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

# Personas

> Choose a synthetic person, save project model settings, and keep each run's selected behavior.

A **persona** is the synthetic person who talks to your agent. A test says what
that person wants. You can use one persona in many tests.

## Core behavior and project settings

The persona's identity name, personality, and language form its core behavior.
Each core version is immutable. The display name and description are live
labels; changing them does not create a version.

Your project saves one set of model settings for each persona it uses:

* The language model provider and model.
* The speech recognition provider and model.
* The speech generation provider and model.
* The voice ID and speaking speed, from 0.6 to 1.5.

Every test in your project that selects the same persona uses these settings.
Another project's choices cannot change yours. Saving model settings creates no
core version. Egma saves complete defaults on first use, so later changes to
recommended defaults do not replace your saved choices.

## Use, create, or clone

Open a predefined persona and choose **Use** to review its model settings. You
can also select it in a test; Egma then saves its defaults for your project.
Choose **Edit settings** to change the saved values later. Supported model
choices come from the catalog. A custom provider voice ID is allowed; it does
not have to appear in the recommended voices.

Choose **New persona** to create a custom persona with behavior and settings.
The complete persona is saved together. A failed save leaves no partial persona.

Choose **Clone** to copy the source's current core and your project's effective
settings into a new custom persona. A clone has its own settings and history.
It receives no later source updates. Existing tests keep the source; edit a
test to select the clone explicitly.

## Edit the current core

You can edit a custom persona's current behavior. A changed save creates the
next core version. If another edit wins first, read the current core before
editing again.

Older versions are read-only. They contain no record of past model choices and
cannot be restored as the current core. Predefined core behavior is maintained
by Egma; clone it to write your own instructions.

Compatible Egma core updates become current for future runs automatically.
Your project's saved model and voice settings and test selections stay in
place. Egma checks that the new core accepts saved settings before it becomes
current. An incompatible update is refused and leaves the earlier core usable.
Catalog defaults do not replace existing choices; custom clones receive no
source updates.

## Runs keep their selection

A test selects a persona definition. When a run is created, each simulation
keeps the selected current core version and complete project settings. A later
edit cannot change queued work or its retries. A future run uses the newer
core and settings without requiring a test edit.

Common Egma runtime instructions remain separate from the authored core. A
saved core is not a historical copy of every instruction in a deployed runtime.

Deleting a custom persona removes it from authoring lists. Its core history and
saved simulation evidence remain readable. A test that selects a deleted
persona must select an active persona before it can run again.

See the [persona API](/api/personas) and [test files](/cli/test-files).