Skip to main content
Personal memory is the cross-expert user profile Profy builds for you. Every expert reads it, so you only have to say “I prefer TypeScript” once.

​
What it looks like

  • “User prefers TypeScript”
  • “User’s stack is Next.js on the frontend, Hono on the backend”
  • “User wants code comments in Chinese”
  • “User dislikes over-abstraction, prefers direct code”
You don’t enter these by hand — the system extracts them after conversations end.

​
Which conversations feed it (the easiest thing to get wrong)

Automatic extraction has two mutually exclusive paths, and they write to different owners:
Chatting with an expert does not produce personal memory. That path writes expert memory, visible only under that expert.To make a preference apply everywhere, either say it in free chat, or say “remember: …” explicitly in any conversation — explicit creation lets you scope it to personal.
The expert-side counter is keyed by user × expert in Redis and persists across sessions (expiring after 7 days of no interaction), so a series of one-line sessions still reaches the threshold. The intervals above are configurable defaults that operations can tune.

​
What gets filtered out

Extraction doesn’t dump the conversation into storage. Three criteria discard most of it: A deduplication check runs before every write: candidates matching existing memories are skipped, so you never accumulate ten synonyms of the same fact. Sessions that are pure greetings, thanks, or off-topic small talk return an empty list and write nothing.
The most reliable route is still to say it outright: “Remember: my project uses PostgreSQL 16.” Explicit memories bypass quality gating and land directly.

​
Managing memories

Settings → Memory lists every entry with its content, source, and current state.
If you just want it out of the way for now, disable it. Delete has no trash bin — it won’t come back as a disabled entry.

​
Personal memory vs expert memory

They complement each other: personal memory tells every expert you prefer TypeScript; expert memory lets your coding expert recall the architecture you agreed on last time. See Expert memory.

​
Memory is reference, not instruction

Experts load personal memory as context when a conversation starts. They use it to adjust tone, skip background you’d otherwise repeat, and avoid approaches you dislike.
But it’s reference: when the task calls for it, an expert will diverge from your general preference — you prefer TypeScript, but if the task is editing a Python project it won’t rewrite the project in TS. That’s correct behavior, not memory failing.

​
Failures and fixes

​
Privacy

  • Only you can view and manage your personal memory
  • Other users and expert creators cannot access it
  • Memories are not used to train models
  • Every entry can be disabled or deleted at any time
  • Temporary chat mode skips memory reads and writes completely

​
Keep reading

Expert memory

The expert-side memory graph

Personalization

Model preferences, tool run mode, and other account-level settings

Temporary chat

Conversations that leave no trace

Self-evolution

The full background review and skill improvement mechanism