Enable PatchVaultFile to apply multiple patches in a single operation by converting oldContent and newContent from strings to arrays. Add whitespace-flexible regex matching as fallback when exact string matching fails, improving robustness for code with varying indentation. Update schemas, tool definitions, and all tests to reflect the new array-based API.
Replace the single Replan tool with three targeted alternatives:
ReviseStep, RevisePlan, and SkipStep. This gives the orchestration
agent more precise control over plan recovery — revising only the
current step, replacing all remaining steps, or skipping a step
entirely — rather than triggering a full replan each time.
Also give the orchestration agent read access to vault files so it
can resolve execution failures without routing back to the planning
agent, and update the execution agent prompt to stop it from
attempting self-recovery (gap resolution is now the orchestrator's
responsibility).
Emit tool call start events immediately when tools begin execution,
allowing UI to display contextual "thinking" messages (e.g., "Generating
note contents..." for WriteVaultFile) before tool completion. Add
Spinner to ThoughtIndicator component for visual feedback during tool
operations.
Migrate attachment storage from base64-in-JSON to separate binary files with SHA-256 naming, add automatic garbage collection, implement lazy loading via getBase64() and getMimeType() methods, normalize text MIME types, add image resizing, and update all AI provider integrations. Fix small issue that caused the chat area to scroll down on message streaming end even when scrolled up.
- Rename toolDefinitions to aiFunctionDefinitions across all AI classes
- Add aiFunctionUsageMode property to control function calling behavior
- Implement provider-specific tool_choice/tool_config based on usage mode
- Remove AIController and create BaseAgent base class
- Update ExecutionPrompt with scope of execution guidelines
- Simplify ChatArea layout update logic by removing debounce
- Add naming service completion wait in ChatService
- Replace console.warn with Exception.warn in InputService
- Delete unused AIController.ts file
- Update all tests to use new aiFunctionDefinitions property
Update AskUserQuestionExecution function definition and ExecutionPrompt to provide clearer guidance on when to ask users vs. report outcomes. Emphasizes resolving ambiguity at discovery point rather than passing uncertain outcomes to orchestrator.
- Remove unused agentExecutionDefinitions and gatedDefinitions methods
- Delete obsolete CreatePlan.ts (replaced by ExecuteWorkflow.ts)
- Rename PlanningAgentSystemPrompt to PlanningPrompt for consistency
- Fix orchestration agent promise returns and replan flag handling
- Make saveConversation calls async throughout AIController
- Simplify vault cache delete event handling
- Change debug logging from console.log to console.debug
- Fix function name references in execution and main agents
Restructure all AI function descriptions to use consistent "Call this function:" and "Do NOT use this function:" sections for improved clarity and scannability.
Track replan state to ensure plan doesn't complete prematurely when a replan
is requested. Also add CompletePlan to orchestration agent definitions and
update step progress callback on successful completion.
- Add context_for_next_step parameter to CompleteStep for passing execution history
- Add context parameter to CompleteTask for preserving task completion state
- Update OrchestrationResult to handle context propagation between steps
- Add debug color differentiation for agent types (Main, Execution, Orchestration, Planning)
- Reorganize SearchTypes from Helpers to Types directory
- Add justification requirement for execution deviations
- Support reasonable deviations in orchestration plan validation
- Refactor dependency service with TryResolve utility
- Add whitespace cleanup to Semaphore class
Restructure the AI workflow from a single-agent model to a specialized
multi-agent system with distinct roles:
- Add AgentType enum (Main, Orchestration, Planning, Execution) to define
agent specializations
- Replace AIControllerService and AIFunctionService with modular agent
classes in Services/AIServices/:
- MainAgent: Handles user interaction and delegates to orchestration
- OrchestrationAgent: Coordinates plan execution and step-by-step workflow
- PlanningAgent: Creates and revises execution plans
- ExecutionAgent: Executes individual plan steps
- AIController: Base class providing common agent loop functionality
- Create specialized prompts (OrchestrationPrompt, ExecutionPrompt) for
agent-specific behavior
- Add OrchestrationResult type to communicate workflow control decisions
(continue, abort, replan)
- Introduce agent-specific function scoping:
- ExecuteWorkflow for main agent
- CompleteTask for execution agent
- CompleteStep/Replan/CancelPlan for orchestration agent
- SubmitPlan/AskUserQuestionPlanning for planning agent
- Update BaseAIClass to use agentType property instead of isPlanningAgent
boolean for model selection
- Simplify ExecutionPlan and ExecutionStep types by removing execution
state tracking (moved to agent coordination)
- Remove PlanningEnabledAppendix and ExecutionStatus enum (superseded by
agent architecture)
- Add comprehensive integration tests for agent workflows
This architecture provides better separation of concerns, clearer agent
responsibilities, and more robust plan execution with explicit orchestration
control flow.
Add distinct AskUserQuestionPlanning and AskUserQuestionExecution functions to enable user consultation during both planning and execution phases. Update system prompt to clarify when to seek user input vs. replan. Fix plan area rendering timing and improve execution flow handling.
- Add currentProvider getter to AI classes for provider access
- Enhance planning prompt to discourage over-engineering and scope creep
- Return Unknown enum instead of throwing for invalid AI functions
- Add completion reminders to execution steps
- Fix ChatPlanArea height calculation timing
- Update tests to handle Unknown function gracefully
Introduce separate planning model setting to allow using different models for planning vs execution. Add visual countdown display when rate limits are hit, with improved retry delay parsing across providers (Claude, OpenAI, Gemini). Refactor settings tab into Views directory and enhance mobile layout for input controls.
Enhance user question display with markdown rendering for better readability.
Display formatted questions with emphasis, lists, and code snippets. Adjust
input display max-height based on platform (mobile: 15vh, desktop: 30vh).
Add new CompletePlan function with schema and validation to allow agents to explicitly signal when all execution steps are done. Update system prompt to include completion confirmation step. Refactor execution loop to recursively prompt for completion if incomplete, with MAX_EXECUTION_DEPTH limit replacing MAX_PLANNING_ITERATIONS. Remove automatic incomplete execution handling in favor of explicit agent-driven completion.
- Add planning mode toggle and AskUserQuestion function for interactive planning
- Fix Gemini cross-provider function call detection using toolId presence
- Update OpenAI naming service to handle new response format
- Improve system prompts by removing complexity gate, streamlining to action-first principle
- Add InputDisplay component and question mode to ChatInput
- Refactor execution plan error messages to show all incomplete steps
- Update tests to reflect cross-provider function call format changes
Restructure system prompt to enforce mandatory complexity evaluation before action. Replace verbose multi-step planning framework with concise gate-based decision model. Add UI for execution plan visibility. Improve planning/execution separation by blocking execution tools during planning phase. Remove cancellation indicator component. Fix conversation deletion bug preventing saves after delete. Strengthen type safety for AIFunction names. Simplify function summary format for planning agent. Update test mocks for new callback signatures.
- Move parseFunctionCall/parseFunctionResponse to ResponseHelper
- Enhance orphaned call/response filtering with detailed debug logs
- Add toolId to conversation content for better tracking
- Fix planning workflow execution mechanics and step numbering
- Remove unused planning agent appendix and detailedAppendixForPlanningAgent
- Add conversation save callbacks throughout AI controller loops
- Improve multi-agent function handling to avoid exceptions
- Update all tests to include toolId fields for proper filtering
Implement extractRetryDelay methods in Claude, Gemini, and OpenAI classes to parse provider-specific retry delay headers/responses. Update StreamingService to use these delays when available, falling back to exponential backoff. Enhance ApiError to include response headers and body for retry delay extraction.
- Add toolId matching validation for function call/response pairs
- Filter orphaned function calls and responses with warnings
- Add CancelPlan function to abort plan execution
- Fix planning response handling in execution agent
- Pass clearThoughtOnContent flag through streaming pipeline
- Move prompts from AIClasses to AIPrompts directory
- Replace centralized IPrompt injection with direct property setters on IAIClass
- Remove allowDestructiveActions parameter from streamRequest methods
- Add toolDefinitions, systemPrompt, and userInstruction properties to IAIClass
- Refactor AIFunctionDefinitions to static methods with agent-specific tool sets
- Add planning agent function definitions (CreatePlan, Replan, CompleteStep, SubmitPlan)
- Create AIControllerService to handle agent orchestration
- Add execution plan related copy strings and replacement utility
- Update all AI providers (Claude, Gemini, OpenAI) to use new architecture
Add cache control to system prompt, tools, and conversation messages to optimize API performance and reduce costs by caching frequently reused content.
Update GitHub link to point directly to project page.
Improve prompt for user suggestion during actions that require confirmation.
Update unit tests.
Bump dependency versions.
Replace errorMessage string with uploadErrors array in processAttachments
to provide structured error handling. Enhanced error messages to include
file names. Added getHeader helper method for case-insensitive header
lookups. Fixed Gemini upload to use contentType parameter instead of
Content-Type header.
- Add validation check for file ID after upload attempt
- Clear existing file ID before re-uploading to prevent stale IDs
- Replace Node.js Buffer with standard Uint8Array/ArrayBuffer for cross-platform compatibility
- Add missing Content-Type header for Gemini file uploads
- Update tests to use TextDecoder for ArrayBuffer inspection
- Move MimeTypeToFileTypes and FileTypeToMimeType mappings to dedicated FileTypeMimeTypeMapping enum file
- Add empty content validation for file attachments
- Add PDF support for Gemini
- Display attachment references in user messages with file info
- Fix OpenAI unsupported mime type message formatting
- Improve URI list handling in InputService to include text/plain
- Update chat area padding and styling for better attachment display
- Prevent drag selection on chat padding element
- Update test assertions for new unsupported mime type message format
refactor: standardize file type and MIME type handling across AI providers
Introduce centralized MimeType enum and bidirectional mappings between file types and MIME types. Update Claude, Gemini, and OpenAI implementations to use consistent MIME type validation. Add file attachment support via drag-and-drop and paste in chat input. Expand supported file type detection to include programming languages, config files, and documentation formats.
Consolidate duplicate file upload and error handling code from Claude, Gemini, and OpenAI into shared `processAttachments` method in BaseAIClass. Remove unnecessary abort service wrapping from BaseAIFileService retry logic.
- Add file ID storage and upload tracking to Attachment class
- Store provider field in BaseAIClass for file service integration
- Update Claude, Gemini, and OpenAI to use file ID references
- Replace fetch with requestUrl for native Obsidian HTTP handling
- Implement graceful upload failure handling with user notifications
- Add retry logic with default values to prevent conversation breaks
- Optimize file cache refresh to run only when attachments present
- Filter empty function responses in conversation content validation
- Change listFiles() to synchronous method returning cached IDs
- Add type predicate to ApiError.isApiError() for better type narrowing
- Remove redundant ApiError union type in retry error handling
Replace boolean flags (isFunctionCall, isFunctionCallResponse,
isProviderSpecificContent) with explicit typed properties (functionCall,
functionResponse, attachments) in ConversationContent. Introduce Attachment
class and BaseAIFileService. Update all AI providers (Claude, Gemini, OpenAI)
to use new attachment-based binary file handling with proper error handling,
retry logic, and AbortService integration.
Implement new Files API service for all providers (not yet integrated).
Excludes image/PDF content blocks from text extraction when
isProviderSpecificContent flag is set, avoiding unnecessary token
usage. Updates ReadVaultFiles documentation to clarify image/PDF
reading capabilities and adds system prompt guidance for handling
binary file references.
Add support for reading images (PNG, JPG, JPEG, WebP) and PDFs in ReadVaultFiles. Implement provider-specific formatting for binary content (Claude, OpenAI, Gemini). Update conversation structure to handle binary files separately from text responses. Add unpdf library for PDF text extraction in search.
Simplify historical tool call/result format from bracketed text to JSON with HTML comments indicating completed status. Update system prompt to clarify this is historical context, not a pattern to reproduce. Improve chat area layout performance with debouncing and reduce minimum padding. Update all tests to match new format.
- Remove unused StoredFunctionCall/Response imports from AI classes
- Replace settled flag with shouldSettle parameter for better control
- Add ResizeObserver to handle dynamic content size changes
- Bind indicator elements for accurate height calculations
- Update layout calculation to account for indicators when not settled
Consolidate convertFunctionCallToText and convertFunctionResponseToText
methods from Claude and Gemini into BaseAIClass for code reuse. Update
OpenAI to handle missing function call IDs by converting to legacy text
format, preventing undefined call_id bugs during cross-provider switches.
Implement thought signature tracking for Gemini function calls to support Gemini 3 requirements. Migrate OpenAI integration from Chat Completions to Responses API with proper input/output item handling. Add cross-provider compatibility via legacy text format fallback for conversations without thought signatures or tool IDs. Improve chat auto-scroll behavior and conversation validation. Add and update AI class and conversation tests.
- Add mobile-specific controls with Accept/Reject buttons for diff review
- Centralize user rejection/suggestion messages in AIFunctionResponse
- Update VaultService to use centralized message constants
- Style mobile buttons with proper touch targets and theme colors
- Auto-focus main view after accepting/rejecting changes on mobile
