EerkAI is a specialized interface that forces Google's Gemini Nano LLM to run locally on low-spec hardware (verified on 1.9GB VRAM) and in restricted regions. It provides a distraction-free, modern chat interface that runs 100% offline.
- 🧠 Persistent Context: The AI remembers previous turns in the conversation (no more "amnesia").
- 100% Offline: Runs entirely in the browser via the
LanguageModelAPI. - Auto-Healing Core: Automatically detects frozen sessions and reboots the AI brain in milliseconds.
- Stealth Mode (NUKE): A dedicated Trash button instantly wipes the session, clears the screen, and flushes console logs.
- Smart Streaming: Smooth text generation that handles data chunks intelligently.
- Download the
index.htmlfile from this repository. - Configure your browser (See SETUP_GUIDE.md for the bypass instructions).
- Drag and drop
index.htmlinto Chrome. - Start typing.
This tool requires specific browser flags to bypass hardware checks. 👉 READ THE SETUP GUIDE HERE
Open Source. Dedicated to the pursuit of knowledge.
- 100% Offline: Runs entirely in the browser with zero server calls.
- Auto-Healing Engine: Automatically detects frozen sessions and reboots the AI core in milliseconds without refreshing the page.
- Stealth Mode (NUKE): One-click memory wipe that clears the screen, resets the model, and flushes console logs instantly.
- Smart Streaming: Handles data chunks intelligently to prevent text flickering/jumping.
- Markdown Rendering: Automatically formats code blocks with syntax highlighting.
EerkAI is a Zero-Dependency tool. You do not need Node.js, Python, or a local server.
- Download the
index.htmlfile from this repository. - Drag and drop it into your Chrome Dev browser.
- Start typing.
Important: EerkAI requires a Chromium browser with the Gemini Nano model enabled.
- Browser: Chrome Dev / Canary (v133+)
- Flags Required:
PromptAPIForGeminiNanoOptimizationGuideOnDeviceModel
(If your browser is not set up yet, see SETUP_GUIDE.md in this repo).
| Command | Action |
|---|---|
| RUN | Executes the prompt and streams the response. |
| WIPE | Instantly clears chat history and destroys the current session (Stealth Mode). |
| Enter | Submit command. |
| Shift+Enter | New line. |
Open Source. Dedicated to the pursuit of knowledge.
Project Name: Encrypted Engine Research Kit - AI (EerkAI)
Version: 3.0 (Stable)
Status: Operational
License: Open Source (MIT)
The purpose of EerkAI is to provide a lightweight, zero-dependency interface for interacting with on-device Large Language Models (LLMs), specifically Google's Gemini Nano. It is engineered to overcome artificial hardware constraints (VRAM gating) and regional restrictions, enabling offline AI inference on standard consumer hardware.
The software functions as a client-side web application running within the Chromium environment. It bridges the gap between the user and the browser's internal LanguageModel API.
Key Capabilities:
- Offline text generation (Air-gapped).
- Bypassing GPU VRAM requirements (<3GB).
- Persistent session management (Context Awareness).
- Real-time Markdown rendering.
- LLM: Large Language Model.
- VRAM: Video Random Access Memory.
- Air-Gapped: Isolated from unsecured networks (Internet).
- API: Application Programming Interface.
EerkAI operates as a standalone HTML5 application. It does not require a backend server (Node.js, Python) or cloud dependencies. It utilizes the experimental AI capabilities embedded in Chrome Dev/Canary builds (v133+).
- Target Audience: Researchers, Students, and Developers.
- Technical Proficiency: Intermediate (Requires ability to modify browser launch flags).
- Minimum Requirement: Integrated GPU with shared memory.
- Verified Testbed: Intel UHD Graphics / 1.9 GB VRAM.
- Storage: 2GB free space (for model weights).
- FR-01: The system shall initialize a connection to the
window.ai.languageModelorLanguageModelclass upon launch. - FR-02: The system shall auto-detect API availability and visually indicate status (Green/Red).
- FR-03: The system shall support "Auto-Healing"; if a session hangs, it must terminate and re-initialize the instance automatically.
- FR-04: The interface shall feature a floating input dock containing the text area and action buttons.
- FR-05: The "Send" button must dynamically transform into a "Stop" button during active generation.
- FR-06: The "Theme Toggle" shall switch between Dark Mode and Light Mode and persist the user's choice via LocalStorage.
- FR-07 (Stealth Mode): The "NUKE" function must perform a hard reset: clearing the DOM, destroying the session object, and flushing console logs.
- FR-08: No data shall be transmitted over the network. All processing must remain local.
The software requires a specific environment configuration to function.
Browser: Google Chrome Dev or Canary (v133+)
To enable the underlying model on restricted hardware, the browser must be launched with specific command-line arguments.
Steps:
- Right-click the Chrome Dev shortcut.
- Select Properties.
- Append the following flags to the Target field:
--enable-features=PromptAPIForGeminiNano,OptimizationGuideOnDeviceModel:compatible_on_device_performance_classes/*,OptimizationGuideOnDeviceModelOverride
- Download
index.htmlfrom the repository release. - Launch the modified Chrome Dev browser.
- Drag and drop the file into the browser window.
The main view consists of a vertically scrolling chat history.
- User Prompts: Aligned right, styled with accent colors.
- AI Responses: Aligned left, rendering Markdown (Code blocks, Bold, Italics).
Output detected as code (wrapped in triple backticks) is rendered in a dedicated container:
- Header: Displays language type (e.g., Python, JS).
- Body: Monospace font with syntax coloring.
- Action: "Copy to Clipboard" button with visual feedback.
- Frontend: HTML5, CSS3 (CSS Variables), Vanilla JavaScript (ES6+).
- Backend: N/A (Client-Side only).
- Model: Gemini Nano (Quantized 1.5B).
- Cold Start: Initial prompt may take 5-10 seconds to load into VRAM on low-end devices.
- Context Window: Limited by the browser's allocation (approx. 4k tokens).
Developed by [Your Name] Dedicated to the pursuit of knowledge.