> ## Documentation Index
> Fetch the complete documentation index at: https://docs.workser.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Memory: what an AI agent knows

> See and correct what an AI agent remembers, what the whole business remembers, and how Memory differs from Company knowledge and from the files you give it.

**Memory is what an AI agent has learned while working: facts about you, the business and its Tasks.** You can read it,
correct it, pin what must always be in mind, and forget by topic. It is separate from what you *write down* for it in
Company knowledge, and from the files you give it.

## The AI agent's memory

Open the AI agent's panel → **Profile** → **Memory** (the row reads "\<n> facts · \<m> to check"). The Memory view has
four parts:

| Part | What it holds |
| - | - |
| **Overview** | "What it knows about you and the business", what it learned lately, and **Always in mind** (pinned facts, read at the start of every Task). |
| **For you to check** | Facts it worked out by itself. "It worked these out by itself. Keep what's right; forget what isn't." Each has **Right** and **Not right**. |
| **Facts** | Every fact it keeps and what each replaced, with **Forget** and a "What it replaced" toggle. |
| **Map** | How what it knows connects: always in mind at the centre, facts around it, replaced facts outside their replacement, Task notes outside. **Open the large map** searches and filters. |
| **Task notes** | A note from each Task it did. |

**Forget by topic.** On **Facts**, type a person, project or topic, choose **Show what would go**, untick anything to keep,
then **Forget these (n)**. It reports "Forgot n facts."

**Pinning.** A pinned fact is in mind at the start of every Task, up to 2,000 characters. **Unpin** when it stops being
true; a stale pin costs a little on every Task.

**When memory is off**, it says so: "Memory is off for \<name>, so it keeps nothing new. Switch it on in its setup."

## What the business remembers

On the same **Memory** page, **What the business remembers** is the shared memory of what all your AI agents learned about
the business as they worked. It uses the same review and forget controls, with one difference: facts here are not pinned,
because pins belong to one AI agent.

## Memory vs Company knowledge

They sit on the same page and are easy to confuse:

| | Company knowledge | Memory |
| - | - | - |
| Who writes it | You, or an AI agent as a suggestion you keep | The AI agent, while working |
| What it is | Facts, products, customers, tone, **Rules** and **Never** rules | Facts, a map, Task notes, anything waiting for your check |
| How you change it | Add, edit, keep or drop an entry | Keep or correct a fact, forget a topic, pin |
| When it is read | Every Task | What is relevant, plus anything pinned |

A **Never** rule is the strongest kind: you choose what it is checked before (sending, paying, deleting or refunding,
signing up on a service) and it can ask you instead of stopping. With nothing ticked it applies to every action.

## Files and notes it reads

**Manage → Knowledge & files** is a third thing: price lists, policies, a brand guide, examples of good work. Text files are
read in every Task; PDFs and pictures are opened when needed. Up to 1 GB per file, and a text file up to 200,000 characters
is kept inline.

**Manage → What it keeps** is the inventory: what it remembers, the customers it knows, the skills it keeps, and its
computer and workspace — with a way to forget, remove or reset each.

## Related

* [Company knowledge](https://docs.workser.ai/ai-workforce/company-knowledge.md): the written facts and rules every AI agent follows
* [The AI agent panel](https://docs.workser.ai/using/agent-panel.md): where Memory and What it keeps live
* [Skills and agent kits](https://docs.workser.ai/using/skills-and-kits.md): what a kit does and does not carry


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.