Bug in outlinks parsing was preventing embedding processing in some cases

patch v3.0.42

• Added: re_import_wait_time setting to Smart Environment settings
  • allows setting the time to wait before re-importing and embedding a note after it has been modified
  • WHY: improves real-time nature of the connections
• Improved: Connections view: Handling when current note hasn't been imported
  • removed notification
  • added refresh instructions to the connections view
• Improved: Connections view when no results are found
  • added "No connections found" message
  • added instructions for reloading sources from the settings
• Reduced size of bundled plugin from ~6.5MB to ~1MB (>80% reduction)
  • removed tokenizer that's only used by OpenAI embedding models
  • removed sourcemap since it's removed by Obsidian anyway
  • WHY: make the code easier to read (trust through transparency)
• Fixed: Embeddings should update when file is changed

patch v3.0.43

• Fixed: Smart Chat: Context tree connections icon should show connections in the suggestions when clicked

patch v3.0.44

• Improved: Settings descriptions for the Connections view
• Changed: Moved "muted notices" settings to the obsidian-smart-env module

patch v3.0.45

• Added: Status element for indicating embedding queue for changed notes
  • click to begin embedding otherwise waits until re_import_wait_time has passed
• Fixed: Smart Environment: only changed blocks should re-embed when the note is modified
  • Adds block has check to parse_blocks to prevent queue_embed from being called on blocks that haven't changed
• Fixed: Release notes should open in a new tab instead of relpacing the current tab
• Moved: Smart Plugins access to the obsidian-smart-env module

patch v3.0.46

• Added: Smart Chat: Include relevance score for item in context tree if retrieved from a lookup
  • allows users to see how relevant the item is to the current chat context
• Added: Snowflake Arctic Embed models to the built-in embedding adapter (transformers)
  • Snowflake/snowflake-arctic-embed-xs
  • Snowflake/snowflake-arctic-embed-s
  • Snowflake/snowflake-arctic-embed-m
• Added: Report a bug and Request a feature buttons to the settings
• Fixed: Smart Context: Tree should not split paths with slashes or hashtags within wikilinks
  • ex. [[some/path.md#subpath]] should not be split into some/path.md and subpath
• Improved: Smart Chat: Prevent trying to use folder scope in lookup when the folder provided by the AI does not exist