GeoAI Memory: agents that remember your work

Until now, every new chat in GeoRetina AI started from zero. You restated the units your team reports in, explained the scenario a project had settled on, and went looking for last month's numbers before you could compare anything with them.

GeoAI Memory, new in GeoRetina AI 2.0, lets agents carry that context forward. They remember your preferences, your project decisions and the results of your analyses, so your next task builds on earlier work instead of repeating it. It's off until you turn it on, and you can see and forget anything it keeps.

Saved memory

• PreferencesIn your own words
• Project decisionsKept to their project
• Analysis resultsSaved when a run ends

New chatMemory on

Has flood exposure in Ottawa changed since our last screening?

Recalling from memory

Our flood exposure screening of Ottawa ran on October 14 with the 100-year scenario. I'll run it again on the latest data with the same method, check that the area and datasets match, and compare the two.

Open the chat it ran in

What it remembers

Memory holds three kinds of things, and nothing vaguer than that:

• Preferences: how you like work done, such as the units you report in. A preference is saved only when you ask the assistant to remember something or clearly state a lasting preference, and it's kept in your exact words, never a paraphrase.
• Project decisions: choices that hold for one project, such as the flood scenario it uses. These are also kept word for word, and only apply inside that project.
• Analysis results: saved automatically when an analysis finishes. Each one records what ran, where, when it ran and the period it covered, the key findings with their units, and the methods and datasets behind them. Every finding is marked as checked or not checked, so you know which values were independently recomputed.

Memory never keeps credentials, tokens or private links, and it doesn't store account details such as your credits or plan limits.

• Remember that I always report areas in hectares.

  Saved as a preference, in your words.

• Remember for this project: we use the 100-year flood scenario only.

  Saved as a decision for this project only.

• Use acres from now on.

  Replaces the earlier hectares preference.

• What did we find in our Ottawa flood screening?

  Answered from the saved result, with its date and a link to the chat it ran in.

• Run the same flood screening we did for Ottawa, this time for Gatineau.

  Reuses the earlier method on a new area.

• Has flood exposure in Ottawa changed since our last screening?

  Compares with the saved result, after checking the area, period and datasets match.

How it shows up in your work

Your preferences are in place from the first message of every chat. When a request depends on earlier work, the assistant searches its memory, matching on both wording and meaning, and you'll see "Recalling from memory" while it does.

Answers say where remembered information came from. The assistant names the source and its date, presents older results as history rather than as the current state, and links to the chat the analysis ran in, opening at that exact result. If several earlier analyses could match what you asked about, it asks which one you mean instead of guessing.

Comparisons get the same care. Before comparing a new result with an earlier one, the assistant checks that the units, area, period, scenario and datasets match, and points out anything that differs.

Memory informs the work; it never overrides you. Your current request always comes first, and saved text is never treated as instructions.

Change your mind at any time

To update a preference, just say the new one. "Use acres from now on" replaces the earlier note about hectares, and the "Saved to memory" card in the chat tells you what was kept.

Under Settings → Memory you can see everything that's been saved, forget any single entry, or reset memory to forget everything at once. Forgetting an entry also removes the earlier versions it replaced.

Private by design

Your memory is yours. Preferences are personal, project decisions stay within their project, and memory works only in your own private chats, never in shared ones. It isn't shared with teammates or across your organization.

Memory you build in the web app is kept separate from memory created through API v1 or apps connected over the MCP connector. It's stored in GeoRetina's own database, where only our servers can reach it, and it can never grant access, approve a run or change what your plan includes.

Turn it on

GeoAI Memory is off by default. Turn it on under Settings → Memory by switching on Use saved memory. While it's off, nothing is used or saved, and anything already saved stays until you forget it. Analyses that finish while memory is off aren't added later.

Developers can list and forget saved memory through API v1 and the MCP connector, and can run a request without memory when it shouldn't use or save notes.

Also new in GeoRetina AI 2.0

• Cloud AI GIS Agents: hand custom and long-running GIS work to agents that plan it, write and run the code in the cloud, and have their results checked.
• Scheduled Tasks: set an analysis to run daily, weekly or monthly, with an email when each run's results are ready.

Watch the GeoRetina AI 2.0 demo (2:40) to see all three at work.