Obvious/Help Center

How Obvious remembers your preferences

Published February 27, 2026 · Last updated March 5, 2026 · 5 min read

Every time you tell the agent something important — your preferred date format, the name of your key account list, the way you like reports structured — you probably don't want to say it again next Tuesday. Memory is how Obvious holds onto those details between conversations.

But it's worth understanding what memory actually is, because it's not what most people assume.

What memory is (and isn't)

Memory in Obvious is selective, not comprehensive. Think of it like a notebook your agent keeps in its desk drawer. During a conversation, the agent might jot down a fact that seems worth remembering — a preference you stated, a decision you made, a data schema it discovered. Those notes persist. The full conversation doesn't.

This distinction matters. The agent doesn't have a photographic record of every message you've ever sent. It has the notes it chose to write down. Sometimes those notes are thorough. Sometimes they miss something you assumed was obvious. That's the tradeoff: memory is useful and lightweight, not perfect and total.

Two kinds of memory

Obvious stores memory at two levels, and the difference between them is practical.

User memory

User memory follows you across every project in your workspace. When the agent learns that you prefer bullet points over paragraphs, or that you always want dates in DD/MM/YYYY format, it stores that in user memory. Open a brand-new project next month, and the agent already knows.

User memory is personal. Other people in your workspace don't see it, and it doesn't affect how the agent works for them.

Project memory

Project memory is tied to a specific project. When the agent maps out a data schema, records a workflow decision, or notes that "Q4 means October–December for this team," that goes into project memory. Anyone working in that project benefits from it — the context is shared.

This is where institutional knowledge lives. If your team has been building a pipeline tracker for two weeks, the agent doesn't start from scratch every conversation. It reads project memory first and picks up where things left off.

What the agent actually remembers

The agent decides what to store based on what seems useful for future work. In practice, that tends to include:

  • Your preferences — formatting choices, communication style, how you like things structured.

  • Project facts — schemas, field mappings, terminology decisions, key artifacts and their purposes.

  • Decisions and context — why a particular approach was chosen, what was tried and didn't work, what the user confirmed.

It does not store entire conversations, raw data, or every passing remark. If you mentioned your dog's name in passing, the agent probably didn't write that down. (If you told the agent your dog's name is relevant to a project naming convention, it might.)

How to see what's remembered

You can ask the agent directly. Try:

What do you remember about me?

or

What's in your project memory?

The agent reads its memory files and tells you what it has stored. No digging through settings panels — just ask.

How to update or correct memory

Memory isn't locked. If the agent has something wrong, or you want it to remember something new, just say so:

Remember that our fiscal year starts in February.

Forget the old schema for the leads sheet — we restructured it.

Update your memory: I prefer tables over bullet lists for data summaries.

The agent updates its notes accordingly. You can also ask it to delete specific memories or clear everything and start fresh.

Honest limitations

A few things to know so memory doesn't surprise you:

  • Memory is selective. The agent stores what it deems important during a conversation. It can miss things. If something matters and you want it remembered, tell the agent explicitly.

  • Memory has size limits. Individual memory files cap at 100 KB, and each scope (user or project) has a total storage limit. This isn't something you'll hit in normal use, but it means the agent can't store a novel's worth of notes.

  • Memory doesn't mean the agent never asks again. Even with good memory, the agent might re-confirm details when the stakes are high. That's caution, not forgetfulness.

  • Memory files are plain text. The agent stores memories as simple files — markdown, text, or JSON. There's no complex database behind it. This keeps things transparent and editable, but it also means memory is only as organized as the agent keeps it.

The practical upshot

Memory makes the agent better the more you work with it. Your first conversation in a project might involve some setup — explaining your goals, your data, your preferences. By the fifth conversation, the agent already knows the landscape and gets to work faster.

It's less like talking to someone with perfect recall and more like working with a colleague who keeps good notes. Not every detail, but the ones that matter.

Next steps

Was this helpful?