v1.27.0: Agent Sandbox

The AI Agent can now run scripts in a secure sandbox to analyze large results exactly, and it is no longer held back by tight row and query caps.

New Tools & Features

Script execution: The AI Agent can write short Python scripts and run them in an isolated sandbox against large Salesforce results, so it can count, filter and join exactly instead of paging through data. Scripts have no network access and cannot reach your credentials.
Result analysis: The AI Agent can now answer questions about a large result without loading all of it into the conversation, for cheaper and more accurate analysis.
Agent script execution settings: Workspace owners can turn script execution on or off and set a daily script-seconds budget under Account → Company, and see today's usage. Other members see the settings disabled with the reason. It is on by default, with a 300 seconds per day budget.

AI Agent: Removed the guardrails that clipped the agent. The 200/500 record cap is gone, SOQL now runs exactly as the agent writes it (no forced LIMIT), and describe, picklist and Apex log results are no longer trimmed. Results are capped only by a single size limit, and full payloads are kept for scripts and analysis.
AI Agent: Large results now keep 23–86% more rows within the same size limit. Stored result size grows from 5 MB to 25 MB.
AI Agent: Sub-agents get a larger step budget (20 to 50) for multi-org questions.
AI Agent: Clearer hints when the agent points at a path that doesn't exist in a stored result.

Stored agent results are now only readable from the conversation and org that produced them.

Fixed the left navigation rail overflowing the sidebar and painting over the page, most visible in demo mode.
Added a fallback message on the login screen when a sign-in error has no detail.