Memory System
The Memory System makes atypica.AI remember things about you and your team across sessions. It stores persistent information about your work — preferences, history, ongoing projects — so the AI doesn't ask the same questions twice and gets smarter about your context over time.
What memory looks like
Memory is a structured markdown document that lives alongside your account. It contains things like:
- Your role and what you're working on
- Your research preferences (frameworks you favor, methods you avoid)
- Past research you've done, with links back to those sessions
- Recurring themes you've been exploring
- Open questions or interests you haven't pursued yet
When you start a new session, the AI loads your memory and uses it as context. You don't have to re-explain who you are or what you've already learned.
User memory and team memory
Memory works at two levels:
- User memory — private to you. Your preferences, your history, your open interests.
- Team memory — shared across everyone on your team. Your team's research conventions, shared history, team-level preferences.
If you work in a team, team memory propagates to every member's sessions, so a colleague picking up where you left off gets the same context you had.
How memory updates
Memory updates happen automatically after conversations. The AI reviews the session, extracts anything worth remembering (a new preference, a finding worth retaining, a project update), and adds it to the memory document at the appropriate place.
You can also view and edit your memory directly. If the AI remembered something wrong or you want to remove it, you can.
When memory reorganizes
Memory has a size threshold. When your memory document grows large enough, the system reorganizes it — summarises less-important sections, deduplicates overlapping entries, and keeps the structure clean.
Reorganization is automatic. You'll see the version number on your memory document increment when it happens, and you can review the change notes to see what was reorganized.
Why this matters
Without memory, every session starts cold: "Tell me about your company. What are you trying to learn? What have you already tried?" You repeat yourself, the AI forgets, and research projects that span multiple sessions lose continuity.
With memory, you start where you left off. The AI remembers your team's conventions, references past research, and asks sharper questions because it knows your context.
Limits
- Memory is AI-extracted, not perfectly curated. The system might miss things, or remember things imprecisely. Check it periodically.
- Memory is not a search index. Don't expect to query "what did we learn about X three months ago?" — for that, use your session history.
- Memory is per-user or per-team. There's no cross-team memory — your team's research conventions stay private to your team.
- Memory updates are not instant — they happen after a session ends, with some processing delay.
Privacy
Memory contains your conversational context. It's stored under your account, scoped to you or your team, and never shared with other users or used to train external models.
You can delete any memory entry at any time. Deleting an entry removes it from future sessions.