"Ana, use Eastern time."
"Keep the same chart style."
"Never use that table."
A new chat used to mean repeating the same instructions. Personal Memory lets Ana carry them into your next conversation, from the units you use to the filters you apply and the way you define a metric.
The feature is now in beta. You review what Ana saves and can edit the notes yourself.
Tell Ana what to remember
Ask Ana to remember an instruction and it proposes saving it. Ana can also pick up on a lasting preference during a conversation. It finishes answering your question, then asks whether to remember that preference. It never asks twice about the same fact.
Either way, you see the exact change in the thread before it's saved. Approve it with one click and the instruction carries forward. Nothing is saved silently.
Read, edit, or rebuild your memory
Your memory is a readable set of notes, starting with an ANA.md file. Open your profile at /me to see what Ana remembers and edit it directly. When your preferences change, you can update the instructions there.
You can also build a memory from instructions you've already given. Rebuild from my history drafts notes from up to 50 chats within the last 90 days. It reads your messages, including the corrections and preferences you shared during those conversations.
Rebuilding only replaces the section Ana generated. Anything you've written yourself stays exactly as it was. Running it again revises that section rather than adding duplicate notes.
Governed through your workspace
Memory lives in a per-member folder in your organization's version-controlled Ontology Library. Saves go through the same review process as other Library changes. Workspace admins can read memory files, just as they can read the rest of the Ontology.
Members can edit their own memory without permission to edit the shared company Library. Admins can disable memory saving for a role while leaving those members able to read their existing notes. When memory is off, Ana refuses requests to save personal facts.
Take it into Claude Code
Your preferences also load in Claude Code sessions opened through TextQL's refinery CLI. Run refinery init once to register the integration. After that, your own memory is in the agent's context before you type the first message. You can save edits from the CLI as well.
Enable the beta
Personal Memory is off by default and requires Ontology 3.0. A workspace admin can turn it on in the Beta settings.
The setup walkthrough separately offers to generate memories for members who don't have one yet. Enabling memory does not automatically read everyone's history; the admin must choose to run that step.
Once it's enabled, ask Ana to remember an instruction or open your profile to build a memory from your past chats.
