Read code more comfortably, personalize your chats and explore new local-model features with improvements for iOS 27. NEW OPTIONS AND CONTROLS • Higher message contrast: darken your message backgrounds in dark themes for easier reading. • Show provider logos: choose whether provider logos appear in your chats. • Code blocks: syntax highlighting, an expanded reader, full or partial copying, and a toggle to switch between wrapped lines and horizontal scrolling. Code text also matches your chat typography more closely. • OpenCode: connect with your own API key, choose Zen or Go, select a model and choose which model to use when verifying your connection. • Siri and Spotlight: optionally search saved chats and artifacts from the system. This is off by default; temporary and protected content is excluded, and indexing pauses when app lock is enabled. LOCAL MODELS AND iOS 27 • Adaptive layouts and improved keyboard handling preserve your draft and focus when resizing windows. • Image understanding with Apple's local model on iOS 27 when the device and model support it. • Improved conversation context handling for Apple's local model, with clearer context-limit and attachment errors. • Support for continued local generation when the system allows it, and longer-running local-model shortcuts on iOS 27. • Support for companion-device authentication where available. FIXES • Fixed typing interruptions when tapping inside the message composer and removed excess space below messages. • Fixed provider reasoning preferences being reset at launch. • Corrected the privacy consent flow for custom providers, with clearer information about the destination server. • Improved local-model download verification and reporting when download state cannot be saved. • Addressed image-processing failures caused by image token counting in Apple's local model. • Improved voice-session handling during audio interruptions and output changes, preserving the mute state.

LLMConnect 4.0 is our biggest update yet. • Local models, rebuilt: refreshed catalog with Qwen3, DeepSeek and GLM-4, plus custom models straight from Hugging Face — all running privately on your device • Apps & MCP, upgraded: latest MCP protocol, hardened OAuth sign-in, live tool progress and interactive server forms • Skills library, rebuilt: safer installs and a durable catalog for extending your bots • New streaming engine: one unified pipeline for every provider, image streaming, Perplexity Agent API and automatic OpenRouter retries • Deep performance work across the whole app — chats render and scroll noticeably faster • Redesigned look: glass chat header, provider logos on responses, rebuilt onboarding and clearer per-provider privacy info • iCloud sync, rewritten: per-chat storage removes the old size cap, with conflict-aware merging • Dozens of security and reliability fixes from a full-app audit

General bug fixes and UI optimizations.

General bug fixes and UI optimizations.

General bug fixes and optimizations.

New in 3.1.3: • Smarter attachments: LLMConnect now chooses the best image, PDF, OCR, or plain-text format for each provider and model. • Better OCR and text extraction for PDFs, images, and unsupported attachment types. • DeepSeek now supports reasoning effort controls, with clearer handling for models that do not support image attachments. • Sidebar selection, chat streaming, and daily navigation feel smoother and refresh more reliably.

New in 3.1.2: • OpenAI voice features have been refreshed with newer Realtime models, Marin and Cedar voices, speaker-aware transcription, TTS style instructions, and clearer AI voice disclosure. • Perplexity chats are more reliable, with better streaming behavior, cleaner sources, updated Sonar model handling, and stronger support for longer responses. • OpenAI image generation now uses longer timeouts, making larger or slower image jobs less likely to fail too early. • Custom providers are more dependable, with improved setup, editing, model defaults, local/Ollama support, and safer credential handling. • Custom text sizing now applies more consistently across chat messages, Markdown, code, inputs, and key app screens. • Chat polish: user bubbles fit their content better, the keyboard dismisses when you tap outside the composer, and message actions are simpler. • Permission prompts now explain optional access more clearly during onboarding.

New in 3.1.1: • Image generation now works directly in chat, with OpenAI Image Generation v2, fal image models, and a shortcut to your generated image gallery. • Added GLM provider support, improved local Gemma stability, and prepared compatibility with GPT-5.5 models. • Model selection is now searchable, settings are cleaner, replies better follow your language, and streaming chats scroll more reliably.

LLMConnect 3.1.0 is a major update focused on control, automation, and reliability. New in this release: • Artifacts with live preview, fullscreen canvas, persistent gallery, and natural-language follow-up editing • Image generation with OpenAI, Gemini, and Qwen, new Cerebras support, and stronger provider/model controls • Connected Apps/MCP, Skills, automatic memory, recurring tasks, notifications, and better Apple Shortcuts support • Text size controls, advanced API settings, smarter model selection, folder prompts, auto chat titles, and broad stability, performance, and localization improvements

New features: • Real-time Voice Mode (requires an OpenAI API key). • Text-to-speech for provider responses (requires an OpenAI API key). • Speech-to-text dictation (requires an OpenAI API key). • Support for OpenAI GPT-5 models (5.x). • Added Qwen as a new provider. • Home Screen widgets and quick actions. • Apple Shortcuts support. Improvements: • UI rendering and performance improvements. • UI design polish. • Updated Markdown engine with LaTeX support. • Added missing translation keys. • Updated biometric authentication. Fixes: • Fixed saving very long system prompts. • Fixed memory management instability. • Fixed duplicate sending of very long messages to some providers. • Fixed some biometric auth failures.

New features: • Real-time Voice Mode (requires an OpenAI API key). • Text-to-speech for provider responses (requires an OpenAI API key). • Speech-to-text dictation (requires an OpenAI API key). • Support for OpenAI GPT-5 models (5.x). • Added Qwen as a new provider. • Home Screen widgets and quick actions. • Apple Shortcuts support. Improvements: • UI rendering and performance improvements. • UI design polish. • Updated Markdown engine with LaTeX support. • Added missing translation keys. • Updated biometric authentication. Fixes: • Fixed saving very long system prompts. • Fixed memory management instability. • Fixed duplicate sending of very long messages to some providers. • Fixed some biometric auth failures.

- Fixed a bug that prevented memories from being saved. - Fixed general bugs and errors. - UI and process optimizations.

- Added a new option in Settings to enable the Liquid Glass effect on the top and bottom chat bars. - Increased the sidebar width for a more comfortable visual experience. - Improved the interface for iPad users with better layout and spacing. - Optimized animation speed for a faster and smoother feel. - Added new UI elements to enhance the overall experience. - Improved overall app performance for better speed and responsiveness. - Fixed minor bugs and resolved known issues.

- Fixed a bug that prevented real telemetry from showing when downloading local models. - Fixed a bug that prevented local models from downloading correctly. - Fixed a bug that allowed incompatible MLX models to be imported. - Removed incompatible models from the default list. - General UI improvements and fixes.

- Fixed errors affecting custom provider setup. - Added an automatic importer for Replicate image models. - General UI improvements and fixes.

- Fixed a bug preventing new iPhones from downloading local models. - Fixed errors affecting custom provider setup. - Fixed issues saving image generation models. - Fixed a bug preventing generated images from being saved to the library. - Improved the streaming engine for local models (now 3x faster). - Added support for Ollama and OpenWebUI in custom providers. - Added the option to completely disable the system prompt. - Added a search bar in the default model selector.

LLMConnect is turning one year old, and we are celebrating with our biggest update yet! - Compatibility: Full support for iOS 26. - New plan: Free Plan. - Local Models (offline): Run models on-device, even without internet. - Cross-platform: New iPad and Mac apps included at no extra cost. - Security: Lock the app, specific chats, or folders with Face ID/Touch ID. - Chat branching: Create a branch from any message to explore ideas safely. - Sync & data: iCloud sync; JSON export/import. - Background tasks: Close the app while a model replies and get notified when it’s done. - Design: Liquid Glass aesthetic, Guided Tour, Black & White theme, new bubbles, typing indicator. - Under the hood: New streaming + Markdown processors, UI refinements, faster performance and improved stability.

- Redesigned chat screen with a simpler, optimized UI. - Fixed an issue preventing model/provider switching within an existing chat. - Folder section is now scrollable for better organization. - Fixed a bug where bots didn’t respect system prompts. - Updated copyright and attribution screen. - General performance improvements and UI tweaks.

- Fixed a bug that prevented GPT-5 models from keeping the full conversation context. - Removed redundant code that made the app more graphically demanding. - Replaced the Markdown rendering engine with a newer, optimized one for iOS. - Fixed an issue that caused some messages not to display completely. - Updated the message processing engine to improve speed and flow. - Resolved a bug that could freeze or crash the app when receiving very long messages quickly or when scrolling fast. - General improvements in stability, speed, and UI optimizations.

- New OLED dark mode - GPT-5 support with reasoning for OpenAI provider - New: Use your bots as a provider directly in the selector - Create chats from folders and organize your chats with drag & drop - Stronger API key security - New provider: Mistral - New: Disable response chat bubbles for a better UI experience - Sidebar no longer shows empty chats - Fixed an issue where the text in the input field displayed incorrectly - General bug fixes and UI refinements

LLMConnect v2.2.0 is our biggest update yet! • Full app rewrite to support new features and optimize core logic. • New AI providers – Google Gemini (with full streaming and file support), xAI Grok, and improved Replicate integration. • File & image support – OCR for scanned documents, multi-provider uploads, and faster file processing. • Fresh UI – Modern card-based redesign across key screens, new Bot Management interface, smarter message composer, and improved button alignment. • Full localizations – 7 languages added: Spanish, French, German, Italian, Portuguese, Hindi, and Simplified Chinese. • New dynamic conversation starters – Engage with smarter, more personalized opening messages.

LLMConnect v2.2.0 is our biggest update yet! • Full app rewrite to support new features and optimize core logic. • New AI providers – Google Gemini (with full streaming and file support), xAI Grok, and improved Replicate integration. • File & image support – OCR for scanned documents, multi-provider uploads, and faster file processing. • Fresh UI – Modern card-based redesign across key screens, new Bot Management interface, smarter message composer, and improved button alignment. • Full localizations – 7 languages added: Spanish, French, German, Italian, Portuguese, Hindi, and Simplified Chinese. • New dynamic conversation starters – Engage with smarter, more personalized opening messages.

We're squashing bugs, enhancing compatibility with the latest provider models, and polishing the app for smoother performance. Plus, we're laying the groundwork for our next big update!

- Compatibility with OpenAI o1 & o3 Models. - General Fixes and Performance Optimizations.

- Improved perplexity model fetching system - Fixes for OpenAI's new API model system - Overall app speed optimization - General bug fixes and stability improvements - Added compatibility with reasoning models from OpenRouter - Preparations for compatibility with upcoming OpenAI models - Preparations for compatibility with new providers - Fixed an issue where some settings were not being saved correctly - You can now enable timestamp sending to providers (this feature may not be compatible with all providers). - Removed the file attachment system due to conflicts with other features. This functionality will return in future updates.