- Print
- DarkLight
Version 1.87.0
Released 03.11.2025
🪁 Answer Generation Model Improvements
Our current Retrieval-Augmented Generation (RAG) model has undergone a significant upgrade to enhance answer generation. We are introducing a new re-ranking algorithm that improves the relevance of the retrieved documents used in the final answer generation. By refining the retrieval process, we are able to increase the relevance and accuracy of our responses substantially. Stay tuned for more exciting enhancements coming soon!
🧩 Browser Extension Updated
AI Writing Tools
This feature will may only be visible tomorrow (03.12.2025)
With the Unleash Browser Extension installed, your team can access AI-powered Writing Tools directly from their browser’s context menu. Highlight any text, right-click, and choose an action—Refine, Translate, or Summarize. The assigned LLM chat assistant, configured in the Browser Extension add-on page in the Unleash Admin Center, will process your request. If the assistant has predefined instructions, the Writing Tool will follow them accordingly, ensuring consistency and accuracy in your content.
As an admin you must ensure that the assigned LLM chat assistant is configured with the following permissions:
The assistant has "Display on Chat" toggled on.
The assistant is shared with grant Interact permissions to the workspace.
Add-On Page in Admin Center
When you install the browser extension, a new Browser Extension page will appear under Add-Ons in the Admin Center. Here, you can configure settings for the extension, including the AI Writing Tools feature.
🔑 New API capabilities
New Chat Endpoint
The new Chat endpoint is replacing the Answer endpoint, adding the following capabilities:
Chat with an assistant. You can include previous conversation messages to provide context for the bot and ask follow-up questions.
Streaming - in this mode, the response is returned in stream of events via Server Side Events. Events can then be consumed to build a response object.
Chat with an External LLM assistant to get responses from GPT-4o and, in the near future, additional LLM models as we add them.
Added Scope to API Keys
With this release we are now allowing non-admin users to create API keys. To maintain security and provide more granular control over API usage, we are introducing API key limitations based on assistant scope and access levels. Users will now have controlled visibility and management of API keys: admins can view and revoke all API keys, while members can only manage the keys they have created.
When generating a new API key, users must define its assistant scope, determining whether it applies to all current and future general assistants or a specific set of assistants, including the default search assistant. This ensures that API keys can only be used with designated assistants—any API call referencing an undefined assistant will return an error.
Additionally, admins can choose an access mode:
Impersonated: API calls will be executed on behalf of a user specified in the header.
Non-Impersonated: API calls are made on behalf of the user who created the key, enforcing their permissions and access rights.
To support these updates, the API keys table will now include assistant scope and access mode columns, providing clear visibility into each key’s permissions. API keys remain non-editable after creation to maintain security and prevent unauthorized changes.
For more details, check out our updated API Reference Documentation
🕹️ New Slackbot Controls
We’ve introduced two new settings to give you greater control over how your bot interacts in channels:
Allow Responses to Messages from Other Bots
Now, you can choose whether your bot should activate when another bot sends a message in the channel. Simply toggle this setting on to allow bot-to-bot interactions, or turn it off to ensure the bot responds only to human users.
Exclude Email Domains
Prevent your bot from responding to users with specific email domains. This is useful for restricting interactions with certain organizations or groups.
To exclude domains:
Type the domain (e.g., acme.com) into the domain box.
Press Enter.
Repeat for multiple domains as needed.
These updates help tailor bot interactions to your exact needs.