Registry overhaul, batched file reads, and MCP client support.

What's Changed

MCP Client Integration

• Aura can now consume tools from external MCP (Model Context Protocol)
  servers via connect_mcp_server() in the tool registry.
• New aura/mcp_client.py module with a sync wrapper around the MCP Python
  SDK, running the async protocol over a background event loop thread.
• External MCP tools are dynamically registered in the dispatch table and
  appear alongside built-in tools in the Worker's tool set.
• The Planner and Worker can now call remote tools for deployment, database
  queries, documentation lookups, and more — without leaving the IDE.

Tool Registry Refactor

• Replaced the dispatch chain in ToolRegistry.execute()
  with a module-level TOOL_HANDLERS dispatch table (22 tools).
• Each tool has an isolated handler method; adding new tools is a one-line
  registration.

Batched File Reads (read_files)

• New read_files tool reads multiple workspace files in a single API
  round-trip, reducing latency on multi-file operations.
• Per-file error reporting; 500KB total cap.

Testing

• 85 tool registry tests + 11 MCP client tests = 96 tests passing.
• Dependency: mcp>=1.0 added to pyproject.toml.

Installation

Windows Executable

Download Aura-1.1.0-windows-x64.exe from the Assets below.

From Source

pip install --upgrade aura

1. Persistent "Approve All" Mode: Fixed a bug where the Approve toggle failed to bypass diff dialogs after conversation reset. The mode is now a truly persistent session-wide state that correctly silences file modification prompts when enabled.
2. Robust Toolbar Interaction: Refactored window-dragging logic to ensure the Approve and Dispatch toggles respond instantly to clicks on any part of the widget (thumb, track, or label).
3. Diff Card Crash: Resolved a NameError in the Diff Approval dialog that occurred when attempting to display
   worker-generated file changes.

Installation

Installation instructions are unchanged from v1.0.0. See the v1.0.0 release notes for details.

Windows Executable

Download Aura-1.0.1-windows-x64.exe from the Assets below.

From Source

pip install --upgrade aura

Release v1.0.0 – Aura (Initial Launch)

Aura is a desktop AI Orchestration IDE for engineers who want agentic power without sacrificing privacy or workspace security. It bridges high‑level reasoning models with your local filesystem using a Planner‑Worker architecture and local‑first infrastructure.

Key Features

• Planner‑Worker Orchestration – A dual‑agent system turns high‑level goals into concrete, multi‑step execution plans.
• Surgical Python Edits – AST‑based refactoring at the symbol level – perfect indentation, syntax integrity preserved.
• Isolated Sandbox – AI‑generated shell commands and dynamic tools run inside a Docker container, fully isolated from your host OS.
• Zero‑Cloud Vision – Leverage local Ollama (e.g., llama3.2-vision) to describe images locally; only text context is ever sent to the cloud.
• Hardware‑Tethered Security – API keys encrypted via Fernet using a key derived from your machine (MAC + Username).
• Smart Codebase Search – Built‑in BM25 indexing for fast, semantic‑style retrieval across your entire workspace.

Installation

Windows (.exe)

1. Download Aura_IDE from the Assets section below.
2. Run the executable.

   SmartScreen notice: As a new release, Windows may show a warning. Click More Info → Run Anyway to proceed.

3. Configure your API keys in Settings (supports DeepSeek, OpenAI, Anthropic, Google).

Developer Install (from source)

git clone https://github.com/yourusername/aura.git
cd aura
pip install -r requirements.txt
python -m aura