Agent Memory, AI Slack App, and More!
Your Agent Now Has a Memory!
Runbear just got smarter—it now remembers your specific instructions and adapts accordingly. For example, tell it "Be more polite," "Always summarize in bullet points," or even "Always fetch pricing details from the web," and it'll personalize future conversations based on what it has learned about you. Give it a try!
Note: This feature is available by default on Claude and still improving. Memory management tools aren't available yet, but your assistant is always learning behind the scenes. Stay tuned—more updates coming soon!
Switch to the New Slack AI App
We’re excited to launch Slack AI App support for custom bots! You can now add AI agents directly to your Slack’s top navigation bar — the fastest way to bring your assistant closer to your team.
Follow this guide to enable the Slack AI App and make it even easier to get the most out of Runbear!
Smart Reply Now Supports Targeted Responses
Smart Reply now gives you more control — you can choose exactly which users or groups should receive the AI Agent's response.
For instance, if you'd like your customer service team to review customer inquiries before replying to the customer, just set them as the recipients. It's an easy way to keep things efficient and accurate.
This update gives you more flexibility to tailor Smart Reply to your team’s needs and workflows.
Track Changes to Your Agent’s Instructions
You can now view and compare changes to your assistant’s instructions over time.
Whether you’re tracking tweaks or debugging behavior, this gives you better visibility into your assistant’s evolution.
Other Changes
- Support GPT 4.1
- Runbear now supports OpenAI’s latest GPT-4.1 model, with better instruction following and support for long conversations—up to 1 million tokens of context. More accurate, more helpful responses, even in complex threads.
- Improved Model Context Protocol (MCP)
- We’ve added a confirmation prompt system for actions that could modify external resources—like sending emails, posting in Slack, or creating Jira tickets. Whenever possible, your assistant will ask for your go-ahead before taking action, helping prevent surprises while keeping things smooth.
- We also fixed a token limit issue affecting long messages in MCP workflows.
- Stabilized KB sync
- Fixed issues where some documents were embedded multiple times, and ensured that disconnected files are properly removed from the vector store.