What's New in 2.6 Skills are here. Teach MyLLM new tricks by installing Skills — compact, reusable sets of instructions your assistant loads on demand to handle a task exactly the way you like it. NEW — Skills • One-tap install from the Marketplace, or right inside a chat with simple /skill commands. • A growing open collection: meeting minutes, commit messages, email replies, code review, study flashcards, plain-English explainers, web research with sources, SQL, regex, unit tests, changelogs, and more. • Your assistant automatically loads the right skill when your request matches — nothing to set up per chat. • Bring your own: paste a SKILL.md or install from any GitHub repo. Every install asks you to confirm first. Improvements • More reliable tool use — actions some models occasionally wrote out as text now run correctly. • Polish and stability fixes throughout. As always, MyLLM runs on your own local server or fully offline on your device — your data stays yours.

What's New (v2.5) NEW: Email that reads like email • Open any message and see it fully formatted — images, layout and styling intact — in a tap-to-open view, instead of stripped-down plain text • Move unwanted mail straight to the bin from the conversation — it's reversible, landing in your Trash rather than gone for good • Check, read, search and summarise your inbox, all in one place NEW: A tidier agent plan • Agent mode's step-by-step plan is now a collapsible tree that tucks into a single line • Expand it to watch each step tick through to Done, with a status dot and progress count • Clear at a glance, and out of your way when you don't need it Improvements: • Hands-free Walking Mode replies on its own again — pause after speaking and it sends automatically, no tap needed • Agent mode is more reliable: far fewer empty replies, and quicker, more focused follow-up answers • News by outlet is more accurate — added Fox News, and asking for a source that isn't available now tells you instead of quietly returning a different one • Check for Updates now works even if you previously skipped a version, and Settings shows your current and latest version side by side • Stability and performance polish throughout

What's new in 2.4 Chat scrolling that gets out of your way. Drag up to read previous messages while a reply is being written and the view stays where you put it — auto-follow only kicks in when you're already at the bottom. Plus the catch-up at the end of a long response no longer leaves you trailing the last paragraph. Smarter multi-step agents. Agent mode plans, runs tools, and reviews its own work across iterations — with subagents picking up specialised subtasks so the main thread stays focused on what you asked. Walking Mode hears you again. Voice transcription is reliable when the on-device model is still warming up, and errors are surfaced clearly when something does go wrong instead of looping silently. Cleaner stop button. Cancelling a generation no longer leaves orphaned tool calls or duplicate replies in the conversation. Plus smaller polish and stability fixes throughout the app.

This update is all about smoother, more responsive chats. • Long, streaming replies no longer stutter or freeze — even on lengthy responses • The Stop button now actually stops generation instantly, keeping whatever was written so far • Steadier auto-scroll that stays with the latest text while the assistant is typing Thanks for using MyLLM! Please share your experience and feedbacks to make this app better - we love it !!!

What's New in 2.3 On-device Python charts & visuals. Ask for a graph and MyLLM computes and draws it right in the chat. matplotlib, NumPy and Pillow run fully on-device — no internet — and Python can now produce interactive tables and visuals, not just static images. Interactive HTML & diagrams. MyLLM builds rich visuals — charts, styled tables, flowcharts and sequence diagrams from a single sentence, even small interactive apps — as a card you tap to open full-screen and navigate, sandboxed safely on your device. Web pages and HTML emails now render properly instead of as plain text. Context inspector. Tap the token counter to see exactly what's in the model's context: per-message token usage, what's been trimmed to fit, and how close you are to the limit. On-device Python charts. Ask for a graph and MyLLM computes and draws it right in the chat. matplotlib, NumPy, and Pillow now run fully on-device — no internet — so plots, data crunching, and image generation just work. Interactive HTML artifacts. MyLLM can now build and show rich visuals — charts, diagrams, styled tables, even small interactive apps — as a card you tap to open full-screen and navigate, all sandboxed safely on your device. Context inspector. Tap the token counter to see exactly what's in the model's context: per-message token usage, what's been trimmed to fit, and how close you are to the limit. Scheduled routines. Turn any workflow into a daily automation. MyLLM runs it in the background and delivers the result as a notification — your morning briefing, a news digest, or anything you build. Faster, more natural voice. Voice conversations now speak as they think — sentence by sentence, instead of waiting for the whole reply — and speech recognition runs on-device for lower latency and privacy. Smoother chat. Fixed scrolling jitter during long, streaming replies. Plus reliability fixes, including more dependable background briefings.

What's New in 2.2 • Attach documents to chats. Tap the paperclip and choose Document to add a PDF, text file, CSV, JSON, or source file alongside (or instead of) a photo. The model reads the contents and answers your question about it. • Beautiful tables. Formatted tables now render with proper columns, alignment, and row stripes instead of raw pipe-and-dash text. Wide tables scroll horizontally so cells stay readable. Long-press a table to copy it as markdown (Notion, GitHub, Slack) or as a native table (Notes, Numbers, Pages, Excel). • Smoother long replies. Scrolling now stays glued to the latest output without flickering or jumping past the content. While the model is still typing, you can scroll up to re-read earlier parts in peace — a floating arrow appears so you can drop back into the live response with a tap. • Slash commands anywhere. The / tool palette now appears mid-sentence, not just at the start of an empty input. Type something, then / plus a few letters, to surface a matching tool and insert it inline. • Various stability and polish improvements.

MyLLM 2.1 - What's new Apple Intelligence — native tool calling Tools now run through Apple's FoundationModels framework directly, so the model invokes them natively instead of emitting JSON into your chat. Reliable, faster, and integrated end-to-end. New offline models Five new on-device models, all downloadable from Settings → On-Device Models: • Liquid LFM2 350M — ultra-light, sub-300MB • Liquid LFM2 1.2B — Liquid's flagship small model • SmolLM2 1.7B — HuggingFace's strongest tiny model • Llama 3.2 1B — Meta's compact multilingual • Gemma 3 270M — Google's lightest IT model Manage Built-in Tools A new Settings screen lets you switch off built-in tools you don't use. Disabled tools are removed from every conversation, shrinking the prompt and speeding up responses. Re-enable any time. More reliable tool results Email, calendar, contacts and other list-style tools no longer truncate mid-result — the model now sees the full response and stops fabricating tails. Safer on-device loading We check available memory before loading a model and pick a smaller context on tight devices, so you get a clear error instead of the app being silently killed. Plus bug fixes, polish, and SourceKit hardening.

MyLLM 2.0 Tool Library Browse the curated catalogue at opticell-limited.com/myllm-wiki and install tools with one tap. The app confirms every install, with a clear warning when a tool needs network access. 22+ ready-to-use tools across seven themes: • Research — companies, books, dictionary, country facts, scientific papers, EU VAT, marketing trends • Science — chemical elements, clinical trials, proteins, NASA imagery, DNA tools • Productivity — weather, astronomy data, public holidays, word counter • Finance — crypto prices, currency conversion • News — Hacker News top stories • Developer — GitHub user lookup, IP geolocation • Fun — dad jokes Build your own Custom tools in Settings → Tools let you write JavaScript the model can call. Define JSON-schema parameters, opt into network access, and export tool packs to share with anyone. You stay in control Every install link asks for confirmation. Disable any tool individually. Connect MCP servers with per-tool toggles and optional auto-approve for trusted servers. Reliability Major stability fix for custom-tool execution (continuation lifecycle bug).

MyLLM now speaks MCP. • Connect MCP Servers — Plug any remote Model Context Protocol server into MyLLM and its tools instantly become available to the AI. From specialist services like Lawve for legal research, to your own internal MCP servers, MyLLM meets them where they live. • You stay in control — Every MCP tool call asks for your approval before running. Mark a server as trusted to skip the prompt, or disable individual tools without removing the whole server. Bearer tokens are stored in the iOS Keychain. Find it under Settings → MCP Servers.

What's New in 1.8 • Walking Mode — true hands-free voice with talk-to-interrupt • Live Activity — see agent progress on the Lock Screen & Dynamic Island • Smarter Memory — semantic recall, all on-device • Branch chats from any message, pin a model per chat • Tool Trace — see exactly what tools the AI used • New Siri shortcuts: Search, Brief Me, Summarize a Page • New Performance panel with response-time stats & tuning controls • Replies up to 4× faster thanks to coordinated server-side work • Model now always knows today's date Plus polish email tool and bug fixes throughout. Thanks for using MyLLM — feedback always welcome.

What's New in 1.7 Build your own tools, share them with anyone, and focus each chat on exactly the tools you need. Custom Tools Create your own tools right in the app. Describe what they do, write the JavaScript, and the model can call them like any built-in tool. Tap any existing tool to refine its description and steer when the model picks it. Find it under Settings → Tools Available. Network-Enabled Tools Custom tools can opt in to making HTTP requests via a sandboxed fetch(). Hit any REST API, webhook, or service from inside a chat — perfect for hooking MyLLM up to Notion, Home Assistant, your own server, or anything else with an endpoint. Off by default, switched on per tool when you actually need it. Tool Packs Export all your custom tools as a single .myllmpack file to back up or share with friends. Import a pack to install everything at once. Use the share menu in Settings → Tools Available. Per-Chat Tool Focus Disable noisy tools for a single conversation without touching your global settings. Toggle by category or individual tool from the chat composer. Useful when you want a focused chat without the model getting distracted reaching for tools you don't need right now. More Offline Models Hunyuan MT 1.5 1.8B is built in for on-device translation across 33 languages. You can also add any other GGUF model yourself by pasting its download link — find both in Settings → On-Device Models.

What's New (v1.6) NEW: Slash-command tool picker • Type / in any chat to open a live, filterable list of every tool available to your model • Tap a tool to drop a natural-language hint into the input — handy when you know exactly which tool should run but don't want to spell it out • Filters by name and description as you type FIXED: Files saved by the assistant are now visible in the Files app • Anything written via write_file shows up under Files → On My iPhone → MyLLM • Open, share, move, AirDrop, or import to other apps directly — no more files trapped in the app sandbox FIXED: Email "recent messages" now actually shows recent messages • Mailboxes with deletion history could surface very old messages as if they were new — corrected • Unread filter and recent-mail listing both audited

MyLLM 1.5 is everywhere iOS is. • Siri & Shortcuts — "Hey Siri, ask MyLLM…" • Home & Lock Screen widgets • Dynamic Island Live Activity for video renders • Voice Conversation — hands-free, full-duplex • Daily Briefing with your calendar and inbox • iCloud sync across iPhone, iPad, and Mac • 8 new image style presets + Animate-from-image Plus a fix for the Email account save confirmation.

MyLLM 1.4 — Generate images and video from your chat • New Generate Media panel — tap the photo icon to turn prompts into images with FLUX.2-klein or ~4-second clips with LTX-2, routed to your own ComfyUI server • Live progress tracking: queue position, model-load hint, step-by-step sampling • Bearer-token authentication for Ollama and ComfyUI — Keychain storage hardened so tokens never leave this device • Share Configuration: export your server setup as a JSON file and import it elsewhere, with secrets excluded by default • Strengthened privacy manifest, VoiceOver labels across the new screen, HTTPS-only for remote servers • Numerous fixes and stability improvements

What's New in MyLLM 1.3 SMART AUTO-RECONNECT - MyLLM now remembers your server setup between app launches — no more re-configuring Ollama or LM Studio every time you open the app. - On startup, the app verifies your saved connection. If the server moved or changed IP, MyLLM automatically scans your network and reconnects. IPv6 SUPPORT - Connect to LLM servers over IPv6, ready for modern and internet-facing networks. - Server discovery now detects both IPv4 and IPv6 network interfaces. - Enter IPv6 server addresses manually in Settings (e.g.http://[2001:db8::1]:11434). IMPROVED SERVER DISCOVERY - Network scanning covers both IPv4 and IPv6 addresses, including IPv6 localhost (::1). - Faster fallback — if the saved server is unreachable, auto-discovery kicks in immediately.

What's New in v1.2 1. Advanced Problem Solving — decompose_problem, step tracking, synthesis 2. Real-Time Updates & Alerts — daily briefings, smart digest, trending, countdowns, recurring alerts 3. Enhanced Device Integration — network/screen/locale info, brightness control, notification management 4. Auto Update Check — new feature highlighted 5. Bug Fixes & Performance — thread safety, memory leaks, delegate callbacks, truncation messaging

MyLLM 1.1.1 brings Apple Intelligence support, email integration, and offline improvements. MyLLM can now run on Apple Intelligence using Apple's on-device Foundation Models — no server setup required. If you have an iPhone 15 Pro or later running iOS 26, the app automatically uses Apple Intelligence when no LLM server is detected, complete with full tool calling support for web search, calendar, contacts, and more. You can upgrade to a local Ollama or LM Studio server anytime for larger models. This update also adds email tools. Connect your Gmail, iCloud, Outlook, or any IMAP account in Settings, and ask the AI to check your inbox, summarize unread messages, or search for emails by sender or subject. Credentials are stored securely in the Keychain and never leave your device. We've also fixed several issues with offline mode: the status bar now correctly shows your on-device model name, and small local models no longer produce duplicated responses caused by the system prompt instructing them to use tools they can't access.