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

# How memory works

> How a coworker actually builds up what it knows about you — from a passing comment to a lasting memory.

Memory is what lets you skip re-explaining yourself. The first time you work with a coworker, you give some context — your preferences, your team's terminology, facts about an ongoing project. From then on, they carry that forward, so a second request can be shorter than the first.

<Note>
  Memory is about **you staying in control**, not about being watched. Everything a coworker remembers is visible to you, and you can correct or delete any of it — see [Your data & control](/en/memory/your-data) for how.
</Note>

## Why this matters

Without memory, every email would need to start from zero — re-stating who your competitors are, what your brand voice sounds like, how you like a report formatted. With it, a second request to the same coworker gets straight to the point, because the groundwork is already there.

## What actually gets remembered

Coworkers pick up on the kind of things a good colleague would remember naturally, not everything you ever say:

* **Preferences** — how you like reports formatted, which details matter most to you, tone and structure choices you've corrected before.
* **Team language** — the names, shorthand, and terminology your team actually uses, so a coworker doesn't ask you to define the same acronym twice.
* **Ongoing context** — facts about a project that stay true across several tasks, like who your main competitors are or what a recurring initiative is called.

For example: mention once that your team calls a project by an internal nickname, and a coworker will start using that nickname back to you rather than the formal project title — without you ever asking it to.

## How a memory actually forms

Memory doesn't happen all at once — it goes through a real, three-stage pipeline, and the middle stage (**Dreaming**) is visible in the console rather than being an invisible black box:

<Steps>
  <Step title="A note gets picked up">
    During a conversation, a coworker notices something worth keeping — a preference, a piece of context, a pattern in how you work. This is raw and unfiltered: just a small fact, not yet trusted as reliable.
  </Step>

  <Step title="It gets reviewed — the &#x22;Dreaming&#x22; step">
    In the background — after a task finishes, after your profile updates, on a nightly sweep, or during a catch-up pass — notes are reviewed and checked against what was actually said. This step is visible in the console as **Dreaming**, a running log you can open on your Memory page to see exactly when a review happened and what it did — nothing gets promoted to a lasting memory just because it was mentioned once; it has to hold up against the real conversation history first.
  </Step>

  <Step title="It becomes a memory, with a confidence score">
    Confirmed notes become memories you can see on your Memory page. Each starts with a moderate confidence score and rises as it gets reconfirmed in later conversations — or quietly fades if it never comes up again.
  </Step>
</Steps>

Concretely: if you mention your team's project nickname once, that's a note. If it comes up again in a later conversation, that note gets confirmed and becomes a real memory. If you never mention it again, its confidence slowly drops instead of staying pinned at full strength forever — the system doesn't assume something said once six months ago is still true today.

## Seeing what a coworker remembers

Everything is visible in the [console](https://console.serviceplan-agents.com), on the Memory page. Each memory shows how confident and current it is, using a simple freshness label:

| Label           | What it means                                   |
| --------------- | ----------------------------------------------- |
| **Strong**      | Recently confirmed and used — reliable.         |
| **Established** | Solid, but not reinforced very recently.        |
| **Fading**      | Old or unconfirmed — may no longer be accurate. |

You can sort the list by category, by confidence, or by how recent it is, so it's easy to find the one you're looking for.

## What this is not

Memory isn't a transcript — a coworker doesn't remember and replay everything you've ever written verbatim; it distills durable facts and preferences, not full conversation logs. It also isn't instant: a note needs to survive the review stage before it's trusted, so don't expect something you say once to immediately change how a coworker behaves — that's a deliberate safeguard against a single offhand comment being treated as a hard fact.

## How this connects

Memory is what makes [delegation](/en/coworkers/delegation) feel continuous rather than starting over every time, and it's shared across every coworker and channel — see [Your data & control](/en/memory/your-data) for the ownership side (your profile, corrections, exports, and scope), or [Skills](/en/skills/overview) for the difference between what a coworker *knows how to do* versus what it *remembers about you*.

<Tip>
  Memory works quietly alongside every conversation — you don't need to set anything up. Head to [Your data & control](/en/memory/your-data) any time you want to check, correct, or clear what's there.
</Tip>