OpenChat Connect vs AI Chat Client

OpenChat Connect is a free native iOS app with two ways to chat. Download a model and talk to it on your iPhone. Or bring your own API key and use OpenAI, Claude, Gemini, OpenRouter, or any custom endpoint. No OpenChat account. No lock-in. On this device Version 1.2.0 adds curated MLX downloads from Meta, Qwen, and Phi. Pick Fast, Balanced, or Strongest, then choose the On This Device provider. That chat runs on the iPhone. You do not need an API key for it. Your key, any model Paste a key and use OpenAI, Claude, Gemini, OpenRouter, or a custom OpenAI-compatible endpoint, including Ollama, LM Studio, or vLLM. The same key setup also covers Mistral, DeepSeek, cloud Qwen, Moonshot, Z.ai, and 01.AI. Pick the model that fits the task. Switch later. The app does not lock you to one provider. One app either way Chats, rules, memory, and skills are stored on the device. You can set global rules and per-chat rules, edit memory, and run a skill with / in the composer. Web search is optional and uses your own search key (Tavily, Exa, Brave, Serper, or SerpAPI). The app can render streaming Markdown and code, send a photo to a vision-capable model, and finish a cloud reply in the background with a local notification. Private by design OpenChat has no backend, no account, and no analytics. On-device model chats are not sent to a cloud model provider. When you choose a cloud model, that message goes from your iPhone straight to the provider you picked. OpenChat does not sit in the middle. API keys and search keys stay in the iOS Keychain. The app is free. Cloud usage is billed by the provider on your key, not by OpenChat. On-device chats do not use a provider key. App Store: https://apps.apple.com/us/app/openchat-connect/id6798106079 Source (MIT): https://github.com/robert-garay/openchat Support: robert@genion.ai

PowerChat is a privacy-first, bring-your-own-key AI chat app with model configuration options, local per-folder customizable RAG databases, and MCP support. It can be used to generate text and images using cloud models (Gemini, Deepseek, etc.) and local models (through Ollama), or any OpenAI compatible endpoint. It supports importing Silly Tavern v2 cards.