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

# Agent profile

> An agent's identity — name, role, personality, safety boundaries, and languages.

# Agent profile

The profile is the agent's identity: who it is, how it talks, what it won't do,
and which languages it answers in. It's organized into four areas.

## Identity

| Field    | What it controls                                                                        |
| -------- | --------------------------------------------------------------------------------------- |
| **Name** | The display name shown wherever the agent appears.                                      |
| **Role** | A short description of what the agent is for. It anchors how the agent presents itself. |

Keep the role concrete — "IT support agent for the engineering org" is more
useful than "assistant".

## Personality

These fields shape *how* the agent communicates. All are free text — write them
in plain language.

| Field                    | What it controls                                                                                        |
| ------------------------ | ------------------------------------------------------------------------------------------------------- |
| **Personality**          | The agent's tone and manner — direct, warm, terse, formal, or your own blend.                           |
| **Conversational style** | How it moves through a turn: when to ask a clarifying question, when to confirm, when to cite a source. |
| **Greeting & closing**   | How it opens and closes a conversation.                                                                 |

If you leave one blank, Cobalt applies a sensible default.

## Safety

This is where you set the limits.

| Field                 | What it controls                                                                                                                                                        |
| --------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Safety boundaries** | Topics the agent must not engage with, and requests it must escalate. Write one rule per line; mark a rule **hard** (refuse outright) or **soft** (prefer to hand off). |
| **Custom guidance**   | Org-specific instructions, terminology, and edge cases that don't fit elsewhere.                                                                                        |

Common boundaries: never repeat unmasked PII; don't promise SLAs; hand off to a
human for HR, legal, or compensation; refuse write actions for users outside your
directory.

## Languages

The **language policy** controls which languages the agent will respond in:

* **All languages** — the agent detects the user's language and replies in it.
* **A specific set** — choose the languages you support, set a primary language
  used as the fallback, and optionally write a refusal message for unsupported
  languages (Cobalt translates it into each supported language for you).

## Editing safely

The profile is versioned. Edits take effect on the next conversation, and you can
review history. Reworking tone or boundaries on a live agent is low-risk — change,
test in [Preview](/get-started/create-your-first-agent), and iterate.