LlmBench lets you use multiple AI services from one client without installing a separate app for every provider. Android is the first client; portable provider, model, profile, and prompt logic lives in the Kotlin Multiplatform shared core for future desktop and iOS reuse.
What LlmBench provides#
Persistent provider-scoped WebViews for account-backed services, with bounded memory handling, uploads, mobile tweaks, and privacy-safe activity diagnostics.
A native chat and comparison surface for direct Gemini, OpenAI, and Claude transports plus OpenRouter Free and AIHubMix gateway adapters.
Reusable profiles, instructions, prompt tooling, YAML editing, and a small skills/docs browser without vendoring a second copy of the portable .ai configuration core.
The Android prototype also integrates with the system share sheet for text, images, and application files.
Provider coverage#
The account-backed WebView layer currently includes ChatGPT, Claude, Gemini, DeepSeek, Kimi, Mistral Vibe, Qwen, Microsoft Copilot, Z.ai, Grok, Character.AI, Venice, and Meta AI. Some newer providers remain partial until embedded sign-in, upload flows, and activity probes are verified.
The native layer keeps provider-specific transport behind adapters. OpenRouter Free and AIHubMix are gateway options and stay outside the default All Models comparison to avoid duplicate aggregator routing.
Project status and license#
LlmBench remains an early prototype and is released under the ISC license. Android is the current focus; desktop and iOS clients are planned to reuse the shared KMP core where platform-native WebView, upload, authentication, and secure-storage behavior should remain native.
For the freshest implementation status, use the provider matrix and roadmap in these docs together with the repository’s default branch.