Skip to content

AlexC1991/VoxAI_IDE

BranchesTags

Repository files navigation

VoxAI Coding Agent IDE

 ██╗   ██╗ ██████╗ ██╗  ██╗ █████╗ ██╗
 ██║   ██║██╔═══██╗╚██╗██╔╝██╔══██╗██║
 ██║   ██║██║   ██║ ╚███╔╝ ███████║██║
 ╚██╗ ██╔╝██║   ██║ ██╔██╗ ██╔══██║██║
  ╚████╔╝ ╚██████╔╝██╔╝ ██╗██║  ██║██║
   ╚═══╝   ╚═════╝ ╚═╝  ╚═╝╚═╝  ╚═╝╚═╝

Version Status Platform Build

VoxAI IDE is a local-first autonomous coding agent — built to be an alternative to Cursor and Claude Code. Purpose-built for developers who direct high-level intent and let a specialized AI handle implementation, debugging, and execution end-to-end.

Download the latest Windows release from the Releases page — no Python install required.

Why VoxAI

Standard AI editors suggest code; VoxAI builds software.

  • Local & Private — Full support for local LLMs (GGUF format). Run 100% offline with zero data leakage.
  • Curated Stable Core + Advanced Tools Mode — The default agent surface stays focused on local code/edit/test workflows, while git/web/RAG indexing/destructive tools sit behind explicit advanced settings.
  • Self-Healing Loop — When code fails, VoxAI captures stderr, analyzes the traceback, and patches itself in a continuous loop until the task is complete.
  • Terminal Mode — Minimize the GUI to tray and work from a Claude Code-style CLI with full tool access, streaming responses, and slash commands.
  • Deep RAG — A Go-based vector engine (HNSW + mmap) provides sub-millisecond semantic retrieval across your entire codebase and conversation history.
  • Thought Transparency — Every tool call, file read, and command is visualized inline. You see the AI's reasoning as it happens.
  • Token Optimization — Automatic truncation, history compression, and system prompt condensing. ~30-50% fewer tokens per request.

Features

Agent Tools (22)

Category Tools
File System read_file, write_file, edit_file, list_files, move_file, copy_file, delete_file, search_files, get_file_structure
Shell execute_command
Git git_status, git_diff, git_log, git_commit, git_push, git_pull, git_fetch (advanced mode)
Web web_search (DuckDuckGo via IronGate), fetch_url (advanced mode + web enabled)
Memory search_memory, search_codebase, index_codebase (advanced retrieval/indexing tools)

Chat & AI

  • Auto-Context — The open file, cursor position, and surrounding code are automatically attached to every message.
  • @-mention Context — Type @filename.py in the input to attach file contents.
  • Conversation History — Ctrl+H opens a sidebar listing all saved conversations. Click to switch, create, or delete sessions.
  • Attachments — Attach images (multimodal) or text files via the + button.
  • Copy / Regenerate — Every message has Copy; AI messages have Regenerate.
  • Apply / Reject — Proposed file changes show a diff preview and require approval (unless auto-approve or Siege Mode).
  • Collapsible Thinking — AI reasoning is shown in expandable panels, stripped from the visible response.
  • Tool Result Folding — Tool outputs appear as one-line collapsible summaries to keep the chat clean.
  • Token Footer — Per-message token breakdown: tokens: total (in: prompt · out: completion).
  • Context Window Bar — Color-coded fill in the status bar (green/yellow/red).
  • Model & Mode Selectors — Inline in the input bar, Cursor-style. Switch models or modes without leaving the chat.

Code Editor

  • Tabbed Editor — Syntax highlighting for Python, JS/TS, C/C++, Rust, Go, Java. Line numbers, current-line highlight.
  • Find & Replace — Ctrl+F with next/prev, replace, replace all, case sensitivity.
  • Bracket Matching — Real-time highlighting of matching (), {}, [] pairs.
  • Code Folding — Double-click the gutter to collapse/expand blocks.
  • File Watcher — Open tabs auto-reload when files change on disk.
  • Live Change Highlighting — When the AI writes a file, changed lines highlight green and removed lines highlight red using difflib.
  • Diff Viewer — Color-coded unified diffs in dedicated tabs.
  • Code Outline — Ctrl+Shift+L shows classes, functions, and methods for the active file (AST for Python, regex for others).

File Explorer

  • Git Status Indicators — Colored dots: modified (yellow), added (green), untracked (blue), deleted (red).
  • Git Diff — Right-click any changed file to view a full color-coded diff.
  • Context Menus — New File, New Folder, Rename, Delete, Copy Path, Copy Relative Path, Reveal in Explorer, Show Git Diff.
  • Filter Bar — Type to filter files by name in real-time.

Project Search

  • Project-Wide Search — Ctrl+Shift+F opens grep-style search across all project files with regex, case-sensitivity, and file-type filters.
  • Quick File Switcher — Ctrl+P opens a fuzzy-search overlay to jump to any file by name instantly.

Command & Control

Mode Behavior
Phased (Default) Executes one phase at a time. Reports results and waits for authorization.
Siege (Full Auto) Iterates up to 25 tool cycles without stopping. Self-heals on errors.
  • Command Palette — Ctrl+Shift+P opens a searchable action launcher with all IDE commands.

Terminal Mode

Press the Terminal icon in the top bar (or use the Command Palette) to:

  1. Minimize the GUI to the Windows system tray.
  2. Open a new console window with a Claude Code-style CLI.
  3. Stream AI responses with ANSI color, tool execution, and slash commands.
  4. Conversation history carries over from the GUI session.
  5. Double-click the tray icon or type /exit to return to the GUI.

Slash commands:

Command Description
/help Show all commands
/exit Quit terminal, return to GUI
/clear Clear conversation context
/mode Toggle Phased / Siege
/model Show current model
/model <name> Switch to a different model
/models List all enabled models
/tokens Show conversation token estimate
/status git status
/branch Show current branch
/branches List all branches
/log [n] git log (default 10)
/diff [file] git diff
/commit <msg> Stage all + commit
/push [remote] [branch] git push
/pull [remote] [branch] git pull
/fetch [remote] git fetch
/stash / /stash pop Stash management
/checkout <branch> Switch branch
/files [path] List project files
/search <query> Search across files
/run <command> Execute a shell command
/index Re-index project for RAG
/export <file> Export conversation to markdown

Desktop Notifications

Windows toast notifications fire when the app is not focused:

  • AI response complete
  • Approval needed for destructive actions
  • Phase gate reached

Settings (Tabbed)

Tab Contents
Providers & Models API keys for OpenAI, Anthropic, Google, OpenRouter, DeepSeek, Mistral, XAI. GGUF model manager with VRAM estimates.
Agent & RAG Max history tokens, auto-approve, auto-save, web search toggle. RAG enable/disable, top-k, similarity threshold.
Appearance Customizable user and AI chat colors.

Status Bar

  • Git branch name (auto-refreshes)
  • Cursor position (Ln / Col)
  • File encoding
  • Token usage progress bar (color-coded)

Architecture

VoxAI_IDE/
├── main.py                    # Entry point
├── core/                      # Backend
│   ├── ai_client.py           # Multi-provider AI client (streaming)
│   ├── agent_tools.py         # 22 tool implementations
│   ├── code_parser.py         # XML tool call parser
│   ├── prompts.py             # System prompts & mode injection
│   ├── rag_client.py          # Vector search client (Go backend)
│   ├── indexer.py             # Codebase chunker & indexer
│   ├── local_embeddings.py    # GGUF embedding engine (llama.cpp)
│   ├── settings.py            # QSettings-based persistence
│   ├── runner.py              # Script execution (cross-platform)
│   └── hardware.py            # GPU/CPU detection
├── ui/                        # PySide6 GUI
│   ├── main_window.py         # Main window, icon bar, menus, tray
│   ├── chat_panel.py          # Chat interface, AI/tool workers
│   ├── editor_panel.py        # Tabbed editor, find/replace, folding
│   ├── file_tree_panel.py     # Explorer with git status & context menus
│   ├── settings_dialog.py     # Settings UI (tabbed)
│   ├── model_manager.py       # GGUF model manager dialog
│   ├── search_panel.py        # Project-wide search
│   ├── file_switcher.py       # Quick file switcher (Ctrl+P)
│   ├── code_outline.py        # Symbol outline sidebar
│   ├── history_sidebar.py     # Conversation history browser
│   ├── syntax_highlighter.py  # Syntax highlighting engine
│   ├── debug_drawer.py        # Terminal output panel
│   ├── crash_reporter.py      # Crash dialog
│   └── widgets/
│       └── chat_items.py      # Message & progress widgets
├── cli/                       # Terminal Mode
│   └── terminal_mode.py       # Claude Code-style CLI
├── Vox_IronGate/              # Web client backend
│   ├── web_client.py          # DuckDuckGo search & URL fetch
│   └── lib/
│       ├── config.py          # Timeouts, user agents
│       └── security.py        # Rate limiting, URL safety, IP blocking
├── Vox_RIG/                   # RAG vector engine (Go)
│   ├── search_engine/         # HNSW index, mmap storage, HTTP + CLI
│   │   └── main.go            # Unified server/CLI entry point
│   └── drivers/               # llama.cpp shared libraries
├── tests/                     # Test suite
├── models/
│   ├── llm/                   # Drop GGUF chat models here
│   └── embeddings/            # Drop GGUF embedding models here
├── keys/                      # API key configuration
└── resources/                 # UI assets (icons, backgrounds)

Agentic Loop

User Input
    │
    ▼
┌─────────────┐     ┌──────────────┐     ┌──────────────┐
│  RAG Context │────▶│  LLM Stream  │────▶│  Tool Parser │
│  (retrieve)  │     │  (generate)  │     │  (XML parse) │
└─────────────┘     └──────────────┘     └──────┬───────┘
                                                 │
                         ┌───────────────────────┘
                         ▼
                  ┌─────────────┐
                  │ Tool Worker │──▶ file I/O, git, shell, web
                  │ (execute)   │
                  └──────┬──────┘
                         │
                         ▼
                  [TOOL_RESULT] ──▶ fed back to LLM ──▶ loop

Threading Model

Thread Responsibility
Main (UI) PySide6 event loop, widget rendering, user input
AIWorker Streams tokens from the AI provider
ToolWorker Executes tool calls (file I/O, git, web, shell)
IndexingWorker Background codebase indexing into RAG
RAG Server Go subprocess — HTTP vector engine on 127.0.0.1

Getting Started

Option 1 — Download Release (Recommended)

  1. Go to the Releases page.
  2. Download the latest VoxAI_IDE-vX.X.X-windows.zip.
  3. Extract and run VoxAI_IDE.exe.
  4. (Optional) Drop .gguf models into the models/llm/ folder for offline use.

Option 2 — Run from Source

Prerequisites: Python 3.10+, Git, (Optional) Go 1.22+ for RAG engine, NVIDIA GPU for local GGUF inference.

git clone https://github.com/AlexC1991/VoxAI_IDE.git
cd VoxAI_IDE
pip install -r requirements.txt
python main.py

Or on Windows:

.\start_IDE.bat

Build the RAG engine (optional, improves performance):

cd Vox_RIG/search_engine
go build -o vox-vector-engine.exe .

Configure Models

Local Models (Private & Offline)

  1. Download a .gguf model (Llama 3, Mistral, Gemma, Qwen, etc.).
  2. Place it in models/llm/.
  3. For embeddings, place nomic-embed-text-v1.5.Q8_0.gguf (or similar) in models/embeddings/.
  4. Select models from the dropdown in Settings.

Cloud Providers

Open Settings (Ctrl+,) and enter API keys for any of: OpenAI, Anthropic, Google, OpenRouter, DeepSeek, Mistral, XAI.

Benchmarks

VoxAI includes maintained live benchmark harnesses for controller regressions and model comparisons.

  • Tiny starter benchmark: python scripts/basic_project_start_benchmark.py "[OpenAI] gpt-5.4" --repetitions 3
  • Live Siege benchmark matrix: python scripts/run_live_siege_benchmark.py --model "[OpenAI] gpt-5.4"
  • Generated outputs now default to artifacts/benchmarks/
  • Headless benchmark runs use Qt offscreen mode automatically

Use explicit --output, --output-json, or --output-markdown if you want to override the default artifact paths.

Keyboard Shortcuts

Shortcut Action
Ctrl+Shift+P Command Palette
Ctrl+P Quick File Switcher
Ctrl+Shift+F Search in Project
Ctrl+Shift+L Code Outline
Ctrl+H Conversation History
Ctrl+B Toggle File Tree
Ctrl+Shift+E Toggle Editor
Ctrl+F Find & Replace
Ctrl+S Save file
Ctrl+O Open file
Ctrl+Shift+O Open project
Ctrl+, Settings
Ctrl+` Toggle debug panel
Enter Send message
Shift+Enter New line in chat
Escape Stop AI generation
Ctrl+L Clear chat context

Command & Control Modes

Phased Mode (Default)

The AI acts as a senior architect:

  1. Draft — Analyzes the request, presents a phased execution plan.
  2. Execute — Performs one phase using tools.
  3. Report — Summarizes results in detail and stops.
  4. Authorize — You review and send a message to continue.

Siege Mode (Full Auto)

For when you trust the machine:

  • The AI bypasses all phase gates.
  • Iterates up to 25 tool cycles without stopping.
  • Self-heals: captures errors, patches code, retries.
  • Write access is strictly confined to the active project directory.

Security

  • Write sandbox — AI can only write within the active project directory.
  • Read access — Global filesystem read for cross-project context.
  • Web safety — IronGate blocks requests to localhost, private IPs (RFC 1918), link-local, and cloud metadata endpoints.
  • Rate limiting — Token-bucket limiter on outbound HTTP requests (15/min).
  • Approval workflow — Destructive actions (delete_file, execute_command, git_commit, git_push, git_pull, move_file) require explicit user confirmation unless in Siege Mode or auto-approve is enabled.
  • RAG isolation — The vector engine runs on 127.0.0.1 only. No external network access.

Building from Source

To create a standalone Windows executable:

pip install pyinstaller
pyinstaller --onedir --windowed --name "VoxAI_IDE" \
  --add-data "resources;resources" \
  --add-data "cli;cli" \
  --add-data "core;core" \
  --add-data "ui;ui" \
  --add-data "Vox_IronGate;Vox_IronGate" \
  --hidden-import PySide6 \
  --hidden-import llama_cpp \
  main.py

Releases are also built automatically via GitHub Actions when a new tag is published.

License

See LICENSE for details.

VoxAI — The Command Center for Autonomous Development

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases 3

VoxAI_IDE_v2.2.0 Latest
Mar 11, 2026
+ 2 releases

Packages

 
 
 

Contributors

Languages