Set reasoning_effort: "none" for custom openai compatible providers when reasoning is disabled.

• Apply reasoning settings in ZhiPu AI, #1109
• Custom OpenAI-compatible provider: Add reasoning: false to request body when reasoning is disabled, #1106
• Fix OpenRouter GPT-5-Image

• Anthropic: support built-in web search tool
• Display reasoning effort in the button

Fix Perplexity (they changed API url)

Move reasoning settings to the input area:

• You can now use @ to run one-off MCP servers:

• Disable observability for Anthropic when using oauth login method, because of incompaitibility

Fix styling

• add huggingface provider
• fix #1078 #1077

• Disable logging by default
• Fix custom provider reasoning setting

Add logging for LLM requests, see Settings > Advanced > Logging

• Update MCP SDK, fix "Load failed" error in some MCP servers that require oauth, #1070
• Fix reasoning settings for OpenRouter models, #1071

Fix a bug with Google web search (api key)

• New ChatWise Search provider: Ollama web search
• Fix: allow to turn off reasoning for Claude models

• Add Gemini 2.5 Flash Preview and Gemini 2.5 Flash Lite Preview
• Actually fix Google search, but if it still doesn't work it's probably because your IP is restricted

Fix Google web search (update the user-agent so it doesn't get blocked by Google)

Use native-tls instead of rustls, potentially fix #1046

Queue messages

When the model is still generating response, you can send messages to a queue, and queued messages will be sent automatically when the ongoing request finishes.

Fix OpenRouter error when using stored API key from Cloudflare AI Gateway

Now you can use API keys stored in Cloudflare AI gateway, so you don't need to manually enter API keys for AI providers on ChatWise: https://docs.chatwise.app/observability#api-keys

• Observability support, currently only supports Cloudflare AI Gateway https://docs.chatwise.app/observability.html
• Minor UI bug fix

Display token usage for the response message

Add UI for setting env variables and headers in MCP server

Fix Window restoration, closes #1020

Fix Volcengine #1013

Google/Anthropic providers

Remove Gemini 2.5 Flash Thinking Claude Opus 4.1 Thinking Claude Opus 4 Thinking Claude Sonnet 4 Thinking, you can use the reasoning setting in the sidebar to enable/disable thinking instead

New model settings in custom provider

Now you can tell ChatWise if reasoning can be disabled for a model

Display context usage:

Limitations for now:

• Only works for API that returns token usage
• Only updated after a successful response

Use custom titlebar on Windows and Linux

Fix UI

• Fix AiHubMix provider
• New app settings UI

• Save base64 images returned by MCP tools as local files to improve performance
• Allow to log out of authenticated MCP server

Fix Vercel provider