Skip to content

beezyman-studio/OSLAH

Repository files navigation

🌌 OSLAH: Open-Source Local Agent Hub

OSLAH App Header License: MIT Privacy: 100% Local Open-Core: Active

OSLAH is a private, local-first AI automation platform for Windows and macOS. It empowers developers and teams to run, share, and orchestrate private LLMs locally on their own machines without transmitting any data to third-party cloud providers.


💎 Open-Core Monetization Model

OSLAH is built on an Open-Core distribution model. The core orchestration features are 100% free and open-source under the MIT license, while advanced multi-user routing, network administration, and enterprise security features require a commercial license.

Feature Comparison Matrix

Feature 🍃 Free Core Edition 🏆 Enterprise Pro Edition
Local LLM Chat Interface ✅ Yes ✅ Yes
Offline RAG Document Indexing ✅ Yes ✅ Yes
Ollama Autopilot Provisioning ✅ Yes ✅ Yes
Hardware Workload Telemetry ✅ Yes ✅ Yes
LAN HTTP Server Gateway ❌ No ✅ Yes
Cupertino Switch Control Center ❌ No ✅ Yes
Admin vs Employee Role Separation ❌ No ✅ Yes
Administrative Control Endpoints ❌ No ✅ Yes
SQLite User Log Telemetry ❌ No ✅ Yes

📊 Corporate Strategy & Roadmap

OSLAH is uniquely positioned to capture the offline enterprise AI market by eliminating dependency on cloud infrastructures. For a deep-dive look at our business model, go-to-market strategy, and investor pitch deck, please refer directly to: 📄 OSLAH Corporate Strategy Presentation

Key Business Strategy Pillars:

  1. Commercial Licensing & Support: Tailored subscriptions for local offices wishing to run local sandboxed AI gateways with priority technical SLA support.
  2. 100% Air-Gapped Compliance: Designed specifically for secure environments (such as healthcare, banking, and legal firms) that require strictly zero internet access and zero telemetry leakage.
  3. Hardware Optimization: Maximizes the utilization of existing office workstation graphics cards (GPUs) to run lightweight local models (DeepSeek-R1, Llama 3) without ongoing token fees.

⚙️ Quick Installation & Setup

OSLAH is distributed as a native executable. We use Inno Setup to package and build clean installers for Windows desktop.

💻 Installing via Pre-built Installer (Windows)

  1. Navigate to the InnoOutput/ directory.
  2. Launch OSLAH_Setup.exe.
  3. Follow the installation wizard steps to create desktop shortcuts and run the application.

🛠️ Building from Source (Windows Developer Guide)

Ensure you have Flutter SDK and Inno Setup installed.

  1. Clone the Repository:
    git clone https://github.com/beezyman-studio/OSLAH.git
    cd oslah
  2. Install Dart Dependencies:
    flutter pub get
  3. Compile Release Binary:
    flutter build windows --release
  4. Compile Setup Installer: Open a terminal and run the Inno Setup compiler compiler:
    iscc oslah_installer.iss
    The output installer will be generated at InnoOutput/OSLAH_Setup.exe.

🧠 Local Brain Selection Matrix

When setting up OSLAH, you can choose and download the local LLM model weights matching your hardware profiles:

Model Option Parameters Core Strengths (PROS) Drawbacks (CONS) Malayalam Support CPU / RAM Profile
DeepSeek-R1 7B 🧠 Logical step-by-step thinking (<think> tags). Excellent math & coding capabilities. ⏳ Slower initial token stream due to reasoning cycles. Excellent 🌟 (Superb translations & comprehension) Moderate-High (8GB+ RAM / VRAM)
Llama 3 8B 💬 Highly conversational, fast streams. Extremely robust general knowledge base. 🌐 Lacks native reasoning outputs. Basic ⚠️ (English is highly preferred) Moderate-High (8GB+ RAM / VRAM)
Phi-3 3.8B ⚡ Ultra-lightweight & fast. Runs efficiently on basic laptop CPU configurations. 📉 Limited reasoning scope. Smaller context. Poor ❌ (Basic vocabulary matching only) Low-Lightweight (4GB+ RAM / VRAM)

🔒 Security & Sandboxing

OSLAH is designed with local enterprise security compliance in mind:

  • Incoming requests outside the host require a valid API header token: X-OSLAH-Key or Authorization: Bearer <API_KEY>.
  • Access rules are verified by the internal role-based middleware (Admin vs Employee keys).
  • Incoming payloads are parsed using a strict sequential queue to prevent local memory overflow or CPU exhaustion DOS attacks.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details. Custom Enterprise modules are subject to commercial licensing terms.


BeezyMan Studio Kerala

About

An open-source, local-first AI Agent Hub built with Flutter and Ollama for secure, private team collaboration and RAG workflows.

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors