Under Cursor Settings -> Models:

1. Toggle off default model overrides.
2. Enable "OpenAI API Key".
3. Expand the Advanced options to input an alternative Base URL.

Step 2: Configuring Ollama or Local vLLM

To route completions directly to a local GPU instance running Ollama:

• Base URL: http://localhost:11434/v1
• Model Name: qwen2.5-coder:7b or deepseek-coder

Step 3: Mitigating Context Token Costs

When using external custom endpoints, monitor token usage on full-codebase indexing. Providing clean .cursorignore files prevents the indexer from ingesting large build artifacts, node_modules, or database dumps.

A comprehensive walkthrough detailing free provider endpoints compatible with Cursor’s configuration can be found here: Cursor Free Models Setup Guide

1. Google AI Studio (Gemini Flash & Pro)

• Key Advantage: Generous RPM (requests per minute) allowances on Flash models with expansive context windows.
• Best For: Document processing, structured JSON extraction, and low-latency agent routing.
• Setup: Generate an API key through Google AI Studio; standard OpenAI-compatible SDK wrappers can be used with custom base URLs.

2. Groq Cloud (LPU Inference)

• Key Advantage: Industry-leading token generation speeds on open-weights models (Llama 3, Mistral, Gemma).
• Best For: Real-time conversational UI, code completion, and voice-to-voice pipelines.
• Limitations: Strict TPM (tokens per minute) rate-limit ceilings on free organization tiers.

3. OpenRouter Free Endpoints

• Key Advantage: Single unified API interface to route requests dynamically across free open models (labeled :free).
• Fallback Handling: Built-in failover capabilities ensure high uptime even during regional capacity drops.

Live Comparison & Quota Tracker

For updated rate limits, context windows, and real-time status of free developer endpoints, check out the live comparison matrix: OpenGPURadar Free LLM Directory