Are my API keys stored on a server?

No. API keys are encrypted in your browser (IndexedDB, AES-GCM 256 with a non-extractable WebCrypto key) and benchmarks run directly from your browser to the provider.

Which AI providers can I benchmark?

Any OpenAI-compatible chat-completions API: OpenAI, OpenRouter, Groq, Together, DeepSeek, Mistral, vLLM, Ollama, LM Studio and similar.

What metrics does LocalBench measure?

Time-to-first-token (TTFT), total latency, generation time, inter-token latency, output and overall tokens per second, prompt/completion token counts, error rates and estimated cost, with mean/median/p90/min/max/stddev aggregates.

LocalBench — AI Inference Benchmark in your Browser

Providers

🔒 Browser vault
Your API keys are saved and encrypted on this device — they're never sent to or stored on our servers. When you run a benchmark, your browser talks directly to the AI provider. Just note that some providers block this kind of direct browser access, though OpenAI, OpenRouter, Groq, and most others allow it.

Add provider

NameBase URL

API key (encrypted with AES-GCM in this browser)

$ / 1M input tokens (optional)

$ / 1M output tokens (optional)

Configured providers

No providers in the local vault yet.

Run benchmark 🔒 Runs in browser

Configure a provider in the browser vault first.

Provider

Model

Repetitions per prompt

Max tokens

Concurrency

Prompt sizes

shortmediumlong

Streaming (enables TTFT measurement)