Änderungsprotokoll

Here's what's new in this update:

✨ New

• A new Provider UI

  A new UI has been added for configuring providers and their API keys, with unified Enter API Key widgets across all providers.

• Read specific files in Knowledge Base (KB)
  You can now read specific files in your Knowledge Base directly.

🛠 Improvements & Bug Fixes

• OpenRouter API streaming: Improved handling of OpenRouter response API streaming.

• Fixed an issue where redundant newline characters caused invalid Claude prompt cache entries.

Here's what's new in this update:

✨ New

Claude automatic prompt caching: TypingMind now applies Claude's new automatic prompt caching, improving response efficiency for repeated or similar prompts.

Plugin Execution Timeout (Self-Hosted Teams): Admins can now set a custom timeout limit for plugin execution via an environment variable. The value should be specified in milliseconds.

🛠 Improvements & Bug Fixes

Agents/MCPs avatars: Fixed an issue where agents/MCPs avatars could not load.

Custom params in multi-model responses: Fixed an issue where custom params were not applied in some multi-model response generation cases.

Here's what's new in this update:

GPT-5.4 and GPT-5.4 Pro models: OpenAI's latest models — GPT-5.4 and GPT-5.4 Pro — are now available to use in TypingMind.

🛠 Improvements & Bug Fixes

• Granular permission for role assignment in the team version: Admins now have finer control over who can assign roles to members in the organization.

• GPT O-series models formatting: Formatting instructions in the system prompt are now only applied to o-series models, keeping other models clean and unaffected.

• Audio recording for org users: Fixed an error that occurred when recording or transcribing large audio files — audio is now uploaded before transcription to prevent the issue.

• Inline agent in chat: Your pinned inline agent is now remembered when you switch between chats and come back — it will no longer disappear unexpectedly.

• Gemini cost calculation: Fixed an incorrect cost calculation for Gemini — image prompt pricing has been removed to reflect the correct usage cost.

• Grok endpoints: Enabled CORS proxy support for Grok endpoints, improving connectivity and compatibility.

We've added two new models to TypingMind:

• Gemini 3.1 Flash Lite: Google's latest lightweight model, offering fast and efficient responses.

• GPT-5.3 Chat: OpenAI's newest Chat model, bringing improved capabilities to your conversations.

Generate images with Gemini 3 Pro Image + Other models

You can now use Gemini 3 Pro Image alongside other AI models in the same chat, get text responses from your preferred model and generate images, all in one place.

Fixes & Improvements

• Fixed: Shared chat links were not accessible in TypingMind Teams. Shared chats can now be opened correctly.

• Improved: The old Google Search plugin has been removed for new users. Existing users will see a deprecation warning. Please switch to the Web Search or Web Search (Serp) plugin instead.

Here's what's new in this update:

✨ New

• Nano Banana 2 model: Google's latest image generation model that combines the advanced capabilities of Nano Banana Pro with the speed of Gemini Flash. The new Nano Banana 2 (Gemini 3.1 Flash Image) model is now available in TypingMind.
  

🛠 Improvements & Bug Fixes

• Perplexity plugin: You'll now see a clear error message when something goes wrong.

• Custom OpenAI models: Reasoning now works correctly for custom OpenAI models.

• Gemini search tool: Knowledge Base is automatically turned off when using the Gemini search tool to avoid conflicts.

• Self-hosted Docker: Support better compatibility for more container environments.

• Remove member button: The button no longer gets stuck on loading after removing a member.

Deprecation: Google Search Plugin

The Google Search plugin on TypingMind is being deprecated. Google has announced that all Programmable Search Engines will stop working on January 1, 2027, so this plugin will no longer function after that date.

We recommend switching to one of these alternatives:

• Use the default Web Browser tool provided by your model provider

• Use the Perplexity Search plugin instead

Learn more about Google's announcement

Deprecation: Legacy Plugins Removed

The following plugins have been removed from the default/built-in plugin set as they relied on the legacy Plugins Server:

• PowerPoint Generator

• Word Generator

• Web Page Reader

Fixed: MCP OAuth Callback URL (Self-Host)

Fixed an issue in self-host deployments where the MCP OAuth callback URL wasn't working correctly.

Gemini 3.1 Pro is now available in TypingMind!

Google's latest model brings stronger reasoning, better coding, and improved multimodal understanding, all with a 1M token context window.

Why use it?

• Smarter reasoning, scores 77.1% on ARC-AGI-2 (2x better than Gemini 3 Pro)

• Better at coding and agentic tasks

• Handles text, images, audio, video, and code with a 1M token context window

Select Gemini 3.1 Pro from the model picker to get started.