HalCode AI Assistant

Your personal AI coding partner with deep context awareness and agentic workflows - built for VS Code.

⚡ Quick Start (v3.0.17+)

Using Ollama? Just 3 steps:

Install Ollama - Download from ollama.ai
Start Ollama - Run ollama serve (or ollama signin for cloud models)
Select Model - Open HalCode settings and pick a model from the dropdown

That's it! No API keys needed. Start chatting! 🚀

🌟 Features

🚀 Ollama Support - Three Great Options! (v3.0.16+)

HalCode now supports three Ollama options to fit every budget and use case:

1️⃣ Local Ollama (Completely Free!)

Perfect for: Casual coding, learning, privacy-focused users

✅ Zero Cost - Free after model download
✅ Completely Private - Everything runs on your machine
✅ Works with AI Assistant - Great for conversation and simple tasks
✅ Recommended Models: llama3:8b, phi3, gemma:7b
❌ Limitation: File creation requires Ollama Cloud or Claude/Gemini

Setup:

ollama pull llama3:8b
ollama serve

2️⃣ Ollama Cloud (Cheapest for Heavy Use!)

Perfect for: File creation, fast responses, budget-conscious users

✅ Free Tier Available - $0/month with usage limits
✅ Lightning Fast - Datacenter-grade GPUs (1-5s responses)
✅ Works with All Features - File creation, editing, everything!
✅ Flexible Pricing - Free, $20/mo Pro, or $100/mo Max
✅ Privacy First - Ollama doesn't log your data
✅ Zero API Key Management - Just run ollama signin

Setup:

ollama signin
# Select a model in HalCode settings

Available Cloud Models:

Model	Best For	Speed
`deepseek-v3.1:671b-cloud`	⚡ Ultra-fast, instant responses	1-5s

Available Local Models:

Model	Best For	Speed	Size
`llama3:8b`	⭐ RECOMMENDED - Best balance	5-10s	4.7GB
`phi3`	⭐ RECOMMENDED - Efficient & fast	2-5s	2.3GB
`gemma:7b`	⭐ RECOMMENDED - Strong reasoning	5-10s	5GB
`deepseek-r1:14b`	Advanced reasoning	10-20s	9GB
`deepseek-r1:32b`	High-quality reasoning	20-40s	20GB
`deepseek-r1:70b`	State-of-the-art reasoning	40-80s	45GB
`deepseek-r1:671b`	Maximum reasoning capability	2-5min	400GB+

Want to try other models? Just run ollama pull <model-name> and select it in HalCode!

3️⃣ Claude & Gemini (Best Quality)

Perfect for: Complex tasks, Agentic/Multi-Agent modes, maximum reliability

✅ Highest Quality - Most capable models
✅ All Features Work - File creation, Agentic, Multi-Agent
✅ Gemini Free Tier - $0/month with usage limits
✅ Claude Paid - Best for serious development

💰 Complete Pricing Comparison

Feature	Local Ollama	Ollama Cloud Free	Ollama Cloud Pro	Claude	Gemini
Cost	Free	Free	$20/mo	Pay-per-use	Free + paid
Speed	Slow (60-120s)	Fast (1-5s)	Fast (1-5s)	Fast (2-5s)	Fast (1-3s)
AI Assistant	✅ Works	✅ Works	✅ Works	✅ Works	✅ Works
File Creation	❌ No	✅ Yes	✅ Yes	✅ Yes	✅ Yes
Agentic Mode	❌ No	🔄 Coming	🔄 Coming	✅ Works	✅ Works
Multi-Agent	❌ No	🔄 Coming	🔄 Coming	✅ Works	✅ Works
Privacy	✅ Local	✅ No logging	✅ No logging	❌ Cloud	❌ Cloud
Best For	Casual use	Budget users	Power users	Complex tasks	Free tier users

🎯 Which Option Should You Choose?

I'm on a tight budget: → Start with Ollama Cloud Free ($0) - Fast, works with file creation

I want completely free: → Use Local Ollama ($0) - But only for AI Assistant mode

I need file creation and speed: → Use Ollama Cloud Pro ($20/mo) - Best value for heavy use

I need the best quality: → Use Claude or Gemini - For complex tasks and Agentic modes

I want to try everything: → Start with Ollama Cloud Free, upgrade as needed

✅ Follow-Up File Modification Fix (v3.0.1 - AI Assistant Now Executes Follow-Up Changes! ✅)

When AI suggests modifying a file, it now actually does it!

✅ Follow-Up Modification Detection - AI Assistant detects when it suggests file changes
✅ Automatic Approval Setup - Sets up pending request for follow-up modifications
✅ Proper Execution - When you say "yes", the file actually gets modified
✅ Works with Indexed Folders - Integrates with your configured indexed folders
✅ No More Empty Promises - AI Assistant now follows through on suggestions

What This Fixes:

AI Assistant was suggesting file modifications but not executing them
Follow-up requests like "integrate this component" now actually work
Approval flow properly triggers for suggested changes
Files are written to disk when user approves

✅ Task List System & AI Assistant File Extraction Fix (v3.0.0 - Real-Time Progress! ✅)

Never feel stuck again - see exactly what's happening at every step!

✅ Real-Time Task Tracking - Tasks extracted from AI plans and displayed with checkboxes
✅ Progress Visibility - Watch tasks update from NOT_STARTED → IN_PROGRESS → COMPLETE
✅ AI Assistant Fix - Fixed file detection regex to work with all AI response formats
✅ Approval Flows - Both AI Assistant and Agentic Workflow now properly ask for approval
✅ Instant Feedback - Users see exactly what's being created and when it's done
✅ No More Confusion - Clear visual indicators prevent the "is it frozen?" feeling

What This Fixes:

AI Assistant now properly detects files from AI responses (was too strict with regex)
Both workflows show clear approval prompts before executing
Users get real-time progress updates with visual checkmarks
Task completion is tracked and displayed in chat
Fallback pattern matching for various AI response formats

Example Flow:

User: "Create a React component with styling"
AI: [Shows plan with task list]
[ ] Create component file
[ ] Add styling
[ ] Update imports

User: "yes"
AI: [Executes and updates tasks]
[x] Create component file
[x] Add styling
[x] Update imports

✅ All tasks complete!

🎯 Token Budget Enforcement (v2.6.10 - Rate Limit Fix! 🎯)

Semantic search now respects token budgets - no more 429 rate limit errors!

✅ Automatic token budget enforcement (190K limit per message)
✅ Stops adding files when token limit is reached
✅ Eliminates 429 "rate limit exceeded" errors
✅ Smarter file selection based on relevance
✅ Works with both Gemini and Claude
✅ Improved response times with optimized context

What This Fixes:

No more rate limit errors blocking your work
Consistent performance even with large codebases
Better token efficiency across all workflows
Faster responses due to optimized context size

🔄 Claude Embedding Fallback (v2.6.8 - Flexible & Resilient! 🔄)

Semantic search now works with Claude embeddings as a fallback!

✅ Automatic fallback from Gemini to Claude embeddings
✅ Works even if Gemini API is unavailable
✅ Configurable embedding provider (auto/gemini/claude)
✅ Claude-only users no longer need Gemini API key
✅ Seamless provider switching with zero downtime
✅ Better resilience and flexibility

Provider Options:

Auto (Default) - Try Gemini first, fall back to Claude
Gemini Only - Use only Gemini embeddings
Claude Only - Use only Claude embeddings

⚡ Optimized Chat Stream Speed (v2.6.7 - Smoother Responses! ⚡)

AI responses now stream at a more readable pace!

✅ Improved streaming speed for better readability
✅ Responses appear at 20ms per character (optimized from 10ms)
✅ More natural reading experience
✅ Smoother visual feedback while AI generates responses

✅ Organic AI Conversation & File Creation (v2.6.4 - Natural Flow! 🎯)

Claude and Gemini now operate more organically between chat and file creation/editing!

✅ Seamless conversation flow between chat and file operations
✅ Natural planning before file creation/editing
✅ Organic context switching between chat and code generation
✅ Both Claude and Gemini work smoothly together
✅ Full support for both providers in all workflows

Provider Support: | Feature | Claude | Gemini | |---------|--------|--------| | AI Assistant (File Creation) | ✅ ORGANIC! | ✅ ORGANIC! | | Agentic Mode | ✅ Works | ✅ Works | | Regular Chat | ✅ Works | ✅ Works |

🧠 Improved Gemini Memory (v2.6.1 - Better Context! 💭)

Now Testing & Improving One Feature at a Time

Gemini now remembers your entire conversation history, just like Claude! This means:

✅ Gemini recalls earlier messages in the chat
✅ Better context awareness across multi-turn conversations
✅ Consistent behavior between Claude and Gemini providers
✅ Improved agentic workflow reliability

Note: We're actively testing and improving AI Assistant features one at a time to ensure the best user experience. Each improvement is published incrementally.

🔄 Checkpoint System (v2.6.0 - Undo Changes! 🎯)

Available in: AI Assistant Workflow Only

Save Snapshots - Automatic checkpoints after each file operation
Revert Anytime - Click [Revert] to undo changes and go back to any checkpoint
No Manual Cleanup - No need to manually delete files you don't like
Safe Experimentation - Try different approaches without fear
Clear Tracking - See exactly what changed at each checkpoint

Example Workflow:

User: "Create a shopping cart, payment system, and user profile"
AI: "⚙️ Creating Cart.jsx... ✅ Done!"
    [CHECKPOINT 1: Created Cart.jsx] [↩️ Revert]
AI: "⚙️ Creating Payment.jsx... ✅ Done!"
    [CHECKPOINT 2: Created Payment.jsx] [↩️ Revert]
AI: "⚙️ Creating Profile.jsx... ✅ Done!"
    [CHECKPOINT 3: Created Profile.jsx] [↩️ Revert]

User: "I don't like Profile.jsx, redo it differently"
User: Clicks [↩️ Revert] on CHECKPOINT 2
→ Profile.jsx is deleted
→ Conversation continues from CHECKPOINT 2
→ User can ask AI to try a different approach

Note: Checkpoints are only available in the AI Assistant workflow. Multi-Agent and Agentic workflows do not create checkpoints.

💬 AI Assistant File Writing (v2.5.0 - Game Changer! 🎯)

🚫 NEVER Write Code in Chat - All code written directly to disk, no pasting needed
✅ ALWAYS Ask for Approval - "Should I go ahead?" before creating files
⚙️ Natural Progress Messages - "Creating Cart.jsx... ✅ Done!" (conversational, not robotic)
🧠 Remember Context Automatically - No need to click "reply on last message"
- AI remembers what it created in this conversation
- Naturally references previous work
- Internal task list tracks all changes
📝 Task Tracking - Preserves context across multiple messages
🎯 Perfect Workflow - Works seamlessly with Multi-Agent and Agentic Agent

Example:

User: "Add a shopping cart component"
AI: "Great! Let me add that for you..."
User: "yes"
AI: "⚙️ Creating Cart.jsx... ✅ Done!"
AI: "⚙️ Creating Cart.css... ✅ Done!"
AI: "All set! What's next?"

🤖 Enhanced Agentic Workflow (v2.4.0 - Production Ready! 🚀)

📊 Detailed Progress Tracking: See exactly what the AI is doing at each stage
- Step-by-step progress indicators (Analyzing → Generating → Validating → Applying)
- File-by-file updates showing each file being written
- Clear visual feedback with animated spinners and checkmarks
🔄 Stay in Agentic Mode: Continue working seamlessly after task completion
- Post-completion menu with helpful next actions
- Install dependencies, run tests, start dev server - all from the AI
- Context preservation - AI remembers what it just built
💡 Better User Experience: More transparency and control
- See which files are being created/modified in real-time
- Success confirmations for each operation
- AI suggests logical next steps based on what was built
⚡ Improved Workflow: No more black box - know exactly what's happening
- Clear status messages at every stage
- Progress indicators show active work
- Seamless transitions between tasks

⚠️ Proactive Code Analysis (v2.3.0! 🚀)

🤖 AI Watches Your Code: Automatically analyzes code as you type
⚡ Real-Time Warnings: Yellow squiggly lines show potential issues
💡 Quick Fixes: Click the lightbulb to apply AI-suggested fixes
🔍 Smart Detection: Missing imports, error handling, deep nesting, long functions
📊 Professional UI: Looks like native VS Code diagnostics
🎯 Hover for Details: See explanations and suggestions on hover
⚙️ Non-Intrusive: Only shows when there are actual issues
🆓 Free Tier: 10 analyses/day | 💎 Pro: Unlimited

💡 Inline Code Completion (v2.2.4! ⚡)

✨ GitHub Copilot-Style Completions: Get code suggestions as you type
🆓 FREE with Gemini Flash: Uses Google's free Gemini API (no cost!)
⚡ Smart & Fast: 300ms debounce, 5-minute cache, <500ms latency
🧠 Context-Aware: Analyzes 20 lines before + 5 lines after cursor
📝 Multi-Line Suggestions: Up to 3 lines of intelligent code completion
🎯 Works Everywhere: All file types supported
⚙️ Fully Configurable: Enable/disable, adjust timing, customize behavior

🔧 Enhanced Terminal Integration (NEW in v2.2.4! 🖥️)

▶️ Run Commands from AI: AI can execute terminal commands with your approval
📊 Real-Time Output: See command output streaming live
✅ Interactive Approval: User approves commands before execution
🛡️ Safety Checks: Prevents dangerous commands (rm -rf, format, etc.)
🔍 Error Parsing: Automatically detects and highlights errors
🚀 Smart Detection: Auto-detects npm, yarn, pnpm, pip, cargo

🔍 Workspace-Wide Search (NEW in v2.2.4! 🔎)

⚡ Search ALL Files Instantly: Not limited to indexed folders
🎯 Fuzzy Search: Find code even with typos
🔧 Smart Search Types:
- Find API endpoints (Express, Spring, Flask, etc.)
- Find function definitions
- Find class definitions
- Find imports/requires
- Find TODO/FIXME comments
📊 AI-Friendly Output: Results formatted for AI consumption
📁 Grouped by File: Easy to navigate results with context

🗣️ Conversational AI Agents (v2.2.3 💬)

👥 Human-Like Communication: AI agents now talk like real colleagues, not robots
🎯 Professional & Warm: Maintains expertise while being approachable and friendly
💭 Natural Conversations: Uses "I" and "you" naturally, explains reasoning conversationally
🤝 Contextual Responses: Agents acknowledge your requests before diving into work
✨ Personality-Driven: Each agent (Architect, Developer, QA) has distinct communication style
🔄 Smooth Handoffs: Natural transitions between agents ("Let me hand this over to QA...")
📢 No More Robot Speak: Removed formal "SYSTEM" messages and robotic status updates
💡 Better Understanding: Natural language makes intent and progress clearer

🚀 Performance & Scalability (v2.2.2)

⚡ Smart File Chunking: Large files automatically split into manageable chunks (70% memory reduction)
🧠 Persistent Embedding Cache: Embeddings saved to disk for instant semantic search across sessions
📊 Progressive Indexing: Priority-based file processing (high priority files first)
⚠️ Token Budget Protection: Automatic limits prevent memory issues on huge projects
10x Faster Searches: Pre-computed embeddings make semantic search nearly instant
Scalable: Now handles projects with 10,000+ files efficiently
Smart Summarization: Very large files (>500KB) summarized with signatures extracted
Detailed Statistics: See exactly what was indexed and what was skipped

🎨 Code Generation Intelligence (v2.2.0)

🧠 Context-Aware Code Generation: AI automatically includes relevant context and patterns from your codebase
📦 Smart Import Suggestions: Detects missing imports and suggests correct paths with confidence levels
🔧 Auto-Fix Imports: Automatically adds missing imports to your file with one click
🎨 Pattern Recognition: Learns your coding style (indentation, semicolons, naming conventions)
🔄 Refactoring Intelligence: Suggests extract function/component, detects duplication, analyzes impact
🔍 Duplicate Code Detection: Finds similar code patterns across your codebase (80%+ similarity)
🎯 Quick Actions: Five new buttons for intelligent code generation:
- 📦 Suggest Imports - Find and suggest missing imports
- 🔧 Fix Imports - Automatically add missing imports
- 🎨 Analyze Patterns - Show code style and pattern analysis
- 🔄 Suggest Refactoring - Get refactoring suggestions for selected code
- 🔍 Find Duplicates - Find similar code patterns
Framework-Aware: Detects React, Vue, Angular, Svelte and generates appropriate code
Style Matching: Generates code that matches your project's existing style

🧠 Real-time Dependency Analysis (v2.0.0)

⚡ Real-time File Watching: Automatic detection of file changes with incremental re-indexing
📊 Dependency Graph: Visual Mermaid diagrams showing import/export relationships
💥 Impact Analysis: Risk assessment (Low/Medium/High/Critical) with affected file counts
🔍 Reference Finder: Find all files that import a specific file instantly
🎯 Quick Actions: Four buttons for instant dependency analysis:
- 📊 Dependencies - Show what this file imports
- 🔍 References - Show what files import this file
- 💥 Impact - Analyze risk and affected files
- 🕸️ Graph - Visual dependency diagram
Smart Recommendations: Get actionable advice based on impact analysis
Circular Dependency Detection: Identifies problematic dependency chains
Cross-platform: Works on Windows, Mac, and Linux with proper path handling

💬 Persistent Chat History & Conversations

Multiple Conversations: Create unlimited separate conversations, each with its own context
Auto-Save: All conversations automatically saved and restored across VS Code restarts
Conversation List: Browse all your conversations in the sidebar with timestamps
Click to Load: Instantly load any previous conversation to continue where you left off
Auto-Generated Titles: Conversations automatically titled from your first message

🔍 Search & Export

Search Conversations: Powerful search across all conversation titles and message content
Export to Multiple Formats:
- 📄 Markdown (.md) - Perfect for documentation and sharing
- 📦 JSON (.json) - For backup and programmatic access
- 🌐 HTML (.html) - Styled webpage for browser viewing
Clean Filenames: Automatically sanitized filenames for easy organization

🧠 Deep Context Awareness

200K+ Token Context: Understands your entire codebase, not just snippets
Style Guide Integration: Automatically reads and follows your styleguide.md, CLAUDE.md, or .github/copilot-instructions.md
Project Structure Analysis: Knows your file organization and dependencies
Related Files Detection: Understands imports and file relationships
Real-time Updates: Index automatically updates when files change

🔀 Git Integration

AI-Generated Commit Messages: Analyzes your changes and generates meaningful commit messages
Git Status at a Glance: Branch indicator in header with real-time change detection
View Changes in Chat: See diffs directly in the chat interface with syntax highlighting
Commit History: View recent commits without leaving the chat
Quick Git Actions: Status, Commit, Changes, and History buttons for instant access
Smart Commit Workflow: AI analyzes diffs, suggests commit message, waits for approval
Conventional Commits: Follows standard format (feat, fix, docs, refactor, etc.)
Optional Feature: Only shows when Git repository is detected - non-intrusive!

🤖👥 Multi-Agent Collaboration (v1.0.22)

Three AI Agents Working Together: Complex tasks handled by specialized agents
- 🏗️ Agent 1 (Architect): Plans architecture, designs file structure, identifies dependencies
- 💻 Agent 2 (Developer): Reviews plan and implements code for each file
- 🧪 Agent 3 (QA Engineer): Reviews implementation, writes tests, identifies issues
User Approval Required: Review the plan before implementation begins
Automatic File Creation: Generated files automatically saved to your indexed folders
Real-Time Progress: Watch agents collaborate with live status updates
Visual Agent Cards: See each agent's status and current task in real-time

🤖 Agentic Workflows

Plan Before Execute: AI creates a detailed plan and waits for your approval before making changes
Multi-Step Refactoring: Handles complex, multi-file refactors with confidence
Test-First Approach: Suggests tests before implementation
Minimal Changes: Only modifies what's necessary
Terminal Command Execution: Automatically proposes commands like npm install after file changes (requires approval)

💻 Interactive Chat Interface

Sidebar Chat: Always-accessible AI assistant in your VS Code sidebar
Context-Aware Responses: Include current file context with one click
Multiple AI Providers: Switch between Claude and Gemini with separate API keys
Code Analysis: Right-click any file to get instant AI analysis

💬 Conversational AI Experience

Experience AI agents that communicate like real colleagues:

🏗️ Architect Agent:

"Alright, I've analyzed your request and here's what I'm thinking we should do...

I'll need to create about 15 files for this. The structure will be component-based
so it's easy to maintain.

What do you think? Should I hand this off to the developer to build it out?"

💻 Developer Agent:

"Alright, I've reviewed the plan and I'm ready to implement this. Let me build
this out for you...

Done! I've implemented everything from the plan. Let me hand this over to QA
to make sure everything looks good."

🧪 QA Agent:

"Nice work! I've reviewed everything and it's looking solid. All the files are
in place, dependencies check out, and the code is clean.

This is ready to go! 🎉"

⚙️ Highly Customizable

Custom System Prompts: Define your AI's personality and coding style
Multiple AI Providers: Support for Claude and Gemini with independent API keys
Configurable Models: Choose the best model for your needs
Your Coding Style: Teach the AI your preferences once, use forever

🚀 Getting Started

1. Installation

Open VS Code
Press Ctrl+Shift+X (or Cmd+Shift+X on Mac) to open Extensions
Search for "Personalized AI Coding Assistant"
Click Install

2. Set Your API Keys

Add Claude API Key (Optional)

Press Ctrl+Shift+P (or Cmd+Shift+P on Mac)
Type "AI Assistant: Set API Key"
Select Claude (Anthropic)
Enter your Claude API key (starts with sk-ant-)

Add Gemini API Key (Optional - FREE TIER!)

Press Ctrl+Shift+P
Type "AI Assistant: Set API Key"
Select Gemini (Google)
Enter your Gemini API key

Note: You can add both keys and switch between them anytime! Each provider has its own separate key storage.

Setup Ollama Cloud (Optional - FREE & FAST!)

Sign up at ollama.com (free account)
Open terminal and run:
```
ollama signin
```
Follow the prompts to authenticate with your ollama.com account
Done! Your device is now linked to your Ollama Cloud account
In HalCode settings, select any cloud model from the dropdown

That's it! No API keys to manage - Ollama handles authentication automatically.

Setup Local Ollama (Optional - COMPLETELY FREE!)

Recommended Models for HalCode:

llama3:8b - ⭐ Best overall, great instruction following
phi3 - ⭐ Excellent quality, very efficient
gemma:7b - ⭐ Strong reasoning, good code generation

Setup Steps:

Pull a model:
```
ollama pull llama3:8b
```
Start Ollama:
```
ollama serve
```
In HalCode settings, select your model from the dropdown

Important Notes:

✅ Works great with AI Assistant mode (conversation, simple tasks)
❌ Does NOT work with file creation (use Ollama Cloud instead)
✅ Completely free after download
✅ Runs entirely on your machine (100% private)
⚠️ Slower than cloud options (60-120 seconds per response)

3. Configuration

Configure Your System Prompt (Optional)

Press Ctrl+Shift+P
Type "AI Assistant: Configure System Prompt"
Enter your personalized instructions

Set Your API Key (Legacy)

Press Ctrl+Shift+P (or Cmd+Shift+P on Mac)
Type "AI Assistant: Set API Key"
Select your provider and enter the API key

Configure Your System Prompt (Optional)

Press Ctrl+Shift+P
Type "AI Assistant: Configure System Prompt"
Enter your personalized instructions

Example System Prompt:

You are my Senior Engineering Partner, specialized in TypeScript and React.

MODE: Undoubted Expert. Respond with certainty and precision.

AGENTIC BEHAVIOR:
1. Plan First: Outline steps before coding
2. Wait for Approval: Don't code until I say "GO"
3. Minimal Change: Only modify what's necessary
4. Tests First: Suggest tests before implementation

CODING STYLE:
- Language: TypeScript with strict mode
- Use functional components and hooks
- Prefer composition over inheritance
- Include JSDoc comments for public APIs
- Use meaningful variable names

4. Configure Inline Completion (Optional)

Inline completion is enabled by default with these settings:

{
  "halcode.inlineCompletion.enabled": true,
  "halcode.inlineCompletion.debounceMs": 300,
  "halcode.inlineCompletion.maxLines": 3
}

To customize:

Press Ctrl+, to open Settings
Search for "halcode inline"
Adjust settings as needed

5. Index Your Project Folder

Important: To use dependency analysis and context features, you need to index your project folder first!

Press Ctrl+Shift+P
Type "AI Assistant: Index Folder for AI Operations"
Select your project folder
Wait for indexing to complete (shows progress notification)
NEW! Click the X button to cancel indexing anytime

What gets indexed:

All code files (JS, TS, Python, Java, etc.)
Project structure and file relationships
Import/export statements
File dependencies

Real-time Updates:

Index automatically updates when you edit, create, or delete files
No need to re-index manually!

5. Usage

Open the Chat

Click the AI icon in the Activity Bar (left sidebar)
Or press Ctrl+Shift+P and run "AI Assistant: Open Chat"

Use Inline Completion (NEW! ⚡)

Start typing in any code file
Wait 300ms after you stop typing
See gray suggestion appear inline
Press Tab to accept the suggestion
Keep typing to ignore it

Tips:

Works best with function definitions, variable declarations, and common patterns
Uses FREE Gemini Flash API (no cost!)
Caches common patterns for 5 minutes (faster responses)
Disable anytime in settings: halcode.inlineCompletion.enabled

Run Terminal Commands (NEW! 🖥️)

Press Ctrl+Shift+P
Type "AI Assistant: Run Terminal Command"
Enter your command (e.g., npm test)
Click "Yes" to approve execution
See output in a new document

Or let the AI run commands:

Ask the AI: "Run npm install"
AI will request approval before executing
You see the full output in the chat

Search Your Workspace (NEW! 🔎)

Press Ctrl+Shift+P
Choose from:
- "AI Assistant: Search Workspace" - Search for any pattern
- "AI Assistant: Find API Endpoints" - Find all API routes
- "AI Assistant: Find TODOs" - Find all TODO/FIXME comments
See results grouped by file with context

Or ask the AI:

"Search for all API endpoints"
"Find all TODO comments"
"Search for function definitions"

Analyze Dependencies (v2.0.0 🎉)

Open any file in your indexed folder
Click the dependency buttons in the chat interface:
- 📊 Dependencies - See what this file imports
- 🔍 References - See what files import this file
- 💥 Impact - Get risk analysis and affected file count
- 🕸️ Graph - View visual dependency diagram

Example Impact Analysis:

## 💥 Impact Analysis: App.js

Risk Level: 🟢 LOW
Impact Score: 2 files affected

### 📊 Dependencies
- Direct: 6 files
- Indirect: 0 files
- Total: 6 files this file depends on

### 🔗 Dependents
- Direct: 2 files
- Transitive: 0 files
- Total: 2 files depend on this file

### 💡 Recommendation
✅ Low risk. Safe to modify with standard testing.

Create a New Conversation

Click the "✨ New Chat" button in the chat header
Start a fresh conversation with a clean slate
Your previous conversation is automatically saved

Manage Your Conversations

Look at the "Conversations" panel in the sidebar
See all your conversations with timestamps
Click any conversation to load it instantly
Each conversation maintains its own context

Search Your Conversations

Click the 🔍 Search icon in the Conversations panel header
Enter a search term (searches titles and message content)
Results filter in real-time
Click a result to load that conversation

Export a Conversation

Right-click any conversation in the Conversations panel
Click "Export Conversation"
Choose format:
- Markdown (.md) - Great for documentation
- JSON (.json) - For backup and sharing
- HTML (.html) - View in your browser
File saves to your chosen location

Switch AI Providers

Look at the provider dropdown in the chat header
Select Claude or Gemini
Instantly switches to the selected provider
Both providers work independently with their own API keys

Analyze Current File

Open any file
Press Ctrl+Shift+P
Run "AI Assistant: Analyze Current File"

Refactor with Plan

Open the file you want to refactor
Press Ctrl+Shift+P
Run "AI Assistant: Refactor with Agentic Plan"
Describe your refactoring goal
Review the plan
Click "GO" to execute

Use Git Integration (v1.0.23+)

Git features automatically appear when you're in a Git repository!

See Git Status: Look for the branch indicator in the chat header (top right)
- Shows current branch name
- Red dot (●) indicates uncommitted changes
- Click to see full Git status
Quick Git Actions: Use the Git buttons below the input area:
- 🔀 Status - View current Git status with changed files
- 📝 Commit - AI generates commit message from your changes
- 📊 Changes - View diffs with syntax highlighting
- 📜 History - See recent commits
AI Commit Workflow:
1. Make code changes
2. Click 📝 Commit button
3. AI analyzes changes and suggests commit message
4. Review the suggested message and file list
5. Reply "yes" or "commit" to proceed
6. Done! Changes committed with meaningful message
No Git Repository?: Git features automatically hide - no setup needed!

🚀 Three-Workflow Approach (v2.5.0+)

HalCode now supports three complementary AI workflows for maximum productivity:

1️⃣ Multi-Agent Workflow - Large Projects

Architect plans the structure
Developer implements the code
QA reviews and tests
Best for: Complex projects, team collaboration, thorough planning
Supported Providers: Claude, Gemini
Ollama Support: Coming soon (experimental phase)

2️⃣ Agentic Agent Workflow - Automation & Verification

Automated step-by-step execution
Detailed progress tracking
Terminal command support (npm install, tests, etc.)
Error detection and fixing
Best for: Running npm install, testing, verification, automation
Supported Providers: Claude, Gemini
Ollama Support: Coming soon (experimental phase)

3️⃣ AI Assistant Workflow - Details & Polish

Natural, conversational tone
Files written directly to disk
Context preserved across messages
No code in chat
Best for: Filling in details, small features, refinements, follow-ups
Supported Providers: Claude, Gemini, Ollama (local & cloud) ✅

🎯 Provider Support Matrix

Feature	Claude	Gemini	Ollama Local	Ollama Cloud
AI Assistant	✅ Full	✅ Full	✅ Full	✅ Full
Agentic Mode	✅ Full	✅ Full	❌ Not yet	🔄 Planned
Multi-Agent	✅ Full	✅ Full	❌ Not yet	🔄 Planned
Cost	Pay-per-use	Free tier + paid	Free	Free + paid
Speed	Fast	Fast	Slow	Fast
Privacy	Cloud	Cloud	Local	Cloud

Legend:

✅ Full = Fully supported and tested
🔄 Planned = Coming in future release
❌ Not yet = Not supported (use Claude/Gemini instead)

🎯 Recommended Workflow

For Building New Projects:

Use Multi-Agent (Claude/Gemini) to create the large project structure
Use Agentic Agent (Claude/Gemini) to run npm install and verify the build
Use AI Assistant (any provider) to fill in details and add features

For Quick Changes:

Use AI Assistant with your preferred provider (Claude, Gemini, or Ollama)

For Complex Refactoring:

Use Multi-Agent (Claude/Gemini) to plan the refactoring
Use Agentic Agent (Claude/Gemini) to execute changes
Use AI Assistant (any provider) to polish and finalize

For Budget-Conscious Users:

Use Ollama Cloud Free with AI Assistant mode
No API keys needed, just run ollama signin
Perfect for casual coding tasks

💡 Best Practices for AI Assistant

🎯 Recommended Workflow for Best Results

For Quick Changes (AI Assistant Direct):

Ask AI Assistant to make a change
If AI instantly says "I'll create/edit X files", reply "yes"
AI writes files directly to disk ✅

For Larger Projects (Use Agentic Agent):

Ask AI Assistant for a detailed plan
AI provides a comprehensive plan with multiple steps
Reply to the plan message and type "please implement"
Click the "Agentic" button in the toolbar
Agentic Agent reviews the plan and asks for approval
Reply "yes" to execute
Agentic Agent handles all the work automatically ✅

Example Workflow:

You: "What can we do to improve this site?"
AI: [Provides detailed improvement plan]
You: [Click Reply on that message]
You: "please implement"
You: [Click Agentic button]
Agentic: "Here's what I'll do... Should we proceed?"
You: "yes"
Agentic: [Executes entire plan automatically] ✅

Natural Conversation Flow

AI Assistant now feels like talking to a colleague
No need to click "reply on last message" - context is preserved
Ask follow-up questions naturally
AI remembers what it created
NEW: Gemini now remembers entire conversations (v2.6.1)

Keep the AI on Track

When using Agentic Agent mode, if something interrupts the flow or you encounter an issue:

Click "Reply on last message" - This keeps the conversation context
Add context to your message - Explain what happened or what you need
Send the message - The AI will continue from where it left off

Example:

AI creates files and asks "What would you like me to do next?"
You get an error when running npm install
Instead of starting a new conversation:
- Click "Reply on last message"
- Type: "I got an error: [paste error]. Can you help me fix this?"
- Send it
- AI continues in the same context and helps you fix it

Why This Matters

Preserves Context: AI remembers what it just built
Faster Fixes: No need to re-explain the project
Better Flow: Keeps the agentic workflow going smoothly
Fewer Mistakes: AI has full context of what was done

Pro Tips

✅ Use "Reply on last message" for follow-up questions
✅ Paste full error messages for better debugging
✅ Ask for specific fixes, not just "it doesn't work"
✅ Let the AI suggest next steps from the post-completion menu
❌ Don't start a new conversation mid-task
❌ Don't lose context by asking unrelated questions

📋 Commands

Command	Description
`AI Assistant: Open Chat`	Open the AI chat interface
`AI Assistant: Analyze Current File`	Get AI analysis of the active file
`AI Assistant: Refactor with Agentic Plan`	Perform planned refactoring
`AI Assistant: Configure System Prompt`	Set your personalized instructions
`AI Assistant: Set API Key`	Configure your AI provider API key

⚙️ Settings

Setting	Description	Default
`personalizedAI.apiProvider`	AI provider (claude, gemini, openai)	`claude`
`personalizedAI.model`	Specific model to use	`claude-3-5-sonnet-20241022`
`personalizedAI.maxContextTokens`	Maximum context tokens	`150000`
`personalizedAI.agenticMode`	Enable plan-before-execute workflow	`true`
`personalizedAI.primaryLanguage`	Your primary programming language	`TypeScript`

🎯 Best Practices

1. Create a Style Guide

Create a styleguide.md or CLAUDE.md in your project root:

# Project Style Guide

## Code Style
- Use TypeScript strict mode
- Prefer functional programming
- Maximum line length: 100 characters

## Naming Conventions
- Variables: camelCase
- Classes: PascalCase
- Constants: UPPER_SNAKE_CASE

## Architecture
- Feature-based folder structure
- Separate business logic from UI
- Use dependency injection

2. Use Context Wisely

Click "Send + Context" to include current file context
The AI will read your style guides automatically
Keep related files in the same directory

3. Leverage Agentic Mode

For complex changes, use "Refactor with Agentic Plan"
Review the plan carefully before approving
The AI will show you a diff before applying changes

🔒 Privacy & Security

API Keys: Stored securely in VS Code's secret storage
Code Privacy: Your code is only sent to the AI provider you choose
No Telemetry: This extension doesn't collect or send usage data

🛠️ Development

Want to customize or contribute?

# Clone the repository
git clone <your-repo-url>

# Install dependencies
npm install

# Compile TypeScript
npm run compile

# Run in development mode
# Press F5 in VS Code to launch Extension Development Host

📝 License

MIT License - See LICENSE file for details

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

💡 Tips

Start Simple: Begin with basic chat interactions to understand the AI's style
Refine Your Prompt: Iterate on your system prompt to match your preferences
Use Context: Always include context for better, more relevant responses
Review Changes: Always review AI-generated code before committing
Iterate: The AI learns from conversation - don't hesitate to ask for refinements
Use Git Integration: Let AI generate meaningful commit messages - saves time and improves commit quality!

🆘 Troubleshooting Ollama Cloud

"Model not found" error

Solution: Run ollama signin in your terminal to authenticate with Ollama Cloud

"Rate limit exceeded" error

Solution: You've hit the free tier usage limits. Either:

Wait for the hourly/weekly limit to reset
Upgrade to Ollama Cloud Pro ($20/mo) for higher limits
Use a local model instead

Cloud model is slow

Solution:

Check your internet connection
Try a smaller model (e.g., deepseek-r1:1.5b)
Upgrade to Ollama Cloud Pro for faster speeds

"Connection refused" error

Solution: Make sure Ollama is running:

ollama serve

🆘 Support

Having issues? Please open an issue on GitHub with:

VS Code version
Extension version
Steps to reproduce
Error messages (if any)
Which AI provider you're using (Claude, Gemini, or Ollama)

📦 Version History

v3.0.11 (Latest) - Ollama Cloud & Reasoning Models 🚀

NEW: DeepSeek-V3.1 Cloud support - Ultra-fast instant responses via Ollama cloud API
NEW: Full DeepSeek-R1 reasoning model support (1.5B through 671B variants)
IMPROVED: Enhanced Ollama integration with better error handling
IMPROVED: Model loading detection and pre-loading for reasoning models
IMPROVED: Increased timeout to 10 minutes for complex reasoning tasks
IMPROVED: Better error messages for Ollama connectivity issues
FEATURE: Easy model switching between fast (V3.1 Cloud) and reasoning (R1) models

v3.0.10 - Ollama Local Model Support 🦙

NEW: Ollama integration for running AI models locally without API keys
NEW: Support for multiple open-source models (Llama 2, Mistral, CodeLlama, Phi, Gemma)
NEW: Model switching capability between different AI providers
IMPROVED: Conversation history with local model support
FEATURE: Free, private, offline-capable AI inference

v3.0.0 - Task List System & AI Assistant Fix ✅

NEW: Real-time task tracking extracted from AI plans
NEW: Progress visibility (NOT_STARTED → IN_PROGRESS → COMPLETE)
FIXED: AI Assistant file detection regex for all response formats
IMPROVED: Approval flows for both AI Assistant and Agentic workflows
IMPROVED: Instant feedback with visual checkmarks

v2.6.5 - Bug Fixes & Improvements 🐛

FIXED: Agentic Agent menu persistence bug (menu no longer reappears after closing)
FIXED: AI Assistant file approval flow (now consistently asks for approval before writing files)
IMPROVED: Better keyword detection for file creation requests (works with natural language like "change color")
IMPROVED: Two-step approval flow for AI Assistant (analyze → ask approval → write files)
NOTE: Next features in development: AI Assistant task lists, phased task execution, chat history view fixes

v2.6.4 - Organic AI Conversation 🎯

IMPROVED: Claude and Gemini now operate more organically between chat and file creation
IMPROVED: Natural conversation flow when switching between chat and code generation
IMPROVED: Seamless context awareness across chat and file operations

v2.6.1 - Gemini Memory Improvements 💭

IMPROVED: Gemini now remembers entire conversation history (like Claude)
IMPROVED: Better context awareness across multi-turn conversations
IMPROVED: Consistent behavior between Claude and Gemini providers
IMPROVED: Enhanced agentic workflow reliability with Gemini
NEW: Fallback handling for isolated messages in agentic workflows
NOTE: Currently testing and improving features one at a time for best UX

v2.6.0 - Checkpoint System 🎯

NEW: Checkpoint system for AI Assistant workflow (undo changes anytime!)
NEW: Interactive [↩️ Revert] buttons on each checkpoint
NEW: Safe experimentation - try different approaches without fear
NEW: Automatic checkpoints after file operations
IMPROVED: Better error handling for checkpoint revert operations
NOTE: Checkpoints available in AI Assistant workflow only

v2.5.0 - AI Assistant File Writing 🚀

NEW: AI Assistant can write files directly to disk
NEW: Natural progress messages ("Creating Cart.jsx... ✅ Done!")
NEW: Automatic context preservation across messages
NEW: Task tracking for multi-step operations
IMPROVED: Seamless workflow with Multi-Agent and Agentic modes

v2.4.0 - Enhanced Agentic Workflow 🤖

NEW: Detailed progress tracking with step-by-step indicators
NEW: File-by-file updates showing each file being written
NEW: Post-completion menu with helpful next actions
IMPROVED: Better transparency and control over workflow execution

v2.3.0 - Proactive Code Analysis ⚠️

NEW: Real-time code analysis as you type
NEW: Yellow squiggly lines for potential issues
NEW: Quick fixes with lightbulb suggestions
NEW: Smart detection of missing imports, error handling, deep nesting

v2.2.3 - Conversational AI 💬

NEW: Human-like conversational AI agents (Architect, Developer, QA)
NEW: Professional yet warm communication style
NEW: Natural language responses with "I" and "you"
NEW: Contextual acknowledgments before responses
NEW: Personality-driven agent interactions
IMPROVED: Removed robotic "SYSTEM" messages
IMPROVED: Natural approval requests and handoffs
IMPROVED: Better user engagement and understanding

v2.2.2 - Performance & Scalability ⚡

NEW: Smart file chunking for large files (70% memory reduction)
NEW: Persistent embedding cache (10x faster searches)
NEW: Progressive indexing with priority-based processing
NEW: Token budget protection for huge projects
NEW: Smart summarization for very large files (>500KB)
IMPROVED: Handles 10,000+ file projects efficiently
IMPROVED: Detailed indexing statistics and skip reporting

v2.2.0 - Code Generation Intelligence 🎨

NEW: Advanced code generation with pattern recognition
NEW: Intelligent import suggestions
NEW: Duplicate code detection
NEW: Refactoring suggestions
IMPROVED: Better code quality and consistency

v1.0.23 - Git Integration 🔀

NEW: AI-generated commit messages from code changes
NEW: Git status indicator in chat header
NEW: View diffs directly in chat with syntax highlighting
NEW: Quick Git actions (Status, Commit, Changes, History)
NEW: Real-time Git status monitoring
NEW: Conventional Commits format support
FEATURE: Smart commit workflow with AI analysis and approval

💎 Free vs Pro

Feature	Free Tier	Pro ($9.99/mo)
Chat Messages	50/day	♾️ Unlimited
Inline Completions	20/day	♾️ Unlimited
Terminal Commands	10/day	♾️ Unlimited
Workspace Search	♾️ Unlimited	♾️ Unlimited
Git Integration	♾️ Unlimited	♾️ Unlimited
Multi-Agent Workflows	❌ 7-day trial	✅ Unlimited
Agentic Workflows	❌ 7-day trial	✅ Unlimited
Folder Indexing	❌ 7-day trial	✅ Unlimited
Custom Coding Rules	❌ 7-day trial	✅ Unlimited
200K Context Engine	✅ Yes	✅ Yes
Conversational AI	✅ Yes	✅ Yes

🎉 7-Day Pro Trial - Try all Pro features free for 7 days!

🆚 HalCode vs Competitors

Feature	Enterprise Tools*	HalCode Pro	Winner
Inline Completion	✅	✅	🤝 TIE
Terminal Integration	✅	✅	🤝 TIE
Workspace Search	✅	✅	🤝 TIE
200K Context Engine	✅	✅	🤝 TIE
Multi-Agent System	✅	✅	🤝 TIE
Conversational AI	❌	✅	🏆 HALCODE
Pricing	$50-100/mo	$9.99/mo	🏆 HALCODE
Customization	Limited	Full Control	🏆 HALCODE
Free Tier	❌ No	✅ Yes	🏆 HALCODE

💰 Save 80-90% compared to enterprise AI coding tools!

*Compared to similar enterprise AI coding assistants

v1.0.22 - Multi-Agent Enhancements 🤖

NEW: Real-time agent status cards with visual feedback
NEW: Agent progress tracking with pulsing animations
NEW: Color-coded agent states (working, complete, error)
IMPROVED: Enhanced multi-agent collaboration UI/UX
IMPROVED: Better error handling and retry mechanisms

v1.0.21 - Performance Optimizations ⚡

IMPROVED: Deferred heavy initialization for faster startup
IMPROVED: LRU cache eviction for embeddings (100 entry limit)
IMPROVED: Chat history limited to last 50 messages on restore
IMPROVED: Expanded file indexing exclusions
IMPROVED: Better .vscodeignore configuration

v1.0.20 - UI/UX Improvements ✨

NEW: Smooth animations and transitions
NEW: Message timestamps
NEW: Toast notifications for user feedback
NEW: Progress indicators for long operations
NEW: Enhanced code blocks with copy buttons
IMPROVED: Better message styling and layout

v1.0.9 - Multi-Agent Collaboration 🤖👥

NEW: Three-agent system (Architect, Developer, QA)
NEW: Collaborative code generation workflow
NEW: User approval before implementation
NEW: Automatic file creation

Earlier Versions

v1.0.8: Chat history, search, and export features
v1.0.7: Conversation management
v1.0.6: Deep context awareness
v1.0.5: Style guide integration
v1.0.0: Initial release

Built with ❤️ for developers who want AI that truly understands their code

HalCode AI Assistant

James Arce

HalCode AI Assistant

⚡ Quick Start (v3.0.17+)

🌟 Features

🚀 Ollama Support - Three Great Options! (v3.0.16+)

1️⃣ Local Ollama (Completely Free!)

2️⃣ Ollama Cloud (Cheapest for Heavy Use!)

3️⃣ Claude & Gemini (Best Quality)

💰 Complete Pricing Comparison

🎯 Which Option Should You Choose?

✅ Follow-Up File Modification Fix (v3.0.1 - AI Assistant Now Executes Follow-Up Changes! ✅)

✅ Task List System & AI Assistant File Extraction Fix (v3.0.0 - Real-Time Progress! ✅)

🎯 Token Budget Enforcement (v2.6.10 - Rate Limit Fix! 🎯)

🔄 Claude Embedding Fallback (v2.6.8 - Flexible & Resilient! 🔄)

⚡ Optimized Chat Stream Speed (v2.6.7 - Smoother Responses! ⚡)

✅ Organic AI Conversation & File Creation (v2.6.4 - Natural Flow! 🎯)

🧠 Improved Gemini Memory (v2.6.1 - Better Context! 💭)

🔄 Checkpoint System (v2.6.0 - Undo Changes! 🎯)

💬 AI Assistant File Writing (v2.5.0 - Game Changer! 🎯)

🤖 Enhanced Agentic Workflow (v2.4.0 - Production Ready! 🚀)

⚠️ Proactive Code Analysis (v2.3.0! 🚀)

💡 Inline Code Completion (v2.2.4! ⚡)

🔧 Enhanced Terminal Integration (NEW in v2.2.4! 🖥️)

🔍 Workspace-Wide Search (NEW in v2.2.4! 🔎)

🗣️ Conversational AI Agents (v2.2.3 💬)

🚀 Performance & Scalability (v2.2.2)

🎨 Code Generation Intelligence (v2.2.0)

🧠 Real-time Dependency Analysis (v2.0.0)

💬 Persistent Chat History & Conversations

🔍 Search & Export

🧠 Deep Context Awareness

🔀 Git Integration

🤖👥 Multi-Agent Collaboration (v1.0.22)

🤖 Agentic Workflows

💻 Interactive Chat Interface

💬 Conversational AI Experience

⚙️ Highly Customizable

🚀 Getting Started

1. Installation

2. Set Your API Keys

Add Claude API Key (Optional)

Add Gemini API Key (Optional - FREE TIER!)

Setup Ollama Cloud (Optional - FREE & FAST!)

Setup Local Ollama (Optional - COMPLETELY FREE!)

3. Configuration

Configure Your System Prompt (Optional)

Set Your API Key (Legacy)

Configure Your System Prompt (Optional)

4. Configure Inline Completion (Optional)

5. Index Your Project Folder

5. Usage

Open the Chat

Use Inline Completion (NEW! ⚡)

Run Terminal Commands (NEW! 🖥️)

Search Your Workspace (NEW! 🔎)

Analyze Dependencies (v2.0.0 🎉)

Create a New Conversation

Manage Your Conversations

Search Your Conversations

Export a Conversation

Switch AI Providers

Analyze Current File

Refactor with Plan

Use Git Integration (v1.0.23+)

🚀 Three-Workflow Approach (v2.5.0+)

1️⃣ Multi-Agent Workflow - Large Projects

2️⃣ Agentic Agent Workflow - Automation & Verification

3️⃣ AI Assistant Workflow - Details & Polish

🎯 Provider Support Matrix

🎯 Recommended Workflow

💡 Best Practices for AI Assistant

🎯 Recommended Workflow for Best Results

Natural Conversation Flow

Keep the AI on Track

Why This Matters

Pro Tips

📋 Commands

⚙️ Settings

🎯 Best Practices