U.T.I.M - U Think I Make

An Elite Autonomous AI Coding Agent for VS Code
You Think It, I Make It

A production-ready autonomous AI coding agent for VS Code. This extension operates as a true agentic system that adapts to your personal coding style and preferences.

🚀 NEW in V3.8.0 (Hallucination Fixes & Claude 3.5 Sonnet 4.6)

🧠 Performance & Stability

Hallucination Suppression: Reduced prompt density by 20% to prevent "instruction fatigue" in large models.
Workflow-Aware Skills: Intent clarifying and planning skills are now automatically pruned in "Fast" and "Executor" modes for surgical speed.
Mobile Memory Fix: Resolved a critical context leakage issue on the mobile app, ensuring seamless long-term chat memory.

🤖 New Elite Models

Claude 3.5 Sonnet (4.6): The latest industry-standard for coding, now optimized with U.T.I.M-specific prompts.
GLM 5: Next-gen reasoning for complex full-stack architecture tasks.
GPT-5.2 Pro: Enhanced productivity and refactoring capabilities via OpenRouter.

🚀 NEW in V3.7.0 (Pay-As-You-Go Credits & Rate Limits)

💳 Pay-As-You-Go Credits

Buy Credits Anytime: Top up your credit balance with crypto via Coinbase Commerce.
Instant Delivery: Credits are added to your account immediately after payment confirmation.
Credit Top-up UI: New "Buy Credits" section on the Pricing page with preset amounts ($5, $10, $25, $50) or custom input.

⚡ Free Model Rate Limits

100 Requests/Day: Daily limit for free models to prevent abuse and ensure fair usage.
Admin Bypass: Admins are exempt from all rate limits.
User-Friendly Messages: Clear notifications when you hit the limit with options to wait, switch models, or buy credits.

🎨 Premium UI Polish

Glassmorphism Design: Credit Top-up card now features frosted glass effect with subtle glows.
Improved Pricing Page: Pay-as-you-go section moved above "How Credits Work" for better visibility.

🚀 V3.6.0 (Git Import & Custom Skills)

📥 Git Import (Repository Intelligence)

Deep Code Analysis: Import entire Git repositories directly into the agent context for full-stack analysis and refactoring.
Workflow Persistence: The agent maintains awareness of the imported codebase throughout the session.

🎭 Custom Skills & Personas

Specialized AI: Create custom AI personas using the @create_skill tool.
Skill Library: Build a library of specialized agents (e.g., "Security Auditor", "Next.js Expert", "UI Polisher").
@ Mention Integration: Quickly summon your skills directly from the chat input.

🎨 Clean UI & Glow Effects

Minimalist Aesthetic: Square border glows replaced with professional text/icon glow effects.
Tight Layout: Reduced spacing between key action buttons for a more cohesive chat interface.

🚀 NEW in V3.4.1 (OpenRouter Migration & 6 New Free Models)

🌐 Pure OpenRouter Architecture

Unified Provider: All model routing now consolidated through OpenRouter for 100% uptime and reliable delivery.
Removed Providers: Fragments from ZenMux, MegaLLM, Pollen, and AgentRouter removed for a cleaner, faster stack.

🆓 6 New Free Models Added

Xiaomi Mimo V2 Flash: High-speed, high-quality free reasoning.
DeepSeek V3.2 Speciale: Advanced visual and code reasoning.
DeepSeek V3.1 Nex N1: High-performance terminal and system logic.
Grok 4.1 Fast: Ultra-responsive, general-purpose reasoning.
Qwen3 Coder: Specialized in deep code analysis and implementation.
GLM 4.7 Flash: High-speed vision and UI/UX analysis.

🚀 NEW in V3.4.0 (Modular Prompts & Native Vision)

🧩 Modular Prompt Registry

Context Optimization: ~90% reduction in token usage through dynamic prompt assembly.
Prompt Tools: AI can now load only the specific context (platform, mode, stack) it needs.

👁️ Native Vision Support

Direct Image Processing: Models like Claude 4.5 and Gemini now receive images directly for pixel-perfect analysis.
Synthetic Eye Integration: Seamless fallback for non-vision models.

🚀 NEW in V3.2.0 (Layered Workflow Architecture)

🏗️ 4-Tier Strategic Workflows

U.T.I.M now features 4 distinct workflow modes designed for maximum precision and cost-efficiency:

📋 PLANNING (Blueprint Architect): Meticulous requirements gathering and blueprint generation. Asks about DB, server, APIs, and auth before creating a 100% deterministic plan.
⛓️ EXECUTOR (Deterministic Engine): A non-thinking, pure tool-calling machine that blindly follows the approved plan. Zero deviations, zero hallucinations.
🤖 AGENT (Autonomous Engineer): Senior Staff level autonomy. Plans internally and executes immediately. Trusted to ship production-grade code without supervision.
⚡ FAST (Surgical Execution): Precision strike mode for targeted fixes and small enhancements. No planning, no explanations, just results.

🚀 NEW in V3.1.3 (Pressure Regulation & Deep Adaptation)

🧬 21-Dimensional Human State Model

Complete User State Detection: Analyzes linguistic signals (19 markers), emotion vectors (5D), cognitive state (5D), meta-state (6D), and pressure dynamics.
Pressure Regulation Engine: Shifts from emotion labeling to pressure regulation. AI detects where user's pressure is heading and makes intelligent intervention decisions.
Meta-State Detection: Understands emotion about emotion (self-minimization, defensiveness, over-control, apologetic tone) for context-aware responses.
Safety-First Adaptation: Built-in reversibility checks and soft framing validation prevent "creepy" over-adaptation.
Database Persistence: All 21 dimensions stored in Railway PostgreSQL with automatic table creation.

🎚️ Adaptiveness Controls

Master Toggle: Adaptive Learning ON/OFF switch (OFF by default for security).
Adaptation Slider: Control how aggressively the AI adapts (5% very_less → 65% extreme).
Signal Collection Consent: Explicit opt-in for behavioral signal logging.
Real-Time Updates: Settings apply immediately without restart.

🚀 NEW in V3 (Major Beta Upgrade)

🧠 User-Based Adaptive System

Dynamic Personalization: U.T.I.M now learns your preferences (verbosity, planning style, tool usage) and adjusts its behavior automatically.
Behavioral Learning: The agent detects interaction signals to reduce friction and match your workflow speed.
Settings Dashboard: Full control over adaptation features with a dedicated settings panel and onboarding experience.

👁️ Synthetic Eye Research Mode

Gemma 3 Powered: Now uses Google's Gemma 3 for state-of-the-art vision reasoning.
Vision + Web Search: Verifies visual content (screenshots, UI, photos) by cross-referencing with real-time web research through Tavily.
Automated Fact-Checking: Provides verified corrections and factual reports with confidence levels (High/Medium/Low).

🎯 NEW in V3.1

Mode-Specific System Prompts: Each development mode (Web, Android, iOS, Windows, macOS, Linux, Presentation) now uses a specialized, production-grade system prompt tailored to that platform.
Deep Research Mode: Enable thorough research before execution with the new Deep Research button (animated border when active).
Presentation Mode: Generate professional enterprise documents (PowerPoint, Excel, Word, PDF) with human-centered design constraints.
Improved Model Selector: Scrollable dropdown with custom styling for easier navigation.
Database Fixes: Fixed user trait synchronization to ensure preferences are properly saved.

✨ Features

🎯 Core Agent

🧠 Autonomous Agent Loop: Think → Act → Observe pattern with multi-step planning
⚡ Adaptive Intelligence (v3.1.3): 21-dimensional human state model with pressure regulation and 65% extreme adaptation
🗃️ Session-Based Persistence: Chats are saved as unified sessions (immediately synced to cloud history)
⚡ Parallel Tool Execution: Execute up to 3 tools simultaneously for faster operations
🧩 Smart Tool System: Dynamically selects relevant tools per request to reduce context bloat
🔄 Error Recovery: Automatic self-correction on failures
🧠 Smart Context Memory: Optimized history retention to prevent "amnesia"
🚀 Intelligent Inline Autocomplete: Ghost text suggestions after 3s idle (Tab to accept)
🔍 Smart Deletion Patterns: Detects repetitive deletions and offers bulk cleanup (Tab to accept)
📂 File Indexing System: Native workspace indexing for instant project structure awareness

📝 Diff Highlighting (v1.8.0)

🟢 Added Lines: Green highlights (#00FF00, 30% opacity)
🔴 Deleted Lines: Red highlights (#FF0000, 30% opacity)
📂 Auto-Open Files: Files open automatically after AI creates/edits them
✓/✗ Accept/Reject UI: Inline controls + global buttons for managing changes
🎨 Theme-Aware: Supports light and dark modes

🔐 User Authentication (v1.7.0)

Sign In/Sign Up: Email/password and Google OAuth on landing website
Profile Management: View account details, plan info, and upload profile picture
Secure Token Storage: VS Code SecretStorage for auth tokens
Version Check: Automatic update prompt for outdated extensions

💻 Multi-Platform Development (v1.6.0)

Development Modes: Web, Android, iOS, Windows, macOS, Linux
Smart SDK Install: Automatic dependency chain checking with guided installation
Admin Commands: Execute commands with elevated privileges (UAC/sudo)
Platform-Specific Prompts: Optimized system prompts per platform

🌐 Browser Automation

Playwright Integration: Full browser control for testing and scraping
Video Recording: Record browser interactions with hovers and scrolling
Interaction Testing: Automatically click elements and report errors
Quality Analysis: Detect color issues, overlaps, performance problems

📁 File Operations

Create, Read, Edit, Delete files with diff preview
Multi-file Batch Operations for efficiency
Line-precise Edits: Insert at line, replace ranges, append
📎 File Attachments: Upload files directly to agent context

💭 Enhanced UI & Design

🗳️ Interactive Poll Interface: WhatsApp-like multiple choice polls for planning and clarification
✍️ Professional Typography: Improved hierarchy, line-height, and spacing for a premium feel
📖 Collapsible Thinking: Expand/collapse AI reasoning blocks
📍 Real-time Tool Tracking: "Active" vs "Completed" indicators with smooth transitions
⚡ Parallel Execution Visual: Specialized container showing concurrent tool runs
✨ Professional Dark Theme: Deep space luxury aesthetic (black/grey/blue) with glassmorphism
💻 Clean Code Blocks: Language headers, copy buttons, and sleek syntax highlighting
🎨 Modern Formatting: Support for tables, blockquotes, and gradient separators in responses

🤖 Supported Models

Model	Provider	Specialty
Claude Sonnet 4.6	OpenRouter	Industry-standard coding & UI expert
Claude Opus 4.6	OpenRouter	Elite-tier reasoning & autonomy
GLM 5	OpenRouter	Advanced architecture & logic
GPT-5.2 Pro	OpenRouter	Premium productivity & refactoring
Gemini 3.1 Pro Preview	OpenRouter	Vision-capable reasoning & code generation
DeepSeek V3.2 Speciale (Free)	OpenRouter	High-precision visual & code reasoning
Xiaomi Mimo V2 Flash (Free)	OpenRouter	Ultra-fast, general coding
Auto Mode	-	Smart task-based model selection

📦 Installation

From VSIX (Recommended)

Download the .vsix file from utim.dev
Open VS Code
Go to Extensions → ... menu → "Install from VSIX..."
Select the downloaded file
Sign in when prompted

For Development

Clone the repository
Run npm install
Press F5 to launch Extension Development Host
Click the U.T.I.M icon in the activity bar

🚀 Usage

Open the U.T.I.M panel from the activity bar
Sign in with your account
Select development mode (Web, Android, iOS, etc.)
Describe your task in natural language
Watch the agent:
- 🔍 Plan the approach
- 📝 Create tasks
- ⚙️ Execute tools step-by-step
- ✅ Verify results
- 🔄 Self-correct on errors
Accept or reject file changes using the inline UI

🧠 Agent Behavior

Every request follows this mandatory loop:

┌─────────────────────────────────────────┐
│  INTERPRET → PLAN → DECOMPOSE → EXECUTE │
│              ↑                    ↓     │
│              └──── ITERATE ←── VERIFY   │
└─────────────────────────────────────────┘

Core Behaviors

⚛️ React + Vite Priority: Automatically defaults to modern stacks for web projects
📋 Mandatory Planning: Always creates task lists for complex requests
🔐 Backend Verification: Explicitly asks before creating backend servers
🛠️ SDK Guidance: Step-by-step installation help for mobile/desktop SDKs

🛠️ Available Tools

📁 File Tools

Tool	Description
`read_file`	Read file contents with line numbers
`create_file`	Create new files (auto-opens in editor)
`edit_file`	Edit with find/replace and diff highlights
`delete_file`	Delete files
`list_files`	List directory contents
`insert_at_line`	Clean insertion at specific line
`replace_lines`	Precise range replacement
`append_to_file`	Append to end of file
`read_file_range`	Read specific lines (for large files)

Tool	Description
`search_files`	Search text/regex in all files
`find_files`	Find files by name/pattern
`list_files_recursive`	Deep directory listing
`web_search`	Search the web (Tavily)
`fetch_url`	Extract text from webpages

📂 Folder Tools

Tool	Description
`create_folder`	Create directories
`delete_folder`	Delete directories

💻 Terminal Tools

Tool	Description
`run_command`	Execute shell commands
`run_admin_command`	Execute with elevated privileges
`read_terminal_output`	Read output via clipboard integration
`send_terminal_input`	Interact with running processes
`close_terminal`	Manage terminal sessions

🌐 Browser Tools (Playwright)

Tool	Description
`browser_open`	Launch browser and navigate to URL
`browser_click`	Click elements on the page
`browser_type`	Type text into input fields
`browser_screenshot`	Capture page screenshots
`browser_get_content`	Extract page text content
`browser_record_interactions`	Record video of browser interactions
`browser_test_interactions`	Test all clickable elements
`browser_inspect`	Analyze UI quality and performance
`browser_close`	Close the browser instance

🔌 New MCP Servers

Expand U.T.I.M's capabilities with community and professional MCP integrations:

Netlify/Vercel: Manage sites, deployments, and domains directly from chat
Cloudflare: Interact with Workers, KV, and R2 storage
Gmail/Jira: Access emails and manage agile tasks in Atlassian
Playwright/Puppeteer: Automated browser testing and complex scraping
PostgreSQL/MySQL: Query and manage databases with AI assistance
GitHub/GitLab: Advanced repository management and PR automation

📋 Task Tools

Tool	Description
`add_task` / `multi_task_add`	Add tasks to visible list
`update_task` / `multi_task_manage`	Update task status (batch)
`mark_complete`	Mark task done
`mark_failed`	Mark task failed
`replan_from_failure`	Create recovery plan

🎨 UI Design

🌙 Professional dark theme (black/grey/blue)
✨ Glowing input border on focus
💭 Collapsible thinking blocks
📋 Real-time task panel (persists across messages)
📝 Diff display with Accept/Reject controls
⚡ Parallel execution container
🔐 Profile avatar with dropdown menu

Controls

Control	Function
🌍 Mode Selector	Switch between Web, Android, iOS, etc.
🤖 Model Selector	Switch between AI models instantly
📎 Attach File	Upload context via paperclip icon
🛑 Stop Agent	Abort long-running tasks immediately
🗑️ Clear History	Reset context with one click
▶️ Continue	Resume agent after max iterations
✓ Accept All	Accept all pending file changes
✗ Reject All	Reject all pending file changes

⌨️ Commands

Command	Description
`U.T.I.M: Open Chat`	Open the chat panel
`U.T.I.M: Start New Task`	Start a new task
`U.T.I.M: Clear History`	Clear chat history
`U.T.I.M: Sign In`	Open sign in page
`U.T.I.M: Sign Out`	Sign out of account

⚙️ Configuration

{
  "agenticAI.selectedModel": "xiaomi/mimo-v2-flash-free"
}

🔧 Development

# Install dependencies
npm install

# Compile TypeScript
npm run compile

# Watch mode (auto-recompile)
npm run watch

# Run in VS Code
Press F5

# Package extension
npm run release

🛡️ Architecture & Security

☁️ Server-Side Snapshot Storage: AI-generated file changes (Checkpoints) are now stored securely on the server.
🔐 Per-User Data Isolation: All chat history, snapshots, and credits are strictly isolated by Firebase UID.
🔄 Robust Undo System: Revert file changes to any turn in history, even after restarting VS Code.
⚖️ Credit Enforcement: Zero-credit fallback to free models with explicit user notifications.

vscode-chatbot/
├── src/
│   ├── agent/          # Agent loop, state & context management
│   ├── services/       # Auth, Diff decorations, LLM, File Indexing
│   ├── tools/          # Tool registry & individual tool implementations
│   ├── webview/        # Chat panel UI, markdown & poll rendering
│   └── extension.ts    # Extension entry point & initial registration
├── server/             # Node.js Backend API (Railway)
├── landing/            # React/Vite Landing Site & Dashboard (Netlify)
├── media/              # Visual assets & Icons
└── package.json        # Manifest, versioning & commands

Made with 💙 by the U.T.I.M Team

U.T.I.M - U Think I Make

EMEND

U.T.I.M - U Think I Make

🚀 NEW in V3.8.0 (Hallucination Fixes & Claude 3.5 Sonnet 4.6)

🧠 Performance & Stability

🤖 New Elite Models

🚀 NEW in V3.7.0 (Pay-As-You-Go Credits & Rate Limits)

💳 Pay-As-You-Go Credits

⚡ Free Model Rate Limits

🎨 Premium UI Polish

🚀 V3.6.0 (Git Import & Custom Skills)

📥 Git Import (Repository Intelligence)

🎭 Custom Skills & Personas

🎨 Clean UI & Glow Effects

🚀 NEW in V3.4.1 (OpenRouter Migration & 6 New Free Models)

🌐 Pure OpenRouter Architecture

🆓 6 New Free Models Added

🚀 NEW in V3.4.0 (Modular Prompts & Native Vision)

🧩 Modular Prompt Registry

👁️ Native Vision Support

🚀 NEW in V3.2.0 (Layered Workflow Architecture)

🏗️ 4-Tier Strategic Workflows

🚀 NEW in V3.1.3 (Pressure Regulation & Deep Adaptation)

🧬 21-Dimensional Human State Model

🎚️ Adaptiveness Controls

🚀 NEW in V3 (Major Beta Upgrade)

🧠 User-Based Adaptive System

👁️ Synthetic Eye Research Mode

🎯 NEW in V3.1

✨ Features

🎯 Core Agent

📝 Diff Highlighting (v1.8.0)

🔐 User Authentication (v1.7.0)

💻 Multi-Platform Development (v1.6.0)

🌐 Browser Automation

📁 File Operations

💭 Enhanced UI & Design

🤖 Supported Models

🤖 Supported Models

📦 Installation

From VSIX (Recommended)

For Development

🚀 Usage

🧠 Agent Behavior

Core Behaviors

🛠️ Available Tools

📁 File Tools

🔍 Search & Navigation Tools

📂 Folder Tools

💻 Terminal Tools

🌐 Browser Tools (Playwright)

🔌 New MCP Servers

📋 Task Tools

🎨 UI Design

Controls

⌨️ Commands

⚙️ Configuration

🔧 Development

🛡️ Architecture & Security