Version 1.3 makes the agent easier to come back to and easier to trust. You can reopen past conversations, keep your work through a page refresh or a pop-out, save searches you run often, and manage your saved memories. Search results read better, scheduled agents deliver their work where you'll see it, and the agent underneath is more dependable at creating records and answering follow-ups.

Highlights

• Chat history — Every conversation is kept. Reopen one from the history panel, rename it, or search across all of them.
• Your work survives a refresh or a pop-out — Refresh the page or open the agent in a pop-out window, and your results and your chat are still there.
• Save and reuse searches — Save a search under a name and run it again later against today's data.
• Manage your memories — See what the agent has learned about your preferences, and delete anything that no longer applies.
• Scheduled agents deliver their work — Choose where results go (Agent, Email, or both), and see the email drafts a run wrote.
• A more dependable agent — A rebuilt reasoning engine cuts repeated actions, lost context and duplicate records.

New Features

Chat History

Past conversations used to be unreachable once you left them. A history panel, opened from the top bar of the chat, now lists your conversations newest first. You can rename any conversation in place, and type into a search box to narrow the list across all your conversations, not just the ones on screen. 

Your Work Survives a Refresh or a Pop-out
Refreshing the page, by accident or otherwise, no longer means starting over. The results grid comes back as it was, and so does the chat conversation, so you can carry on with a follow-up question.

The same goes for opening the agent in a pop-out window. The pop-out shows your current conversation and the results it has already produced, so you can pick up where you left off. Opening or closing the pop-out doesn't reset your session.

Save and Reuse Searches
If you run the same search regularly, save it under a name you choose and run it again later against current data. You can do this from the chat or from the full results screen. 

The results toolbar is also tidier: two rows of buttons are now one, and record actions (Edit, Add Note, Export CSV) appear only once you select records.

Memory Management
When you tell the agent how you like to work, it saves that as a memory and applies it in later conversations. 

A memory view, reached from Agent Settings, lists what is stored for your account. Each item can be deleted individually. A deleted memory stops shaping the agent's answers from your next conversation.

On-Demand Agents and a Simpler Agent Builder
Agents you've set up can be run on demand as well as on a schedule. The agent builder is more compact, and the agent panel and its notifications have a refreshed look.

Choose Where a Scheduled Agent Sends Its Results
The agent setup screen now has a delivery choice next to how the agent runs: send results to the Agent, to Email, or to both. The choice is saved with the agent. Scheduled agents also accept "every weekday" as a schedule.

Email Drafts From Scheduled Runs
The notification that a scheduled run has finished now takes you straight to the email drafts it wrote, all of them if there are several. You can edit a draft's recipients, subject and text before sending.

Improvements
Search results

• Fields from related records — Add a column from a related parent record (for example, a Lease's Property Year Built), up to three levels away. This works for every searchable object.
• Change columns by asking — Ask for "show the size range and the submarket instead of the broker name" and the table on screen updates straight away.
• Currency and number formatting — Money fields show the currency symbol and thousands separators ($2,450,000) in the full results table and the details panel.
• Redesigned Search Criteria panel — Aligned columns, a field and operator picker you can type into, and reliable filtering on related-record fields, including when a saved search is reopened.

Working with records

• "Which record did you mean?" — When a name matches several records, the agent shows a candidate list to pick from and continues the same conversation once you choose. Questions from the agent now show the records they're asking about.
• Duplicate handling — When a new record looks like one that already exists, the agent shows the possible matches so you can pick the existing record and avoid creating a duplicate.

CSV import

• Import up to 200 records at a time, across more than one object. Accounts and contacts can share a file: each row creates an account and a linked contact, so up to 100 rows.
• Columns land on the right object, the mapping is editable, and one column can map to more than one object.
• Quoted values containing commas are parsed correctly.

Subagents and notifications

• Notifications show the correct relative time and appear in the right channel without duplicates. The panel closes on a second click of the bell or a click outside it.
• Search result tables from a scheduled run keep their record context. "Upcoming Meetings" no longer shows past events, and the Task side panel no longer loads indefinitely.

Reliability and administration

• New agent engine — Each step now chooses its tool and fills in its arguments in a single model call. This replaces the older multi-stage pipeline and removes a class of repeated actions and lost context. The old engine has been retired.
• Better error logging — Errors are recorded with more detail for admins.