Pin the chats you use every day to the top of the list. Speak your messages with dictation, and have replies read back to you. Claude models can now think before answering, with the thinking shown separately from the reply. Start a chat or send a message from Shortcuts, Siri, or the Action button. Download any public Hugging Face model, not just the curated catalog. Paste a model ID or URL and pick the file you want. Also in this release: - Connections show their provider's logo, and every list in the app now shares one design - On iPad, Return sends and Shift+Return starts a new line - The "local model not loaded" notice can be dismissed - A connection missing its API key now says so - Timestamps no longer tick once a second Fixes: - Turning off memory for a chat now sticks - Deleting a chat's messages also clears what was remembered from them - Switching to a model that cannot reason no longer strands the setting - Sending works again as soon as you are back online - Quitting mid-save no longer loses a message

3.1.0 makes local AI chats more resilient and gives model browsing one consistent, cleaner look. • Generations keep streaming in the background on iOS 26 and later — leave the app mid-reply and come back minutes later to a finished answer, with a system indicator showing how long it has been running • Redesigned LM Studio model picker: your current model and its reasoning setting up top, alternatives visible at a glance, and search, filters, and refresh in the toolbar • One consistent design across the model picker, LM Studio management, and Local Models, with provider logos that ship in the app and work offline • Download Hugging Face models on your LM studio server from 3sparks Chat by pasting either a full repository URL or a copied model ID • Reasoning levels are available right in the model picker and stay applied across requests • Cleaner chat header with the active model shown as a compact subtitle • Message actions stay out of the way while a response streams, then appear smoothly • Faster and more accurate model refreshes • Fixed: archiving every chat left an empty screen with no way forward • Fixed: unreliable model switching, accidental duplicate model loads, and crashes from duplicate model records • Fixed: clearer messages when models fail to load or a server returns an empty response

Sync That Doesn't Cry Wolf Fewer "Chat Diverged" warnings: Regenerating a response no longer triggers a false sync-conflict warning, and a real conflict warning now opens a sheet that actually shows the conflict. Better Error Messages No more silent "Generation stopped": When a request fails — wrong API key, server out of memory, a model crashing mid-load — the app now tells you exactly what happened instead of showing an empty response. MCP failures identified: A failing MCP tool integration is reported as a tool problem, not a false "server offline" notice. MCP Server Fixes Auth headers that stick: Adding an API key header to an MCP server now saves reliably — typed headers are included on Save, and the Save button re-enables after edits. Connection edits reach every chat: Fixing an MCP server's URL or key on the connection now applies to existing chats too, and no longer shows duplicate servers in the + menu. Toggles look like toggles: MCP servers in the + menu show a checkmark when enabled again.

3.0.0 is a major update focused on private, on-device AI and stronger local-model workflows. • Run supported GGUF models directly on your iPhone or iPad with new Local Model connections • Browse curated and Hugging Face local models, including supported vision models with projector downloads • Use private on-device Memory to save and recall helpful context from chats and documents • Choose per-connection defaults for Memory and instructions when starting new chats • Start private chats that never read from or save to Memory • Improved LM Studio v1 support, including reasoning/context settings, model loading controls, image requests, MCP tools, and reconnect behavior • Regenerate failed responses without duplicate messages • Added offline LaTeX rendering, better Mermaid spacing, expanded About details, and many release-readiness fixes

-Bug fixes

- Improved connection check - Bug fixes

Improves how the app handles missing model from your LLM server

- Bug fixes

- LM Studio API v1 support with new models tab - iCloud sync support to sync chats, connections and settings across your devices. - Added ability to share your current location in a chat - Added filtering - Added ability to archive chats - Improved chat search with better indexing - Tailscale connection bug fix - OpenClaw support (Chat with your OpenClaw server via its open Ai compatible channel) - Gemini image generation support

- LM Studio API v1 support with new models tab - iCloud sync support to sync chats, connections and settings across your devices. - OpenClaw support (Chat with your OpenClaw server via its open Ai compatible channel) - Gemini image generation support - Fixes a bug with connecting over Tailscale

- LM Studio API v1 support with new models tab - iCloud sync support to sync chats, connections and settings across your devices. - OpenClaw support (Chat with your OpenClaw server via its open Ai compatible channel) - Gemini image generation support

Bug fixes and performance improvements.

NEW FEATURES Searchable Instruction Picker Quickly find and select instructions from the chat header with a new searchable picker. Filter by name or content, and add new instructions without leaving the chat. Connection Health Checks Proactive LLM connectivity checks now detect unreachable servers before you send a message, showing clear status notices so you know immediately if there's a problem. IMPROVEMENTS Refined System Messages System message handling has been refactored for more consistent behavior when switching instructions mid-chat. Better Network Detection Enhanced error handler now recognizes more network failure patterns including offline states, DNS failures, and connection losses. BUG FIXES - Fixed error notice colors to use proper system colors for better dark mode support - Fixed the widget's photo button not opening the photo picker when launching a new chat

New Features: • Visual Polish: Refreshed app UI, softer widget buttons, clearer error banners, and subtle glass effects keep the app and widgets looking consistent across platforms. Improvements: • Improved Search & Performance: Chat search now looks inside every message, with improved app performance and enhanced search abilities. • Token Management: Token counting with per-connection presets, custom limits, and inline warnings so you know before a request would overflow a context window. The token meter respects your preferences—long-press to hide it, toggle it in Settings, and enjoy better spacing as the keyboard shows or hides.

Improved chat reliability with smarter background task handling. Streamlined image sharing with direct photo picker and clipboard paste support. Widgets now support starting new chats with images from your clipboard.

- Adds support for for OpenAI OSS models using the Harmony rendering format to the keyboard extension - Improves LaTeX rendering support

Full support for OpenAI OSS models using the Harmony rendering format New API profile for OpenAI OSS models with adjustable analysis level in chat settings LaTeX math formula rendering for easier reading Improved model compatibility with advanced configuration options

We've expanded our app's capabilities with support for OpenAI's newest open-source models, which now include a groundbreaking analysis block. openai/gpt-oss-20b openai/gpt-oss-120b

Easily interact with your PDF files like never before. With our latest update, you can directly add PDFs to your conversations, allowing for seamless communication with your language model about the content. Get quick summaries, ask detailed questions, or tackle complex topics—all tailored to your specific documents. Enhance your workflow and understanding effortlessly.

- Improve settings updates in keyboard extension. - Adds support to fetch models from Anthropic

Big update! This version brings major improvements to 3Sparks Chat: • New Tools Menu – Chat with images you upload or take with your camera (for models that support vision), upload text files, or fetch web content using Jina Reader. • iOS Widget – Quickly start a new chat from your home screen by snapping a photo, uploading an image, pasting from your clipboard, or starting fresh. • Share Extension – Share text or links from other apps straight into 3Sparks Chat to begin a conversation. • Smarter chat storage – Your messages are saved more reliably and automatically kept up to date. • Better chat cleanup – Easily delete chats right from the ChatView. • Enhanced sidebar – Smoother navigation and better performance. • Faster chat loading and improved handling of long AI responses. • Improved formatting – Cleaner display for tables and messages. Thanks for using 3Sparks Chat – we’re just getting started!

Refined sidebar animations and interactions across the app.

What’s New in This Update • Added support for Google Gemini, OpenWebUI, and AnythingLLM—bring your own server and connect to a wider range of models. • Improved connection settings now support custom route prefixes, unlocking compatibility with even more LLM providers. • Adjust text size in chat to fit your reading style. • General bug fixes and performance improvements for a smoother experience.

Added ability to resize chat messages text sizes

This update brings small enhancements behind the scenes.