Chat interface
Streaming chat, folders, per-chat settings, context fit, attachments, and the composer in LM Mini.
Chat is the core screen: stream replies with markdown/code/math, attach files and images, override settings per conversation, and keep threads in folders. On phone + Mac, sync keeps that list aligned.
What you get
- Real-time streaming with stop / regenerate
- Markdown, fenced code, and LaTeX
- Images plus files (including Markdown and spreadsheets where supported)
- Reasoning / thinking when the model emits it
- Auto-generated titles after a real conversation starts
- Folders and search
- Context usage so you can see how full the window is
- Optional Live Activity / notification while a reply finishes in the background
If there is no model yet, Get Started offers an on-device download or a computer connection instead of a raw error.
Starting a chat
- Tap New chat (or pick a Persona — Mini ships as a built-in character).
- Check the provider/model chip in the header.
- Type, dictate, or attach — then send.
Short greetings like “hi” may delay auto-title until there is enough to name the thread.
Folders
From the home list, switch to Folders. Create a folder, pick a color or cover image, then move chats in. Synced Home setups keep folder names and covers on both devices.
Per-chat settings
Open the chat overflow / settings sheet to override:
| Override | Notes |
|---|---|
| Model / provider | Pin this thread without changing global defaults |
| Temperature, max tokens, top-p/k | Sampling for this chat only |
| System prompt | Extra instructions on top of Persona/global |
| Reasoning | Force off for models that otherwise think aloud |
| Background / avatars | Look of this conversation only |
Chat overrides sit on top of param presets. On-device chats hide LM Studio-only sliders.
When context is full
Settings → Models → When context is full:
| Mode | What happens |
|---|---|
| Stop | Show an error when the prompt is larger than the model context |
| Roll | Drop the oldest messages; keep recent ones |
| Cut middle | Keep the start of the chat and the latest turns; drop the middle |
LM Mini packs history into about 90% of the loaded window so there is room for your next message. Compact can write a short summary of the older part so a long thread can keep going on the same model.
A context bar in chat shows how full you are. If you hit a limit with Stop on, switch to Roll or Compact and retry — or start a new chat.
Composer
In Settings:
- Shine composer (default) vs legacy compact composer
- Empty-chat starter cards on/off
- Return key: newline vs send
- Auto-scroll while streaming; scroll to bottom when you jump back
- Hide avatars for more reading width
Tips
- Regenerate when a reply went off-track instead of starting a new thread
- Use Group chat for multi-role work
- On Mac Home, the same chat list is a column next to the thread — see desktop