• MCP: support image results return by tools
• MCP: now you can use env variables $HOME and $TMPDIR in the command
• Fix screenshot when the chat response contains images
• Fix OpenAI o1-pro

• Support Gemini image generation: choose the model Gemini 2.0 Flash Exp (Image Gen), currently only support the official Google AI provider
• Fix a web search issue related to port being used.
• Improve the LLM response performance (detail: previously I was using Tauri events to get streaming response, now it's replaced with Tauri channel)
• MCP: handle tool result with type: 'resource', like the repomix mcp and others

• Account: you can now log in with email + verification code
• OpenRouter: use DeepSeek V3 (free) to generate chat title and search keywords by default
• Text-to-Speech: add new OpenAI TTS model: gpt-4o-mini-tts and make it the default, you can also fine tune the voice with custom instructions now <img alt="tts" src="https://cdn.jsdelivr.net/gh/egoist-bot/images@main/uPic/g5uhhW.png" width="700">
• Add a button to expand system instruction editor <img alt="system instruction" src="https://cdn.jsdelivr.net/gh/egoist-bot/images@main/uPic/25moMZ.png">
• Data settings: you can now export and import chat history <img alt="chat history export" src="https://cdn.jsdelivr.net/gh/egoist-bot/images@main/uPic/vGHt7Z.png" width="700">

• Fix: google (free) web search sometimes returns 0 results
• Add OpenAI o1 pro model, the most expensive model so far
• Fix: prompt caching

• Fix: weird UX behavior when updating prompts / tools, fix #155
• Fix: toggling dark mode should apply to code blocks too
• Fix: MCP tool parsing error

Fix npx MCP servers on Windows.

• Implement prompt caching for OpenRouter Claude, now both Official and OpenRouter's Claude models have prompt caching by default
• Properly closing MCP servers when you close the app window
• Properly reconnect SSE MCP servers

• Fix a bug where you can't send files without text prompt
• Add "Delete response" to message context menu

• Add a button to reconnect all tools (MCP servers)
• Minor style tweaks

• Windows: fix settings layout
• Tool: fix tool result view

• Ollama gemm3 image support
• Fixes

• Initial Tools (MCP) support https://docs.chatwise.app/tools
• Fix editing assistant system instruction

• Fix web search
• Add Google Gemma 3 27B

• Add Amazon Bedrock
• Add AiHubMix
• Add Azure OpenAI o1 model
• Display assistant name under the model name in model dropdown button

Fix a translation error

• Improve mermaid UI
• Fix "save to assistant" button

• Fix AWS Bedrock Claude thinking error when using as OpenAI-compatible API in LiteLLM
• Make the copy button in code blocks sticky, add a button to collapse code blocks with more than 30 lines of code

• Don't trap focus when editing chat title
• Show pricing and context limit for Together AI models
• UI fixes

Fix a memory leak in 0.7.23.

How it happened: two items used the same id in a virtual list (virtua)

• Performance improvement: now you can scroll the model list without frame drops, no matter how many models you have
• Support custom API URL for Mistral AI and OpenRouter, adding a button to quick set API URL to ChatWise Proxy
• Fix chat title generation, fix left sidebar animation
• New dedicated changelog page: https://chatwise.app/changelog