Connect a local OpenAI-compatible server, run installed models and measure real request performance. No hosted token charge; hardware and electricity still cost money.
Requests go directly from this browser to your local server. Cloud API keys are never attached.
http://127.0.0.1:11434/v1
Start the server yourself and enable CORS only for your INFERDEX origin. Do not expose your unauthenticated local server publicly. Browser support and local-network permissions vary.
Planning estimate for dense model weights plus your reserve, not a compatibility benchmark. Quantization metadata, architecture, context, batch size and runtime change memory use. Check the actual model license and runtime support.
Browse model details →Cancelling stops this browser request; the local server may continue computation.
Include electricity, hardware or rental cost. Estimates below cover active request time only, excluding idle time. Throughput includes prompt processing, not decode speed alone.
| Model | Request time | Output tokens | End-to-end tokens/s | Active compute cost |
|---|
No benchmarks yet. Run a local request to measure your setup.
Explore image generation and video workflows, or compare transcription options. These studios use their configured cloud providers; connecting a local text server does not move their jobs to your GPU.