# Traits

> Switches that change how an agent behaves. They decide whether its character reaches the model, whether it keeps what it learns in conversation, whether it tidies its memory while idle, and whether it changes its own settings when you ask.

A **trait** turns one behavior of the agent on or off. Unlike the [system prompt](/docs/build/system-prompt/), which is text you write, a trait is a switch: the setting stays where it already lives, and the trait decides whether it is used.

## The traits

| Trait | What turning it on does |
| --- | --- |
| **Personality** | The character you set in the agent's [profile](/docs/build/identity/) is included in what the model reads. With it off, the agent works from its prompt alone. |
| **Self-evolving** | The agent keeps what it learns in conversation. Preferences, facts, and commitments go to its [memory](/docs/capabilities/memory/), procedures you teach it become [skills](/docs/capabilities/skills/), and it keeps an index of the files it made so it can open them again. Scheduled cleanup is Dream's job. |
| **Dream** | The agent tidies its memory while it is idle. It merges memories that overlap, corrects stale facts, closes finished commitments, and [archives](/docs/capabilities/memory/#archive) memories that are no longer useful. Then it gives each memory a one-line summary and search terms, so the next conversation finds them faster. On days when nothing changed, it rests. It runs as a [routine](/docs/automate/routines/) you can see and edit like any other. |
| **Self Configuration** | When you ask it in a conversation, the agent changes its own settings. It shows the before and after values on a card first, and nothing is saved until you press **Apply**. You choose, item by item, which settings it may change. |

<Screenshot
  name="studio-trait"
  alt="The trait section. Personality, Self-evolving, Dream, and Self Configuration each sit on a switch with a one-sentence explanation. Under Self-evolving, which is on, are the skills and file index switches and one learned skill; under Dream, which is on, is the clock that sets when it sleeps."
/>

## When to use them

**Personality** is worth turning on when tone matters, as with an agent that talks to customers or one whose replies you read all day. Leave it off when you want the agent to be plain and predictable, or when you are [tuning a prompt](/docs/build/iterating/) and want one fewer variable.

**Self-evolving** is for an agent that keeps working with the same people. It keeps a preference, a commitment, or a way of working you taught it, so you do not have to explain it again in the next conversation.

**Dream** earns its place on an agent that has been running for weeks and has built up memories. Overlapping and stale memories shrink, and the ones that remain get a summary and search terms. A new agent has little to tidy, and Dream creates no run on days when nothing changed, so turning it on early is fine.

**Self Configuration** is for an agent whose settings you keep adjusting as you use it. You say "be more concise" or "change your name" and the agent finds the setting and changes it, so you do not have to open Agent Builder and look for it. If there is a setting the agent should not touch, turn off that item.

<Aside type="note" title="Turning a trait off keeps its settings">
  Turning a trait off does not erase anything. The character stays on the profile
  and Dream's routine stays in the routine list, and the trait stops using them.
  Turning it back on restores the previous behavior.
</Aside>

## What Self-evolving learns

Self-evolving works during conversations. It has no routine of its own.

| Part | What the agent does | Switch |
| --- | --- | --- |
| Memory | When it learns a preference, a fact, or a commitment, it writes it to memory or updates the one it already has, and says so in one line. | Self-evolving itself |
| Skills | When you teach it a procedure it will reuse, such as a report format or an order of steps, it saves an instruction skill. When you correct a skill it learned, it updates that skill. It mentions the save at the end of its reply and follows the procedure from the next conversation. | Turn taught procedures into skills |
| Files | In every conversation it keeps the paths and descriptions of up to 15 files it changed most recently. It opens the one it needs and searches by topic for files not on the list. When it changes a file substantially, it updates the description too. | File index |

Both switches turn on with Self-evolving. Turn off only the parts you do not want.

The skills and files parts run only when the owner is the one talking. In conversations with endue.live visitors, API callers, or channel users, the agent does not create or change skills and does not load the file list, whether or not Self-evolving is on. Routine runs do not learn skills either.

### Reverting a learned skill

**Learned skills** in the Self-evolving settings lists the skills the agent created or changed, newest first. Each row shows whether the skill is new or updated, what the agent learned it from, and whether it is in use.

**Revert** takes an updated skill back to the version before the change, and removes a new skill from the agent and your account. The agent does not create or change a reverted skill again.

The agent does not change skills you created yourself. If it sees something to improve, it suggests the change in the conversation.

## What Self Configuration changes

With Self Configuration on, the agent reads its own settings and proposes the change you asked for. It saves only after you approve.

<Steps>

1. In the trait section, turn on **Self Configuration**.

2. Under **Settings the agent may change**, choose the items you want to hand over. Profile, Personality, and Role and languages start on.

3. In a conversation, say what to change. For example: "Change your name to Luna and be more concise."

4. The agent shows a settings change card. Each item shows the current value and the new one. A prompt change shows the lines added and removed.

5. Press **Apply** to save, or **Cancel** to leave everything as it is. An applied setting takes effect from the next reply.

</Steps>

<Screenshot
  name="studio-trait-self-config"
  alt="The Self Configuration trait, turned on. Under settings the agent may change, Profile, Personality, Role and languages, and Prompt are on; Model, Built-in tools, and Other traits are off. Below are two settings the agent changed, each with its before and after values, a link to the conversation, and a Revert button."
/>

| Item | What the agent changes | Starts |
| --- | --- | --- |
| Profile | Name and description | On |
| Personality | The [personality](/docs/build/identity/) coordinate. If the Personality trait is off, the values change but the tone does not. | On |
| Role and languages | Role statements and response languages | On |
| Prompt | The [system prompt](/docs/build/system-prompt/). Each change creates a new revision, marked in the revision list as changed by the agent. | Off |
| Model | The default chat [model](/docs/build/models/). It only picks models your plan includes. | Off |
| Built-in tools | Turning [built-in tool](/docs/capabilities/built-in-tools/) groups on and off | Off |
| Other traits | Turning Personality and Self-evolving on and off | Off |

When you ask for a change to an item that is off, the agent does not show a card. It tells you the setting can be changed in Agent Builder.

The agent also answers questions such as "which model are you using?" from its current settings. It can read items that are off.

Self Configuration runs only when the owner is the one talking. In conversations with endue.live visitors, API callers, or channel users, and in routine runs, the agent does not read or change its settings.

<Aside type="note" title="What the agent cannot change">
  Live publishing and visibility, channel and connector connections, API keys, the avatar, Dream,
  and archiving or deleting the agent stay out of reach whichever items are on. So do turning
  Self Configuration itself on or off and choosing its items. Change those in Agent Builder.
</Aside>

### Reverting a changed setting

**Settings the agent changed** in the Self Configuration settings lists the changes, newest first. Each row shows which item went from which value to which, and links to the conversation where it happened.

**Revert** takes that setting back to its earlier value. For the prompt, it creates a new revision restored from the earlier one. If the same setting has been changed again since, nothing is reverted and you are told so.

## When Dream sleeps

Dream's settings show a round clock with all 24 hours on one dial. You set when Dream sleeps with two handles, one for bedtime and one for wake time.

<Steps>

1. Turn on **Dream** in the trait section. A cleanup routine is created, starting at 04:00 bedtime and 07:00 wake time, every day.

2. Drag the two handles on the clock to move bedtime and wake time. They move in 5-minute steps. Dragging the arc between them moves both together and keeps the length. With the keyboard, focus a handle and use the left and right arrow keys for 5 minutes, or Page Up and Page Down for an hour.

3. Under the clock, pick the days Dream runs. At least one day must be selected.

4. Click **Save time**.

</Steps>

Cleanup starts at bedtime. Wake time is the last moment a cleanup can start. A cleanup that fired late, or a retry after a failure, rests for the day once wake time has passed and waits for bedtime on the next selected day. A cleanup that has already started runs to the end, even past wake time.

On a day when no memory has changed and there is no new conversation since the last cleanup, Dream skips the cleanup. No run is created, so there is no usage, and the skip does not count as a routine failure.

The clock follows the routine's timezone. If it differs from your browser's, the clock says which timezone it uses.

If you edited the schedule directly in the [routine](/docs/automate/routines/) details, that schedule keeps running. When the clock cannot draw it, such as the 1st of every month, the clock shows a notice, and clicking **Save time** replaces the schedule with the clock's time.

Each cleanup leaves a report of what it merged, corrected, and archived as a file in your [outputs](/docs/capabilities/outputs/), under `/reports/memory/`. The latest cleanup report link in Dream's settings opens it too.

<Aside type="note" title="Agents that tidied memory with Self-evolving">
  The scheduled cleanup that Self-evolving used to do moves to Dream. Dream picks up
  the cleanup routine that was already linked, so there is nothing to do, and the
  routine keeps its schedule and history. If that routine had stopped after repeated
  failures, Dream's settings show a warning and a **Resume** button.
</Aside>

## Limits

- Traits are per agent.
- The personality trait does not create a personality; it carries the one already on the profile. Change the character in [Identity](/docs/build/identity/).
- Dream runs as a normal routine and follows the same rules. An unattended run cannot ask you anything, and each cleanup counts toward [usage](/docs/account/usage/) like any other routine run.
- Dream needs the memory [built-in tools](/docs/capabilities/built-in-tools/) switched on.
- Self-evolving only creates instruction skills. It does not create skills that call outside services.
- Self-evolving saves at most two skills per reply. A learned skill belongs to your account, so attaching it to another agent applies it there too.
- The file index covers only the [outputs](/docs/capabilities/outputs/) the agent made. Files attached to a conversation and files on a device are not included.
- Dream does not edit or archive a pinned memory. It only adds a summary and search terms.
- Memories Dream archives are not deleted on their own. Delete the ones you do not need in [Memory](/docs/capabilities/memory/#archive).
- The clock only sets times by day of the week. Any other schedule has to be edited in the routine details.
- Self Configuration only proposes changes you asked for. The agent does not change settings on its own judgment.
- Every Self Configuration change needs your approval on the card. There is no way to save without it.
- One card holds up to four items. When you approve several at once, some may fail to save, and the agent reports the result for each.
- The card shows a long prompt change only in part, and says so. The full text is in the prompt section's revisions after you apply it. If you want to read all of it first, ask the agent for a smaller change.- If the agent wrote the first prompt for an agent that had none, Revert cannot remove it. Edit it in the prompt section.
- Changes apply to the agent's next [run](/docs/work/runs/).

## Related

<CardGrid>
  <LinkCard
    title="Identity and profile"
    href="/docs/build/identity/"
    description="Where the character the personality trait carries is set."
  />
  <LinkCard
    title="Memory"
    href="/docs/capabilities/memory/"
    description="What Self-evolving writes and Dream tidies, including archive and index."
  />
  <LinkCard
    title="Skills"
    href="/docs/capabilities/skills/"
    description="The instruction skills Self-evolving saves. Attach and detach them here."
  />
  <LinkCard
    title="Routines"
    href="/docs/automate/routines/"
    description="How Dream runs: schedule, history, and what happens on failure."
  />
  <LinkCard
    title="System prompt and revisions"
    href="/docs/build/system-prompt/"
    description="Where a prompt changed by Self Configuration is kept as a revision."
  />
  <LinkCard
    title="The context window"
    href="/docs/work/context-window/"
    description="Traits are part of what the model reads on every step."
  />
</CardGrid>