LiberaGPT Offline Privacy AI

What’s new in v1.7.0: • iPhone 18 Pro Support: Full support for the iPhone 18 Pro and Pro Max, including our largest models (Mistral Small 24B and Bonsai 27B), the high-performance GPU path and the memory-efficient cache previously reserved for the iPhone 17 Pro. • New Models: Introducing Gemma 4 (E2B and E4B), LFM2.5 (2.6B and 8B), Ministral 3 (3B and 8B) and DeepSeek R1 0528 (Qwen3 8B). LiberaGPT now offers 22 on-device language models, plus image generation, vision and voice engines, 25 model packs in total, all running entirely on your phone. (Gemma 4 E2B requires iPhone 16 and newer; Gemma 4 E4B, Ministral 3 8B and DeepSeek R1 0528 require iPhone 16 Pro and newer; LFM2.5 8B requires iPhone 17 Pro and newer) • Bonsai 27B: Our largest model is now fully integrated with Sovereign Memory, attachments and your custom instructions, and supports the new Reasoning Mode. • Reasoning Mode: A new setting that lets supported models think step by step before answering, for stronger results on hard problems. (Available for SmolLM3, Qwen 3.5, Nemotron 3 and Bonsai 27B) • Attachments: Attach PDFs, text files and images to a message. LiberaGPT now fits documents to the loaded model’s context window and selects the sections most relevant to your question. • Sonelith: Gemma 4 E2B is the new default voice brain, with a more reliable model selection and enhancements to the voice agent. • Privacy Fix: Leftover diagnostics were writing prompt text to the device’s system log. This has been removed. Your prompts stay inside the app. • Smaller Backups: Downloaded models are no longer included in iCloud and computer backups, saving gigabytes of backup space. • Performance: Various performance improvements and ML library updates, including a larger context window for smaller models on iPhone 17 Pro and newer. • Chat: Scroll freely while an answer is streaming and jump back to the latest reply with one tap. Opening an older conversation now offers to switch to the model that wrote it.