Our AI integration has evolved significantly in 2025, focusing on reliability, security, and seamless deployment. This updated guide covers the current AI models and providers, along with important changes from our previous offerings.

What's Changed in 2025

Deprecated Features

Ollama/LM Studio Integration Removed: Due to browser security policies that prevent HTTPS production sites from accessing localhost HTTP endpoints, we've removed support for Ollama and LM Studio integration. This ensures consistent functionality across all deployment environments.

Enhanced AI Offerings

AI-ML API Integration: Access to cutting-edge models including Claude 4.0, GPT-4.1 Nano, Gemini 2.5 Flash, and specialized image/video generation models.

Improved WebLLM Support: Enhanced browser-based AI processing with optimized models ranging from lightweight (Llama 3.2 1B ~0.9GB) to powerful (Hermes-3 8B ~4.9GB).

Current AI Integration Options

AI-ML API Models (Recommended)

Our primary provider offering state-of-the-art AI models through a unified API:

Text & Reasoning Models

• ⭐ GPT-4.1 Nano: Highly recommended - Efficient text generation and analysis ($ - Most cost-effective for daily tasks)
• Claude 4.0 Sonnet/Opus: Advanced reasoning and writing capabilities ($$ Sonnet / $$$$ Opus)
• o4-mini: Specialized reasoning model ($)
• Gemini 2.5 Flash Preview: Multimodal capabilities with vision support ($$)
• Qwen 2.5 Coder 32B: Specialized coding assistance ($$$)
• Llama 3.1 70B: Open-source high-performance model ($$$)

Multimodal Models

• Llama 3.2 90B/11B Vision: Advanced image understanding and analysis ($$ 11B / $$$ 90B)
• Claude 4.0 Models: Vision capabilities for document and image analysis ($$ Sonnet / $$$$ Opus)

Specialized Generation Models

• FLUX Schnell: Fast image generation ($)
• Stable Diffusion 3.5 Large: High-quality image creation ($$)
• Minimax Video (video-01): Professional video generation ($$$)
• Google Veo3: State-of-the-art video creation capabilities ($$$$ - Premium video model)

Setup: Simply add your AI-ML API key in settings. No additional configuration required.

WebLLM Models (Browser-Based)

Privacy-first AI processing that runs entirely in your browser:

Ultra-Lightweight Options (Perfect for Most Users)

• Llama 3.2 1B (~0.9GB): Latest instruction-following model with enhanced efficiency
• Gemma 2 2B (~1.9GB): Google's optimized model with improved memory usage

Recommended Models (Best Performance-to-Size Ratio)

• ⭐ Qwen3 4B (~3.2GB): Highly recommended - Excellent multilingual capabilities with strong reasoning
• ⭐ Phi-3.5 Mini (~3.7GB): Top pick for coding - Microsoft's specialized programming model
• Llama 3.2 3B (~2.3GB): Enhanced reasoning with better memory efficiency

High-Performance Options (Advanced Capabilities)

• Qwen3 8B (~5.7GB): High-performance multilingual model with coding expertise
• Hermes-3 Llama 3.1 8B (~4.9GB): Latest function-calling and instruction-following model

Setup: Select your desired model - it downloads automatically on first use and runs offline thereafter.

Hardware Requirements (2025 Update)

AI-ML API Models

• Internet: Stable connection required
• Hardware: No local constraints - processing happens in the cloud
• RAM: Standard browser requirements
• Performance: Instant responses with no local processing

WebLLM Browser Models

• RAM: Minimum 16GB (32GB recommended for larger models)
• Storage: Models cached locally (0.9GB - 5.7GB per model)
• Processing: Runs entirely on your CPU/GPU using WebAssembly technology
• Privacy: Complete offline processing after initial download

Internet Connectivity

Cloud Models (AI-ML API)

• Require stable internet connection for all operations
• Real-time processing with immediate responses
• No offline capabilities

Browser Models (WebLLM)

• Initial download requires internet connection
• Complete offline operation after model download
• Models remain cached for future use

Integration Steps

Setting Up AI-ML API

Step 1: Navigate to any AI-enabled tool (e.g., Notepad, Code Editors)

Step 2: Click the AI settings icon in the bottom left

Step 3: Select an AI-ML model from the dropdown menu

Step 4: Enter your AI-ML API key when prompted

Step 5: Start using AI assistance immediately

API Key: Obtain your API key from AI-ML API Dashboard

Setting Up WebLLM Models

Step 1: Enable GPU acceleration in your browser for optimal performance:

• Chrome/Edge: Visit chrome://flags/#enable-webgpu and enable WebGPU
• Firefox: Visit about:config and set dom.webgpu.enabled to true

Step 2: Open AI settings and select a WebLLM model

Step 3: Choose based on your system resources:

• 8-16GB RAM: Llama 3.2 1B or Gemma 2B
• 16-24GB RAM: Qwen3 4B (recommended), Llama 3.2 3B, or Phi-3.5 Mini (for coding)
• 32GB+ RAM: Any model including Qwen3 8B or Hermes-3 8B

Step 4: First selection triggers automatic download

Step 5: Wait for model initialization (one-time process)

Step 6: Enjoy offline AI assistance

Pro Tip: Download models on fast WiFi - they're cached permanently until manually cleared. For technical details, see the WebLLM model repository.

Migration from Deprecated Features

If You Were Using Ollama/LM Studio

Immediate Alternative: WebLLM models provide similar privacy-first, local processing without the deployment limitations.

Performance Alternative: AI-ML API models offer superior performance and capabilities compared to typical Ollama setups.

Migration Path:

1. Select a WebLLM model: Qwen3 4B for balanced performance, Phi-3.5 Mini for coding tasks, or Hermes-3 8B for advanced reasoning
2. Or upgrade to AI-ML API for enhanced capabilities
3. Your existing workflow remains the same - only the underlying AI changes

Advanced Features

Automatic Model Selection

The system intelligently routes requests to appropriate models based on:

• Text Generation: Default text models handle writing, editing, analysis
• Image Tasks: Automatically uses FLUX or Stable Diffusion models
• Code Tasks: Routes to Qwen Coder or other specialized coding models
• Multimodal: Uses vision-capable models for image analysis

Performance Optimizations

• Smart Caching: Browser models cached for instant subsequent use
• Efficient Prompting: Optimized prompts reduce token usage and improve responses
• Content Sanitization: AI responses automatically formatted for optimal tool integration

Troubleshooting

Common Issues & Solutions

WebLLM Model Won't Download: Ensure sufficient disk space and stable internet connection

WebLLM Models Running Slowly: Enable GPU acceleration in your browser:

• Chrome/Edge: Visit chrome://flags/#enable-webgpu and enable WebGPU
• Firefox: Visit about:config and set dom.webgpu.enabled to true

AI-ML API Errors: Verify API key is correctly entered in settings

Performance Issues: Consider switching to cloud models for resource-constrained systems

Browser Crashes: Reduce model size or increase system RAM allocation

Best Practices

Model Selection Strategy

• ⭐ Daily Tasks & Cost-Conscious Users: GPT-4.1 Nano ($) - Top choice for best value and performance in routine work
• Quick Tasks: Use WebLLM lightweight models (Llama 3.2 1B, Gemma 2B)
• Complex Analysis: Claude 4.0 Sonnet ($$) or Opus ($$$$) for advanced reasoning
• Creative Work: FLUX Schnell ($) for images, Veo3 ($$$$) for premium video generation
• Privacy-Sensitive: Stick to WebLLM for complete offline processing

Resource Management

• Download WebLLM models on WiFi to conserve mobile data
• Clear unused models if storage becomes limited
• Monitor system performance when running multiple browser models

Security & Privacy

AI-ML API

• Data processed on secure cloud infrastructure
• No data retention policies - check provider terms
• HTTPS encryption for all communications

WebLLM

• Complete local processing - data never leaves your device
• Models run in browser sandbox environment
• No internet communication after initial download

Conclusion

The 2025 AI integration focuses on reliability, performance, and universal compatibility. Whether you prioritize privacy with WebLLM or cutting-edge capabilities with AI-ML API, you now have access to more powerful and dependable AI assistance than ever before.

The removal of localhost-dependent integrations ensures consistent functionality across all environments while opening doors to more advanced AI capabilities that weren't possible with traditional local server approaches.

Ready to experience the next generation of AI integration? Start with our AI-ML API models for immediate access to state-of-the-art AI, or explore WebLLM for privacy-first local processing.