TypingMind Release Notes

Updates for TypingMind Teams admins.

Redesign

Admin plugins page redesign: The admin plugins page has a cleaner layout that is easier to scan and manage. Admins can now install multiple MCP servers at once, instead of adding them one by one.

Sidebar menu: Admin navigation is reorganized into Appearance and System Config. Knowledge Base now lives under System Config instead of the top-level menu and pinned shortcuts.

Appearance includes Branding, Custom Domain, Welcome Message, and Chat Theme.

System Config includes Chat Features, System Prompts, Knowledge Base, Memory, API Integration, and Emails.

Changes

Control memory on admin: Added a new admin Memory page. Admins can enable Memory for users, who choose which scopes to turn on in Memory → Settings. Memories belong to each user and are never shared between users, even with the same agent. They are not saved to agents in the admin dashboard or added to their instructions or knowledge base.

• Global memory applies across all chats, agents, and projects of one user.
• Agent scoped memory applies only in that user’s chats with a specific agent.
• Project scoped memory applies only in that user’s chats within a specific project.

Multi-model response finalization: Org admins can now set default finalization settings for multi-model responses, which users can still customize per response.

Improvements to math, tables, and diagrams in chat, plus Claude prompt caching for agents.

Math rendering

Math formulas in chat now render more accurately. Added support for \underbracket and \overbracket. Also fixed layout issues with fractions, delimiters, and accents.

LaTeX math such as \[ ... \] and \( ... \) sometimes showed up as raw text when the AI response also included code (code blocks or inline code). Those expressions now render properly.

Claude prompt cache sharing across agent chats

Claude prompt caching now reuses an AI agent’s instructions and training files across all chats with that agent. Starting a new chat, or reopening an old one, no longer re-caches the agent’s full context. You are not paying to cache the same agent context again every time you open a chat.

Copy and expand tables

Tables in AI responses now have Copy table and Expand table buttons. These replace the width toggle. Copy table pastes a formatted table into documents, spreadsheets, and email. In a plain-text editor, it pastes as tab-separated values. Expand table opens the table full screen, with a copy button, so wide tables are easier to read.

Mermaid diagrams

TypingMind can now natively display Mermaid diagrams in the AI’s response without needing a plugin.

Toolbox is a free, built-in plugin that lets your AI analyze files, run code in your browser, and return finished documents, charts, and media.

Turn it on under Plugins → Toolbox, attach a file, and ask for the output you want.

• Documents and PDFs: Read, edit, merge, or generate Word, Excel, PowerPoint, and PDF files. You can also match the style of a template you upload.
• Data: Clean, reshape, and combine CSV, JSON, and XML. Filter, join, and run statistics, then chart the result.
• Reports: Build charts, dashboards, and HTML reports you can preview and download.
• Images, video, and audio: Resize, watermark, and convert images. Clip, crop, and compress video. Make GIFs. Extract and convert audio.

Note: Code runs on your device. Network access is off by default. You don’t need to write any code.

Toolbox guide

Memory is now supported on TypingMind 👀

Your AI can remember context about you, your preferences, and the way you work - so you don’t have to repeat yourself in every new chat.

TypingMind now handles generated images and chat files more seamlessly across Plugins, MCP tools, GPT Image plugin, and Artifacts.

For users

• Attachment/Generated files are now managed under About this chat → Files.
• GPT Image 2.5 can now edit any image available in the current chat, instead of being limited to images from your latest message. The AI will automatically select the relevant images to use.
• Interactive Canvas and Web App Builder can now use images directly from the chat. When you share an artifact, the images displayed inside it and are included as part of the shared artifact.

For plugin developers

• Plugins can now receive files directly from the chat using the new typingmind-attachment parameter type together with the read_chat_attachments permission. Learn more

• Image cards now support a new image.filename field. Base64 images returned by cards are also automatically saved as chat files. Learn more

• Chat files can now be referenced directly in plugin output:

  • HTML: tm-attachment://<id>
  • Markdown: #tm-attachment:<id>

  Learn more

• For GitHub-hosted plugins, JavaScript functions can now be stored in separate files using the new codeFile field. Learn more

Behavior changes

• previousRunOutput no longer includes output from respond_to_ai. Learn more
• When importing a plugin from a private GitHub repository, TypingMind now uses the main branch by default. To import from another branch, add **`

Teams Updates

Added user group filtering to org analytics, allowing users to view tokens and messages broken down by user group, plus a new "User Groups" option when exploring analytics data.

Integration API key generation is now restricted to account owners and admins; other roles can still view and revoke existing keys.

UXUI Improvements

Open "About this chat" in a convenient side panel, better view artifacts on both desktop and mobile.

Improved the button UI in rendering results for artifacts, HTML, and code.

Improved support for reading PowerPoint (.pptx) attachments.

Added GPT-6.1 Sol and GPT-6.1 Sol Pro to TypingMind — OpenAI mid-tier upgrade (Sept 29, 2026, DevDay)

Near-Astra on coding, computer use, and pro work at ~1/5 Astra’s token price. Same $2 / $10 as GPT-6 Sol; cached input now $0.10. 1.05M context.