🖥️ Creating your own HomeLAB for trading

Lenovo wrote: Sat Mar 07, 2026 3:34 am I liked it, but I don't know what it's about.

Hi Lenovo,
If you haven't already done so, copy the HomeLAB thread link and paste it into an AI to begin a discussion. If appropriate, tell it of your limited tech skills and ask it to explain exactly what you should expect after you have downloaded it to your machine. If needed, ask it to walk you through the download process. You will most likely get a good layman's term explanation and begin to see the far reaching benefits something like this offers.

tried to make an Antigravity version of XARD's homeLab. looks like this
XARD: This is great -- You download the shared zip file
Antigravity creates the HomeLAB for you and then you decide what you want to run on it -- FANTASTIC
Once you have your HomeLAB the way that you want you can then say to Antigravity --
"Create a zipped share file (excluding .env etc) so I can share with the community"
And you just upload the file with an image of your HomeLAB and everyone is on the same page

# HomeLAB v3.5 — How It Works

## Overview
HomeLAB is a Python desktop application that orchestrates 6 AI models through a single Antigravity API, with real-time market data feeds and a shared memory system called the Neural Bridge.

---

## Architecture

```
┌─────────────────────────────────────────────────────┐
│ HomeLAB GUI │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │Gemini Hi │ │Gemini Lo │ │Gem Flash │ ← Buttons │
│ ├──────────┤ ├──────────┤ ├──────────┤ │
│ │Cl.Sonnet │ │Cl.Opus │ │ GPT-OSS │ │
│ └──────────┘ └──────────┘ └──────────┘ │
│ │
│ [ News] [ Whales] [ Macro] [ Intel] │
│ ← Scrolling Ticker Banners (clickable) │
│ │
│ [Input Box] → Type prompt here │
│ [Log Area] → Model responses appear here │
└──────────────────┬──────────────────────────────────┘
│
▼
┌──────────────────────────────────────┐
│ ORCHESTRATOR │
│ │
│ 1. Collects live prices (CoinGecko) │
│ 2. Reads Blackboard (shared brain) │
│ 3. Fetches news/whale/intel data │
│ 4. Builds system prompt with ALL │
│ context injected │
│ 5. Sends POST to Antigravity API │
│ 6. Writes response to Blackboard │
└──────────────────┬───────────────────┘
│
▼
┌──────────────────────────────────────┐
│ Antigravity API (local service) │
│ Routes to correct model by ID │
└──────────────────────────────────────┘
```

---

## File Structure

| File | Purpose |
|------|---------|
| `run_lab.py` | Entry point — launches the GUI |
| `src/config.py` | All settings: models, API key, endpoints, ticker config |
| `src/orchestrator.py` | Brain — API calls, live data injection, memory management |
| `src/gui.py` | User interface — buttons, tickers, input/output |
| `src/utils.py` | Utilities — price feeds, news, logging, backups |
| `src/scrapers.py` | Telegram channel scrapers for intel ticker |
| `.env` | Your `ANTIGRAVITY_API_KEY` (not committed to git) |
| `BLACKBOARD.md` | Neural Bridge shared memory (auto-managed) |

---

## What Happens When You Click a Model Button

1. **You type** a prompt in the input box
2. **You click** one of the 6 model buttons (e.g., Gemini High)
3. **Orchestrator collects context:**
- Your prompt
- Live prices (BTC $67K, Gold $5,172, DAX 23,591, etc.)
- Latest news headlines from RSS feeds
- Whale alerts from Telegram
- Fed/Macro press releases
- Sovereign intelligence signals
- Neural Bridge (BLACKBOARD.md) shared state
- Attached PDF content (if any)
4. **Sends everything** as a single API call to the Antigravity service
5. **Response** appears in the log area
6. **Neural Bridge updated** — other models can see this result

---

## Live Data Sources

### Prices (refreshed every 2 minutes)
| Asset | Source |
|-------|--------|
| BTC, ETH | CoinGecko API (free) |
| Gold (PAXG, XAUT) | CoinGecko API |
| Gold/Silver (Spot) | metals.dev API |
| DAX 40 | Yahoo Finance |
| Fear & Greed Index | alternative.me |

### Tickers (scrolling banners)
| Ticker | Source | Refresh |
|--------|--------|---------|
| News | CoinTelegraph, Kitco, CNBC, CME RSS | 30 min |
| Whales | Telegram @whale_alert | 5 min |
| Macro | White House + Federal Reserve | 60 min |
| Intel | 7 Telegram channels + AI sentiment | 5 min |

> **Clicking any headline** auto-sends it to Gemini Flash for deep analysis.

---

## Neural Bridge (Shared Brain)

The **BLACKBOARD.md** file acts as shared memory between all models:
- Every model **reads** the Blackboard before answering
- Every model **writes** its response summary to the Blackboard
- This means Claude can see what Gemini said, and GPT-OSS can see what Claude said
- Use ** Clear Blackboard** to archive and start fresh

---

## Voice Input (Optional)

- Uses **Whisper large-v3-turbo** for speech-to-text
- Click ** Hold to Talk** → speak → click again to stop
- Transcribed text goes into the input box
- Requires `sounddevice` and `faster-whisper` dependencies

---

## Quick Start

```bash
# 1. Set your API key
echo ANTIGRAVITY_API_KEY=your_key > .env

# 2. Install dependencies
pip install -r requirements.txt

# 3. Ensure Antigravity service is running

# 4. Launch HomeLAB
python run_lab.py
```

Pelle wrote: Sun Mar 08, 2026 9:16 am tried to make an Antigravity version of XARD's homeLab. looks like this

It looks great. Could you please share the files?

First of all, I would like to thank Xard for creating a separate topic for the HomeLAB and for regularly sharing such a wealth of information with the community.
I always find myself short of words to express my gratitude for what you have been doing for the community for so many decades.

I am completely new to Python and do not have any prior programming or coding experience.

Currently, I have paid subscriptions for Grok and Perplexity AI. I am planning to unsubscribe from my Grok subscription, which I normally use to help me understand MQL code when the source code is available. However, with Perplexity.ai I can now access and use several other models as well.

I downloaded the latest release shared by Xard (HomeLAB_v3.4_Gemini_3.1_Flash_Update.zip). I have installed Antigravity and Ollama with the 27B and 12B models.

At the moment I have not added any APIs, as I am not sure how to do that yet.

After installing and running the HomeLAB 3.4 batch file, this is what I see (image attached).

I am running this on my Dell Inspiron AIO. I have also attached an image of my PC specifications below in case it helps.

Processor: Intel(R) Core(TM) 7 150U (1.80 GHz)
Installed RAM: 32.0 GB (31.7 GB usable)
System type: 64-bit operating system, x64-based processor
Display driver: NVIDIA GeForce MX570 A

Windows Edition: Windows 11 Pro
Version: 25H2
OS Build: 26200.7840

Could someone please guide me on how to use it properly and what the correct workflow is?

Also, should I first configure the APIs (Gemini, Kimi, etc.) before trying to use the system, or can it already be used effectively with the local Ollama models only?

Xard, would it be possible in the future to add an option for Grok, or perhaps use Perplexity instead of Abacus.ai?

Also, are there any video tutorials or beginner-friendly resources that explain how everything works and how to use these models effectively?

For example, how are models like Grok and Perplexity different in terms of usage and purpose within this type of setup?

Any guidance would be greatly appreciated.

Stay blessed always.

XARD: If you already subscribe to Perplexity then just say to Antigravity to "Change Flash + Pro 3.1 + DeepSeek models to Grok + Perplexity and have Perplexity analyse news that is selected from NewsTicker" and place your Grok + Perplexity API_KEYs into the .env file and if you are unsure how to do that you can always get Antigravity to do it for you -- This way your HomeLAB would run Cloud (Grok + Perplexity) + Local (Gemma3) Nice

Thanks Xard will try to implement those changes today using Antigravity today.

Had to delete the post because the whole program wasn't working today

News ticker.............. yew

New day, new attempt.

# Creating your own HomeLAB: Antigravity AI Orchestrator (v3.5)

### Introduction
Welcome to **HomeLAB v3.5**, the **Antigravity AI Orchestrator**. This version is a complete overhaul, replacing all legacy multi-provider infrastructure (Abacus, DeepSeek, Kimi, local Ollama) with a clean, unified **Antigravity AI** architecture delivering **6 world-class models** through a single API key.

### Prerequisites
1. **Msty Studio**: Ensure you have Msty Studio installed and running.
2. **Vibe CLI Proxy**: Enable the Vibe CLI Proxy in Msty Studio on port 8317. **NO API KEY REQUIRED!** This local proxy unlocks all 6 models.

### New in v3.5: Unified Hexa-Architecture
- **6 Native AI Models**: Gemini 3.1 Pro (high/low), Gemini 3 Flash, Claude Sonnet 4.6, Claude Opus 4.6, GPT-OSS-120b.
- **Local Proxy Powered**: One local endpoint, all models. No more juggling Abacus, DeepSeek, or individual API keys. Just route through Msty!
- **No Local Hardware Required**: Ollama and local Gemma 3 dependencies have been completely removed. All inference runs in the cloud through the proxy.
- **Massive Code Generation**: All models tuned for **16,384 tokens**, allowing for 1,500+ lines of continuous code generation.

### Neural Bridge (Shared Brain)
- **Shared Blackboard Protocol**: All models read and write to a single source of truth (`BLACKBOARD.md`).
- **Zero Copy-Paste**: Context is automatically shared between models.
- **Project Lifecycle Management**: Use the ` Clear Blackboard` button to archive and reset.

### Installation Steps
1. **Create a Folder**: Create a folder on your computer (D: or C: drive) called `VAULT-AI`.
2. **Download & Unzip**: Download this HomeLAB Zip and unzip it into `VAULT-AI`.
3. **Launch Antigravity**: "Add Folder to Workspace" → Select `VAULT-AI`.
4. **Automated Setup**: Ask Antigravity: "Set up my HomeLAB".

### The Hexa-Architecture
HomeLAB v3.5 uses 6 Antigravity-native models:
- **Gemini 3.1 Pro (high)**: Deep architecture & complex reasoning at maximum quality.
- **Gemini 3.1 Pro (low)**: Pro-level reasoning with faster, economical inference.
- **Gemini 3 Flash**: High-speed planning, strategy, and general queries.
- **Claude Sonnet 4.6 (thinking)**: Advanced code review and synthesis with thinking traces.
- **Claude Opus 4.6 (thinking)**: Maximum intelligence for the hardest problems.
- **GPT-OSS-120b**: Open-source powerhouse for diverse general tasks.

### Smart Memory & Safety
- **Context Memory**: Automatically summarizes conversation history every 10 exchanges via Gemini Flash.
- **Visual Heartbeat**: Real-time "Neural Bridge: ACTIVE" status in the GUI header.

---
*Note: This is a completely free workstation utilizing your local Msty Studio Vibe CLI Proxy. No API keys are required. Enjoy the Hexa-Architecture!*

EDIT: Link to Msty Studio's Vibe CLI Proxy instructions: https://docs.msty.studio/features/vibe-cli-proxy

Pelle wrote: Mon Mar 09, 2026 7:03 pm New day, new attempt.

# Creating your own HomeLAB: Antigravity AI Orchestrator (v3.5)

### Introduction
Welcome to **HomeLAB v3.5**, the **Antigravity AI Orchestrator**. This version is a complete overhaul, replacing all legacy multi-provider infrastructure (Abacus, DeepSeek, Kimi, local Ollama) with a clean, unified **Antigravity AI** architecture delivering **6 world-class models** through a single API key.

### Prerequisites
1. **Msty Studio**: Ensure you have Msty Studio installed and running.
2. **Vibe CLI Proxy**: Enable the Vibe CLI Proxy in Msty Studio on port 8317. **NO API KEY REQUIRED!** This local proxy unlocks all 6 models.

### New in v3.5: Unified Hexa-Architecture
- **6 Native AI Models**: Gemini 3.1 Pro (high/low), Gemini 3 Flash, Claude Sonnet 4.6, Claude Opus 4.6, GPT-OSS-120b.
- **Local Proxy Powered**: One local endpoint, all models. No more juggling Abacus, DeepSeek, or individual API keys. Just route through Msty!
- **No Local Hardware Required**: Ollama and local Gemma 3 dependencies have been completely removed. All inference runs in the cloud through the proxy.
- **Massive Code Generation**: All models tuned for **16,384 tokens**, allowing for 1,500+ lines of continuous code generation.

### Neural Bridge (Shared Brain)
- **Shared Blackboard Protocol**: All models read and write to a single source of truth (`BLACKBOARD.md`).
- **Zero Copy-Paste**: Context is automatically shared between models.
- **Project Lifecycle Management**: Use the ` Clear Blackboard` button to archive and reset.

### Installation Steps
1. **Create a Folder**: Create a folder on your computer (D: or C: drive) called `VAULT-AI`.
2. **Download & Unzip**: Download this HomeLAB Zip and unzip it into `VAULT-AI`.
3. **Launch Antigravity**: "Add Folder to Workspace" → Select `VAULT-AI`.
4. **Automated Setup**: Ask Antigravity: "Set up my HomeLAB".

### The Hexa-Architecture
HomeLAB v3.5 uses 6 Antigravity-native models:
- **Gemini 3.1 Pro (high)**: Deep architecture & complex reasoning at maximum quality.
- **Gemini 3.1 Pro (low)**: Pro-level reasoning with faster, economical inference.
- **Gemini 3 Flash**: High-speed planning, strategy, and general queries.
- **Claude Sonnet 4.6 (thinking)**: Advanced code review and synthesis with thinking traces.
- **Claude Opus 4.6 (thinking)**: Maximum intelligence for the hardest problems.
- **GPT-OSS-120b**: Open-source powerhouse for diverse general tasks.

### Smart Memory & Safety
- **Context Memory**: Automatically summarizes conversation history every 10 exchanges via Gemini Flash.
- **Visual Heartbeat**: Real-time "Neural Bridge: ACTIVE" status in the GUI header.

---
*Note: This is a completely free workstation utilizing your local Msty Studio Vibe CLI Proxy. No API keys are required. Enjoy the Hexa-Architecture!*

EDIT: Link to Msty Studio's Vibe CLI Proxy instructions: https://docs.msty.studio/features/vibe-cli-proxy

Hi,
Please don't use my Home Lab version. It will result in a violation of Antigravity's terms of service, which came as a surprise to me.

My account was banned for a while

Pelle wrote: Thu Mar 12, 2026 5:52 am Hi,
Please don't use my Home Lab version. It will result in a violation of Antigravity's terms of service, which came as a surprise to me.

My account was banned for a while

Sorry to hear that Pelle picked up a temp ban from Google. The idea of using Msty Studio is fantastic—it truly levels the playing field for everyone. Whether you're on a budget laptop running a HomeLAB or looking to run larger models and offload some processing to your PC, it's a game-changer.

Last month I burned through 60k credits and just can't bring myself to subscribe to Antigravity. Their 1,000 credits per month with a reset every 5 hours doesn't cut it—and if you use up your credits, you get hit with a week-long ban. That's not exactly the best way to retain a user base.

Yesterday, I moved back to Linux Mint 22.3 and used Bottles to set up MT5 build 5675 (10 Mar 2026) with both GDI and Blend2D—now my MT5 is running sweet again with all my code. I also set up my HomeLAB_v3.6 (basically, v3.4 without the VS Code node, keeping everything simple), and it's running faster and smoother in the Linux ecosystem too.

For now I'm just trading my setup and waiting for DeepSeek v4 to land—no doubt it'll kick off a wave of releases from the other heavy hitters too. In the meantime, I'll be tinkering with Msty Studio and looking at ways to improve the newsticker feed and Blackboard.md analysis on my HomeLAB. Then I will drop a shared file for everyone to try out.
Best,
Xard777