Model Selection Guide
Optimize model speed, capability, and token cost for any workload.
Overview & Objectives
Our platform supports dynamic hot-swapping between diverse AI engines. Tailor your developer agent output by selecting the model that perfectly balances response speed, multi-modal context, and key billing rates. Choosing the right LLM configuration is the single most important factor in mastering local agent performance.
Core Tutorial Steps
Choose Your Provider & Backend Architecture
Toggle between Puter (Free, fast proxy), direct Google Gemini keys, OpenAI servers, or offline WebLLM running directly in your browser GPU. Each option has unique trade-offs.
Configure Your API Key & Local Storage Settings
Save keys locally to secure localStorage or browser session variables. To maintain maximum security, Accelerated Logic never sends your keys to an external database.
Verify Vision & Multi-Modal Compatibility
When processing complex screenshots, system UI mockups, or image-to-code components, verify that your active model supports full vision input (e.g., Gemini 3.5 Flash).
Understand Context Windows & Rate Limits
Analyze the context window of your selected model. Standard models support up to 1 million tokens, allowing you to ingest entire code repositories at once.
Analyze Performance Logs & Token Efficiency
Open the performance metrics panel to view real-time tokens-per-second, total generation time, and API cost estimation.
For rapid prototyping and casual testing, we recommend using the default Free Gemini 3.5 Flash proxy. If you plan to run complex multi-file compilation loops, connect a direct Gemini API key.