v1.10.7 Oct 26, 2026
Home share: direct Wi-Fi, choose your Mac's model, smoother sync
- Added a direct Wi-Fi connection between your phone and Mac on the same network: faster, encrypted, and works without internet, with the relay as a fallback
- Added: pick any model downloaded on your Mac from your phone, and your Mac switches to it
- Added Keep Mac awake while sharing and Open LM Mini at login on the Mac
- Added a Get a model button on the Mac sharing screen when no model is downloaded
- Improved chat sync between your phone and Mac: only changes are sent, pictures and memories come along, edits and deletions sync, and deleted personas and memories stay deleted
- Improved big models on your Mac get time to load, and your phone shows when your Mac is still loading
- Improved shared Mac models use a 16K context (8K on 8 GB Macs)
- Improved your phone suggests updating LM Mini on your Mac when it's out of date
- Improved Remote Access now says how your connection is protected
- Improved messages when your Mac's pairing changed and you need to scan the new QR code
- Changed glass effects are off by default on new installs; turn them on in Appearance
- Fixes chats with your Mac's models failing when Tool Use is off, in group chats and in Arena
- Fixes your phone showing no model right after pairing with your Mac
- Fixes your Mac showing Ready with no model to share
- Fixes long replies through the relay being cut off as if they were finished
- Fixes the Mac stopping another app that used the same port
Download LM Mini Home 1.10.7
v1.10.5 Oct 24, 2026
Reliability: chats, sync and on-device thinking
- Improved on-device thinking models like Qwen 3.5: their thinking now goes in the collapsed thoughts section, and you can turn Think off
- Fixes the chat list and the open chat getting stuck loading until you restart the app
- Fixes the chat still moving while a reply is written when "Follow new replies" is off
- Fixes large models failing to load over Remote Access with server error 524
- Fixes image generation failing on Automatic1111 after loading a ComfyUI workflow
- Fixes Ollama replies that stall mid-way being lost; the text already written is kept, with the reason
- Fixes Pro Search showing "temporarily unavailable" when LM Studio blocks per-request MCPs; you now see which setting to turn on
- Fixes servers that aren't LM Studio showing "Endpoint not found"; you now see how to add them instead
v1.10.4 Oct 20, 2026
Apple Intelligence & the 2026 models
- Added Apple Intelligence: chat with the model built into your iPhone, iPad or Mac, with no download (iOS 26 / macOS 26 on supported devices)
- Added the 2026 on-device models: Qwen 3.5, Gemma 4, Bonsai (1-bit), LFM 2.5, MiniCPM 5, Ministral 3, SmolLM 3 and Granite 4, grouped by family with one "Best quality" pick for your device
- Improved setup: use Apple Intelligence, download a model, or connect your computer. Servers found on your Wi-Fi now show right on the card
- Added a chat style choice during setup: full width or bubbles
- Added LM Mini now asks your name in your first chat and remembers it
- Added Arena is now free: race models on your own prompts, compare answers side by side, and keep a history of your races
- Added group chats in folders
- Added Apple Intelligence as a group chat participant
- Improved Apple Intelligence replies start faster in longer chats
- Improved voice chat: tap the mic button to stop recording
- Improved the home screen with a clearer Chat button and a New Persona tile
- Improved models you downloaded before stay available under Older models
- Fixes PDF export failing on Cyrillic, Greek and emoji, and long messages overlapping across pages
- Fixes Export All Chats garbling non-English text and overwriting chats with the same title
- Fixes Obsidian exports with invalid frontmatter, and hidden reasoning tags showing in exports
- Fixes LM Studio failing with "failed to parse grammar" when structured output and web search are both on
- Fixes long tool and MCP replies from LM Studio ending early
- Fixes group chats answering with only your main model when it runs on this device
- Fixes LM Mini Home participants being rejected in group chat setup
- Fixes No folder not taking a chat out of its folder
v1.10.3 Oct 14, 2026
Shortcuts MCP tools, Keychain & connection fixes
- Added Turn off Remote Access to the "Can't reach your Mac" message
- Improved API keys, tokens and server headers are now kept in the Keychain / Keystore instead of plain settings
- Improved German and Russian voices are now labelled as Piper voices
- Fixes servers on your local network being blocked on mobile data even when a VPN could reach them; the Wi-Fi notice now only appears when the server really can't be reached, and can be closed for the session
- Fixes OpenAI-compatible providers like Groq failing with "property 'chat_template_kwargs' is unsupported"
- Fixes Stop not stopping the reply on your computer over Remote Access and USB, which made your next message wait behind it
- Fixes USB replies arriving all at once instead of streaming
- Fixes emoji and non-English text showing broken characters over Remote Access, and LM Mini Home
- Fixes Remote Access no longer checking your computer after a failed reconnect; models now reload when it's back
- Fixes Ollama, oMLX, Jan or Unsloth pointing at a dead relay after you turn Remote Access off or unpair
- Fixes a cleared API key still being sent
- Fixes Add Server ignoring the headers you entered, so servers behind Cloudflare Access couldn't be added
- Fixes server addresses ending in /v1 failing, and public addresses without http:// or https:// defaulting to unencrypted http
- Fixes the model you pick not being the one sent to Ollama, oMLX, Jan, Unsloth and cloud providers
- Fixes switching servers carrying over the wrong model, such as a GPT model being sent to Mistral
- Fixes a locked Pro server showing as active while chat used a different server
- Fixes group chat sending LM Studio participants to your cloud provider
- Fixes Ollama group chat participants failing over Remote Access
- Fixes your LM Studio custom headers and Cloudflare Access secret reaching Unsloth servers and cloud chats
- Fixes LM Mini Home retrying the relay every second when its session was refused
v1.10.2 Oct 11, 2026
Remote Access switch & clearer connection errors
- Added a Remote Access on/off switch
- Added Tap to retry on messages that didn't send
- Added a mobile data notice when your server only works on Wi-Fi
- Improved connection errors
- Improved memories
- Fixes Pro Search answering without searching when image generation is on
- Fixes on-device models crashing when the app goes to the background
- Fixes Kokoro voice crashes, and blocks a voice pack that keeps failing to load
- Fixes Switch provider opening the model list instead of your providers
- Fixes a failed send leaving the chat stuck on sending
- Fixes your LM Studio custom headers and Cloudflare Access secret being sent to cloud providers and image servers
v1.9.12 Sep 28, 2026
Unsloth context, ComfyUI workflows & clearer limits
- Added Context Length for Unsloth — the Model Parameters slider loads the model at that size, and asks before reloading if it is already loaded at another size
- Added Load settings from a ComfyUI workflow, and a name so the next image is saved into that list
- Added Resize images for physical batch — shrinks a photo when it would be too big for a local model and crash the server
- Fixes the context bar sitting far below LM Studio — a finished reply uses LM Studio’s own token count, including thinking
- Fixes high thinking using up the whole reply — Mini tries again with low thinking, and tells you that you can raise max tokens to keep high thinking
- Fixes an unknown MCP looking like an app bug — Mini says the server rejected that tool, and does not offer Send to support unless it is Pro Search or Code Sandbox
- Fixes a camera photo disappearing when you come back to the chat
- Fixes “Loading model” showing again after the reply has already moved on
- Fixes ComfyUI graphs, including Z-Image, losing the prompt or ignoring the graph’s own sampler
- Fixes Connect / Share with phone sending image generation to the chat server instead of AUTOMATIC1111 or ComfyUI
v1.9.10 Sep 21, 2026
Full-width chat, gallery & crash fixes
- Added Unsloth Desktop as a free local OpenAI-compatible provider
- Added Generated images gallery in the chat ⋯ menu
- Added website-icon pile for web search sources (tap to expand, tap again for results)
- Added support tickets can attach the error log
- Full-width replies is now the default — your messages stay in a bubble, assistant text uses the whole row
- Improved the chat ⋯ menu (glass) and the streaming status shimmer
- Hides embedding models from Model Section
- Fixes image generation crashing when the server isn't AUTOMATIC1111 (
Null is not a List)
- Fixes two New Chat taps creating the same conversation id
- Fixes memory extraction still running after you stop a reply
v1.9.8 Sep 17, 2026
Home personas, lid-close reconnect & clearer host errors
- Added Sync personas when Home chat sync is on — pick which personas copy between phone and Mac, including photos and memories
- Added Full-width replies in Appearance — your messages stay in a bubble, assistant text uses the whole row
- Added Compress image and resend when LM Studio stops because an attached photo is huge
- Fixes Share with phone dropping after you close the laptop lid and not reconnecting until you refresh
- Fixes Home sync replacing an existing chat title, and leftover image-prompt text showing in titles
- Fixes LM Studio / llama.cpp stopping mid-reply looking like a Mini crash — now says the model server stopped, and points at context / max tokens
- Fixes a sent chat that never reached LM Studio (proxy or tunnel) looking like a Mini bug
- Fixes Kokoro English voice generating silence
v1.9.0 Sep 8, 2026
App Lock, JAN AI & Persona Voices
- Added App Lock (Pro) — set a 4 or 6 digit PIN and choose how long Home can stay closed before asking again
- Added JAN AI as a local provider
- Added Ability to assign Kokoro, ElevenLabs or Grok voices to Personas from one dropdown
- Improved composer tools bar — icons when collapsed, chips when expanded
- Improved Persona setup — tap Preferred Model to change it, edit avatar from the pencil
- Fixes tools being sent to models that don't support them
- Fixes Ollama regenerate going to the wrong backend
- Fixes App Lock toggle not updating after setting a PIN
- Fixes Pro tags showing for users who already have Pro
- Fixes mixed icon styles in Settings
v1.8.16 Aug 12, 2026
- Share with phone — QR pair once. Encrypted relay to your Mac — no port forwarding or VPN.
- Run models on this Mac — bundled local inference, no Terminal or Homebrew.
- Share LM Studio & Ollama — phones can use backends already running on the Mac.
- Desktop shell — nav rail for Chats, Personas, Arena, and Settings.
- Menu bar — keep Share with phone available while you work in other apps.
Download LM Mini Home — primary Mac app.