ArloLite

ArloLite

Arlo Lite AI chat app

0 ratings
Free

Rating summary

Details

  • Released
  • Updated
  • August 6, 2026
  • August 16, 2026

Features

ArloLite screenshot #1 for iPhone
ArloLite screenshot #2 for iPhone
ArloLite screenshot #3 for iPhone
iphone
ipad
🖼️Get Icon
Icons↘︎

About

Arlo Lite is a free, open-source iOS client for interacting with large language models. Bring your own API key and talk directly to the provider — no middleman server, no account, no subscription. SUPPORTED PROVIDERS • OpenAI — Chat Completions & Responses API • Anthropic — Claude via Messages API • Local Inference — Ollama, llama.cpp, vLLM, or any OpenAI-compatible endpoint KEY FEATURES • Real-time streaming with token-by-token display • Rich Markdown rendering with syntax-highlighted code blocks • Session management — switch models mid-conversation, rename, resume • Thinking effort controls — adjust reasoning depth per session • Per-turn and per-session cost tracking • System prompt library — save and reuse your favorite prompts • iCloud backup for chat history (text only, keys never leave your device) PRIVACY BY DESIGN • API keys stored exclusively in the iOS Keychain • No backend server — all calls go direct to the provider • No telemetry, no analytics, no tracking • No account or sign-up required • Offline read access to past conversations BUILT FOR POWER USERS Arlo Lite is designed for developers and AI enthusiasts who want a clean, fast interface for daily-driving LLM APIs. Light/dark/system appearance, Dynamic Type support, and VoiceOver accessibility built in. 100% open source. MIT licensed. Contributions welcome on GitHub.
Show more

What's New in ArloLite

1.3

August 16, 2026

New: Full capability toggles (reasoning, vision, image gen, etc.) in the Add/Edit Model screen. Fixed: Cost always showing $0.000 — pricing was stored at 1/1,000,000th the correct scale. DB migration v10 fixes existing records automatically. Session total cost and token count never updating from zero. Cached token billing now uses the discounted cache-read price instead of the full input price. OpenAI Responses API streaming wasn't capturing cached token counts. Custom provider with thinking off wasn't sending enable_thinking: false, leaving some backends silently defaulting to thinking on. Error messages truncated with no way to see the full text — tap to expand now works even without a detail field. Sidebar delete button obscured by the chat layer during the open/close transition. Removed emoji from Vision and Reasoning badges in the model picker.

More