Run AI on only the records you choose
Select records in a table and run an AI field, a validation check, or a transformation on that selection. It is especially useful after an import: process the rows that just arrived while leaving the rest of the module untouched.
Agent Chat works out the job before it acts
Agent Chat now identifies whether you want it to change data, look something up, save a password, draw a chart, or build an automation, then chooses the right tools for that message. If you spend several messages building a module, it stays on that job until you change the subject.
Results come from your data
A record link or chart appears only when Agent Chat has actually read the relevant records. If it has not, the chat answers in words rather than presenting something that looks like a result.
Keep secrets out of chat and code
When an automation needs a key or password, Agent Chat opens a form for the value instead of asking you to put it in the message. Boost.space stores it as a Secret rather than inside the code, and using the same Secret name again replaces the previous value.
Also in this release
- The filter opens ready for input, so you can start building the condition immediately.
- Mentions, emoji, and formatting now work in the rich-text editor.
- Only administrators can edit a field group.
- The MCP token page warns you when a system still uses classic built-in modules that agents on the newer address cannot see.
- Attachment lists load faster.
- A new version no longer reloads the app while you work. A New version prompt lets you choose when to refresh.
- The browser no longer offers to translate the interface; switch language in your settings instead.
- Agent Chat keeps streaming through longer answers.
Agent Chat and the MCP server continue to update between platform releases. Over these weeks, the MCP server moved from v2.3 to v2.4. It can search Boost.space Docs, explain field types to an agent editing them, and return errors when code fails.
