Updates

Added & Optimized

• •Configurable Project Exploration Depth: You can now customize how deeply the context helper thinks when exploring your project files and gathering codebase information, giving you full control over speed versus thoroughness.
• •Smarter Background Research: Background research assistants now automatically adapt their thinking and reasoning depth to match your selected model, ensuring faster, more accurate discoveries during complex investigations.
• •Seamless Auto Reasoning Mode: Selecting 'Auto' mode now instantly configures the best AI model and reasoning depth for your task without extra setup steps, skipping right to coding.

Added & Optimized

• •@ Context Mentions & Autocomplete: Type '@' anywhere in your prompt to open an interactive suggestion menu. Easily pull in specific files, line ranges, functions, Git status, terminal output, and project errors with keyboard navigation and live status feedback.
• •Gemini 3.x Thinking Levels (/models): Choose how deeply the AI reasons on your tasks with simple Low, Medium, and High thinking options in the /models menu, backed by smart defaults tailored for each model.
• •Targeted Code Reading: The assistant now extracts exact functions, classes, and code sections without loading entire files, saving up to 95% on token usage and speeding up responses.
• •Smart History Pruning & Memory Retention: Keeps important project files fresh in active memory while clearing out old command logs, preventing context slowdowns in long sessions.
• •Instant Memory Bank Caching: Saves analyzed file summaries to local disk for instant sub-millisecond retrieval on repeat questions, capped safely at 5MB.
• •Quieter Multi-Assistant Teamwork: Filters background assistant updates to only active file changes, reducing inter-assistant chatter and saving tokens by up to 80%.
• •Automatic Rate-Limit Protection: Smooths out background requests to prevent unexpected API rate limit interruptions during heavy workloads.

Added & Optimized

• •Smart In-Memory Read Caching: Automatically caches recently read files across assistant tasks, instantly invalidating modified files to prevent unnecessary disk reads.

Added & Optimized

• •Enhanced Local Syntax Validation: Adds robust syntax and delimiter checks to verify code integrity and prevent incomplete or broken code changes before saving.
• •Comprehensive Developer Documentation: Detailed technical documentation and interface definitions for internal fuzzy matching and syntax validation engines.
• •Gemini Context Cache Auto-Refresh: Automatically refreshes cached AI context with a quick 500ms retry when stale or mismatched sessions are encountered.
• •Live Action Indicators: Displays real-time status indicators when reading files, modifying code, searching, or running tests so you can track progress at a glance.

Added & Optimized

• •Expanded File Reading Limit: Increased single-file reading capacity to 2,500 lines to seamlessly analyze large source files without missing details.
• •Sub-Agent Memory Protection: Safeguards workspace file blocks from historical output pruning to prevent redundant re-reading loops.
• •Smart AI History Pruning & Rate-Limit Mitigation: Exempts critical conversation turns from pruning and optimizes API retry timing to prevent rate limits.

Added & Optimized

• •Mutation-Only Peer Activity Filtering: Restricts real-time peer telemetry to active file modifications, reducing log noise.
• •Ephemeral Scratchpad Isolation: Runs disposable test and debug scripts safely inside isolated temporary directories.
• •Automatic IDE Watcher Setup: Automatically configures local project settings to prevent file watcher lag and ensure instant UI updates.

Added & Optimized

• •Smarter Error Recovery: Instantly pinpoints the root cause of build errors and command failures without repetitive guessing or loops.

Added & Optimized

• •Permissive Intent Routing & Codebase Search: Upgraded intent classification with a permissive posture that triggers deep codebase exploration for any question touching project logic, local files, architecture, debugging, or multi-workspace references, with automatic fallback guaranteeing zero missed context.
• •Parallel Multi-Agent Investigation Heuristics: Calibrated domain complexity evaluation to allocate 2–3 parallel sub-agent workers for multi-file patterns, cross-module refactors, and ambiguous scopes, defaulting to parallel search on uncertainty so full multi-agent power is always ready.
• •Multi-Workspace Sub-Path Auto-Focusing with Global Overrides: Automatically focuses tool actions on active monorepo sub-paths while intelligently detecting global tasks (such as repository audits and cross-project migrations) to expand search boundaries dynamically.
• •Hardened Path Traversal Security: Validates all aliased (@alias/) and relative paths against registered workspace boundaries with canonical path resolution to prevent accidental boundary escapes.
• •Canonical File-Lock Resolution: Unified aliased and relative paths to absolute canonical paths prior to mutex acquisition, eliminating concurrency aliasing bugs across parallel sub-agents.

Added & Optimized

• •Crystal-Clear Error Reporting: Fixed an issue where general connection or formatting hiccups were mistakenly reported as invalid API keys. The assistant now reports the exact reason for any issue with zero guesswork, never asking you to reconfigure a key that's already working.
• •Seamless Conversation Summarization: Improved how past conversations and previous work are summarized in the background, ensuring extended multi-turn sessions and resumed older chats continue smoothly without interruptions.

Added & Optimized

• •Adaptive Persona Memory Bank: The assistant now gets to know you organically over time, remembering your preferred communication style, how you like explanations formatted, and your coding habits across sessions so you never have to repeat yourself.
• •Cognitive & Decision-Making Tailoring: Automatically adapts to how you think and collaborate—whether you prefer high-level architecture designs or direct code snippets, direct recommendations or exploring trade-offs, and quick fixes or in-depth explanations. Anything to get to know you better.
• •Complete Privacy & Control (/profile): Your learned profile and notes stay strictly on your local machine. Use the new /profile command anytime to inspect what the assistant has learned, remove specific notes, or reset memory completely.
• •Witty & Natural Conversational Flow: Completely eliminated repetitive robotic disclaimers in chat mode in favor of a sharp, approachable senior-developer persona with natural humor and clean technical focus.

Added & Optimized

• •Lightning-Fast Parallel Exploration: The assistant can now search files, read code, and map dependencies all at once in parallel rather than one by one, speeding up initial project investigations by 60% to 80%.
• •Massive Token & Memory Reductions: Slashes token usage by up to 95% on large projects by summarizing background files into lightweight structural outlines, pruning older tool outputs, and automatically condensing chat history starting right after the first turn.
• •Universal Multi-Language Sandboxing: The AI can now safely run and verify code scripts across TypeScript, Python (with virtual environments), Ruby, PHP, Go, Rust, Java, and C/C++ inside isolated temporary sandboxes.
• •Streamlined Chat History & Clean Indicators: When resuming previous conversations, the chat view now hides verbose intermediate tool logs in favor of clean action badges, while permanently saving the AI's final answers.
• •Command Security & Safety Guardrails: Added strict safety protections that block accidental administrator (sudo) commands from running automatically, keeping your system secure and preventing terminal freezes.

Added & Changed

• •Smarter Multi-Tier Memory Bank: The AI can now recall past research through exact questions, slight phrasing variations, or similar topics. It automatically refreshes cached knowledge when you edit project files and keeps local cache storage strictly under 5MB.
• •Seamless Multi-Turn Continuity: The assistant now automatically stays aware of recent changes made to your workspace, allowing it to seamlessly follow up on previous edits and keep context across multiple turns.
• •Faster & More Reliable AI Routing: Upgraded the internal decision engine with structured responses, eliminating formatting errors and retry delays for faster, more dependable answers.

Fixed & Changed

• •Precise Input Lifecycle Management: Configured input handling to stop immediately when AI generation or self-correction finishes (prior to rendering and chat history persistence), preventing user keystrokes typed after generation from being mistakenly intercepted as chained prompts.
• •Automatic Queue Clearance: Implemented automatic message queue clearance upon starting and stopping to guarantee that no stale or queued messages carry over into subsequent turns.
• •Event Listener & Timeout Safety: Resolved listener capacity warnings on terminal input streams by canceling pending reattach timeouts, enforcing single listener registrations, and setting safe capacity limits.

Added & Changed

• •End-to-End Cancellation Propagation: Integrated cancellation signals throughout all helper services, intent routers, complexity evaluators, history compression routines, and syntax validators.
• •Instant Retry Abort: Added instant abort rejections to backoff delay timers and stream error recovery handlers, ensuring ongoing retry loops stop immediately when you issue a 'stop' command.

Added

• •Easy Model Switching: You can now choose Gemini 3.7 Flash anytime using the /models command.
• •Automatic Smart Selection: The system now automatically picks Gemini 3.7 Flash for heavy coding tasks to give you the best balance of speed, accuracy, and cost.

Improved

• •Advanced Symbol Extraction: Expanded code parsing support to detect Rust inherent implementation blocks and modern TypeScript class modifiers, generics, and metadata ranges.
• •Clean Terminal Symbol Formatting: Added automatic preprocessing and rendering of LaTeX arrows and math symbols into crisp Unicode characters for effortless terminal readability.

Improved

• •Custom Key Connectivity: Streamlined custom API key requests for faster, more reliable streaming responses.

Fixed

• •Terminal Paste & Stdin: Resolved issues with duplicate text echoing when pasting prompts into the terminal and improved command prompt responsiveness.

Added

• •Automated Edge-Case Testing: Added an autonomous testing tool that automatically generates edge-case inputs to discover hidden bugs and boundary failures in your code.
• •Memory Leak Inspection: Added a heap memory inspection tool to monitor memory allocations before and after code execution, catching memory leaks and growth spikes.
• •Behavioral Drift Comparison: Added a side-by-side verification tool that runs original and modified code to detect subtle output changes and regressions.
• •Interactive Terminal UI Indicators: Added real-time terminal activity spinners and formatted indicators so you can easily track diagnostic and fuzz testing operations in progress.
• •Smarter Assistant Guidance: Updated system prompt instructions to help the assistant automatically deploy fuzz testing, memory checks, and drift analysis during code verification.

Added

• •Reliable Assistant Messaging: Implemented a background message queue that safely orders and saves inter-assistant communications, preventing signal loss or data conflicts during concurrent tasks.
• •Collision-Free File Writes: Enhanced temporary file creation with unique identifiers and timestamps to ensure multiple assistants writing at once never overwrite each other's files.
• •Property-Based Testing Integration: Added support for popular testing frameworks (such as fast-check, hypothesis, and proptest) in sandboxed temporary environments to test code invariants against edge cases.
• •Targeted Code Mutation Testing: Added diff-scoped mutation verification in temporary memory buffers to ensure test suites catch code changes without filesystem locking issues on Windows, macOS, or Linux.
• •Permanent Full Chat History: Full, uncompressed transcripts of your entire chat session are now permanently saved to disk for complete session review while keeping active memory fast and efficient.
• •Interrupted Execution Summaries: When you stop a task mid-way, the assistant now displays a clear, itemized summary in the terminal showing all partial file edits made up to that point.

Added

• •Read-Before-Edit Safeguard: Implemented a safety gate that prevents the assistant from editing or writing to a file unless it has first inspected its contents, eliminating errors caused by guessed file contents.

Fixed

• •Fixed Completion Infinite Loops: Replaced text-matching completion checks with a reliable tool-call architecture, preventing the assistant from looping endlessly or clearing history when finishing a task.

New

• •Web Search Tool (perform_web_search): Enables real-time web querying and documentation retrieval directly within agent workflows.

Improved

• •Optimized Token & Context Efficiency: Streamlined context gathering tools and system prompts to maximize token efficiency during investigation.
• •Refined Model Selection & History Handling: Enhanced BYOK model configuration options and improved formatting and limits for session history.

New

• •Bring Your Own Key (BYOK) Support: You can now provide your own Gemini API key, giving you more control over your usage limits and privacy.

Improved

• •Smoother AI Performance: We've refined how the agent works to make it more robust and reliable during complex tasks.
• •Smarter Assistant Behavior: Improved how the assistant handles files and understands your code, leading to more accurate and helpful suggestions.

New

• •Memory Bank Caching: Automatically caches investigation paths, letting the agent instantly retrieve knowledge without re-running expensive multi-turn explorations.
• •Aggregated Token Telemetry: Now provides a complete, unified view of all token usage per turn, showing you exactly how much your entire agentic workflow costs, not just the final step.

Optimized

• •Surgical Cache Invalidation: The agent now only forgets investigation paths related to the files you actually modify, keeping the cache highly accurate and efficient.
• •Unified 5MB Cache Cap: Caches are now self-regulating with a strict 5MB size limit using LRU eviction, ensuring the .minovativemind folder stays extremely lightweight.

Changed

• •Clearer Settings: Updated the 'Auto' AI selection label so it's much easier to understand at a glance.
• •Faster Everyday AI: Upgraded our standard AI model to a newer, faster version. This new version automatically handles background tasks—like organizing and summarizing—keeping the entire system quick and responsive.

Added

• •Gemini 2.5 Flash Model: Added support for the gemini-2.5-flash model across the agent and configuration, offering advanced reasoning at a lower cost.
• •Token Usage Tracking: Implemented automatic token usage tracking per chat session to monitor and optimize API consumption.
• •Richer Session Metadata: Saved sessions now automatically track the active Git branch, auto-approve status, and sub-agent execution state.
• •History Archiving: Improved history persistence by archiving pruned conversation history entries to local storage instead of discarding them.

Changed

• •Sub-Agent Concurrency Limits: Limited concurrent sub-agent execution to 2 to prevent rate limiting and ensure stable parallel execution.
• •Tightened System Prompts: Refined system prompts regarding code generation limits and brevity to keep responses highly focused and concise.

Added

• •Double-Checking Work: We added a new step where the AI automatically reviews its work against its checklist to make sure nothing was missed before finishing a task.

Fixed

• •Better Teamwork: Fixed a bug where AI assistants working at the same time would accidentally overwrite each other's to-do lists, keeping everything organized.
• •Accurate File Reading: Improved how the system reads your project folders, ensuring it correctly ignores hidden or temporary files deep in subfolders.
• •Focused Assistants: Added stricter rules to keep background assistants focused exclusively on their specific task, preventing them from getting sidetracked.

Added

• •Switch AI Models on the Fly: You can now change your AI model right in the middle of a task. If a task gets too hard, easily switch to a smarter model without losing your progress.
• •Remember Your Favorite Model: The system now automatically remembers the AI model you choose, so you don't have to keep selecting it for every new chat.

Added

• •Jupyter Notebook Support: The AI can now directly read and understand Jupyter Notebook (.ipynb) files, including all of the code, notes, and results inside them.
• •Spreadsheet Support: The AI can now natively read CSV spreadsheet files and automatically convert them into clear, organized tables for easy analysis.
• •Database Support: The AI can now safely read database files (.sqlite, .db) to automatically understand how your data is structured, rather than crashing on these complex files.
• •Smart File Skipping: Added strict safety guards that prevent the AI from accidentally reading massive, unhelpful system files (like lockfiles), which saves computer memory and speeds up performance.

Fixed

• •Stronger Security: We've upgraded how the system safely reads and validates file locations behind the scenes, making the app much more secure and reliable.

Added

• •Cross-Repository Architecture: Support for workspace aliases (`@alias/`) to reference files and resources across different projects.
• •Interactive Workspace Management: Improved commands for adding, switching, and monitoring multiple active workspaces.

Added

• •Toggle Undo History: You can now turn the undo history tracker on or off from the '/revert' menu. Turning it off saves computer memory and space, which is great for very large projects.
• •Smart Code Search: The AI can now search for the actual meaning behind your code, instead of just looking for exact words. This makes it much better at finding what you need.
• •Smart AI Selection: The system now automatically picks the best AI model for your task. It uses a faster model for simple questions and a more powerful one for complex coding, saving you time and credits.
• •Plan Mode: Added a new '/plan' command. When you turn this on, the AI will investigate your project and write down a clear plan of what it wants to change before it actually touches any of your code.
• •Teamwork AI: Introduced a brand new system that allows multiple AI assistants to work on different parts of your task at the same time.
• •Safe Code Editing: Even when multiple assistants are working at once, our new safety system ensures they don't overwrite each other's changes.
• •Silent Communication: The AI assistants can now securely talk to each other in the background without cluttering up your chat screen.
• •Smart Error Fixing: If a coding error happens, the system automatically figures out which assistant caused it and sends them the error so they can fix it immediately. If they can't, the Project Manager AI steps in.
• •Teamwork Toggle: Added a new '/sub-agents' command so you can easily turn the multitasking AI team on or off.

Changed

• •Better Teamwork Coordination: Completely redesigned the core system to better manage the AI assistants when they are working together.
• •Unlimited AI Steps: Removed the limits on how many steps the AI can take, allowing it to tackle much larger and more complex tasks without stopping.
• •Better Chat Controls: Improved the '/chats' command so you can easily start new conversations and browse your chat history.
• •Smarter Memory Management: Updated the way the AI handles large files so it works perfectly with the new smart AI selection feature.

Fixed

• •Fixed Chat Glitches: Fixed an annoying bug where background tasks (like the app automatically naming your chat) would accidentally copy themselves into your text box while you were typing.

Easier Tool Integration

Minovative Mind can now connect to and use more external tools and services, making it much more capable.

• •Connects to external tools: Minovative Mind can now link up with and use a wider variety of external tools and services.
• •Automatic tool discovery: It automatically finds out what new tools can do when you set them up, so they're ready to use right away.
• •Seamless updates: If you change your tool settings, Minovative Mind updates instantly without needing a restart.

Unified Tool Access

All tools, both built-in and external ones, are now managed together, making it easier for the AI to choose and use the best tool for any task.

• •One place for all tools: External tools are now listed alongside Minovative Mind's built-in tools, giving the AI more options.
• •Smart tool usage: The system knows how to use both built-in and external tools efficiently, logging their activity for clarity.

Safer Self-Correction

To prevent the AI from accidentally making code worse, Minovative Mind now automatically checks for errors after making changes. If too many new errors appear, it will undo the changes and inform you.

• •Automatic error rollback: If the AI makes a change that introduces too many errors, Minovative Mind will automatically undo it to prevent damage.
• •Clear notifications: You'll be informed when changes are rolled back to keep your code safe.

Smarter Code Checks

Before saving any AI-generated code, Minovative Mind now performs a quick check to catch common structural mistakes (like missing brackets) across many programming languages. If a mistake is found, the AI is prompted to fix it immediately.

• •Instant syntax checks: AI-generated code is checked for basic errors (like unmatched brackets) before it's saved.
• •Automatic error correction: If syntax errors are found, the AI will try again to fix them right away.

More Efficient Code Understanding

To understand large codebases faster and more efficiently, Minovative Mind can now create concise summaries (outlines) of files in many different programming languages. This helps the AI grasp the code's structure without needing to read every single line.

• •Quick code summaries: Minovative Mind can now create helpful outlines of your code, making it easier and faster for the AI to understand.
• •Supports many languages: This outlining works for a wide range of programming languages, including TypeScript, Python, Go, and more.

Improved AI Output Handling

Minovative Mind is now better at extracting the actual code from AI responses, even if the AI accidentally wraps the code in extra formatting.

• •Smarter code extraction: Minovative Mind can now more reliably pull out the correct code from AI responses, even if they come with extra, unnecessary formatting.

Safer Messaging

Internal communications between different parts of Minovative Mind are now more secure and reliable thanks to new validation checks. This ensures data is always correct and helps prevent errors.

• •Enhanced internal communication: Messages within Minovative Mind are now validated to ensure data accuracy and prevent issues.

Smoother AI Experience

The overall AI experience has been enhanced with smoother real-time code generation, improved understanding of user requests, and smarter ways of providing relevant information to the AI.

• •Better real-time AI responses: Enjoy a more fluid and responsive experience when the AI generates code or provides information.
• •Smarter planning and context: The AI is now better at understanding your goals and finding the most relevant information to help you.
• •Refined AI instructions: Improved internal guidance helps the AI provide higher quality and more accurate results.

UI & Webview Enhancements

Various parts of the user interface and web views have been visually updated and improved for a better overall look and feel, and smoother interactions.

• •Refreshed chat interface: The chat window has been visually updated for a cleaner, more responsive look.
• •Smoother interactions: Buttons and other interactive elements now respond more effectively.
• •Better code streaming display: Real-time code generation is displayed more smoothly.
• •Updated 'What\'s New' panel: The panel now features the latest release details.