A big update: memory, tools, documents, voice and a much better server. MEMORY • LLMTray can remember what matters between chats: your preferences, projects and how you like answers. Everything stays on this device; see, edit or forget any note in Settings › Memory. SEE AND READ • Show the model a window or your whole screen and ask about it. • The model can read a web page you point it to. • Spoken messages: talk to models that understand speech. • Paste an image straight into the chat with ⌘V. DOCUMENTS AND TOOLS • Make PDF documents with charts, tables and formulas right from the chat, then change any section later. • Connect tool servers (MCP) by URL, with sign-in for web services. • Thinking effort picker: Off, Low, Medium or High. SERVER • Several apps and coding agents can use one model at the same time. • OpenAI Responses API and Anthropic Messages API, alongside the OpenAI-compatible endpoints. • Optional API key. • Usage statistics per model, client and day, kept only on this device. • A new server log with search, filters and colors. MODELS • Faster answers with speculative decoding on supported models. • New model families, including LFM3. • Experimental: run large mixture-of-experts models that don't fit in memory by keeping part of them on disk. • The system's built-in on-device model can be used as a chat model, where available. And many fixes: steadier menu bar icon, chat search across long chats, clearer download and setup steps.