Ael-Coder for VS Code 🚀

Multi-Million-Token Local Code Intelligence powered by ISOM-R2

Ael-Coder brings deep, multi-million-token codebase intelligence directly into your VS Code editor. Powered by the ISOM-R2 engine and continuous $\mathbb{SO}(16)$ Lie-manifold Cayley transport, it breaks the quadratic KV-cache memory barrier—allowing you to query, refactor, and audit 1.7M+ tokens of code locally on consumer GPUs (e.g., 4 GB RTX 3050).

Zero cloud subscriptions. Zero telemetry. 100% private code generation.

⚡ Features

• 💬 Native Chat Sidebar: Ask complex multi-hop questions across your entire repository with instant streaming responses.
• 🛠️ Right-Click Code Actions:
  • Explain Code: Discovers architectural roles and edge cases across dependencies.
  • Refactor with Manifold Context: Rewrites functions for cleaner performance and modern idioms.
  • Generate Unit Tests: Produces 100% executable test suites for active functions.
• 🔍 Workspace Architectural Audit: Evaluates module boundaries, dataflow loops, and hidden bottlenecks in 1 click.
• 📊 Live Telemetry in Status Bar: Real-time connection and VRAM monitoring right in your status bar (⚡ Ael-Coder (1.7M)).

🚀 60-Second Quickstart

Step 1: Install the Local Engine (Terminal)

pip install git+https://github.com/prannesshkva/ael-coder.git

Step 2: Start the Engine on Your Machine

ael-coder serve --port 11435

(Or point to a specific folder: ael-coder serve --port 11435 --repo "C:\path\to\your\project")

Step 3: Open VS Code & Start Coding

1. Click the Ael-Coder icon in the Activity Bar on the left.
2. The status indicator will switch to Connected (1.7M).
3. Start chatting or highlight any code in the editor and right-click to refactor!

⚙️ Extension Settings

This extension contributes the following settings (settings.json):

• aelCoder.serverUrl: Base URL of the local bridge server (default: http://localhost:11435).
• aelCoder.modelName: Target model identifier (default: prannesshkva/ael-coder-1.5b).
• aelCoder.temperature: Sampling temperature (default: 0.2 for deterministic code).
• aelCoder.maxTokens: Maximum generated tokens per response (default: 2048).

🔬 Mathematical Architecture (ISOM-R2)

Ael-Coder bypasses standard 54–98 GB KV-cache bloat via continuous orthogonal transport along the $\mathbb{SO}(16)$ Lie group manifold: $$W = (I - A)(I + A)^{-1}, \quad \text{where } A = -A^T$$ $$|W^T W - I|_F < 10^{-5}$$

This ensures exact norm preservation across millions of tokens without representation decay.

📜 Links & Community

• GitHub Repository: https://github.com/prannesshkva/ael-coder
• Hugging Face Model: https://huggingface.co/Prannesshkva/Ael-Coder-1.5B
• Author: Prannessh K. V. A. (@Prannesshkva)

License: Apache-2.0. Copyright (c) 2026 Prannessh K. V. A. All Rights Reserved.