# What a persona is

> A persona is a saved role: instruction, reference material, a pinned model, a voice, and more. Why to build them, what one holds, and how they carry into chats and Beams.

- Canonical: https://big-agi.com/docs/feature-personas
- Minimum tier: open  [Open ⊂ Free ⊂ Pro - a tier only ever ADDS to the Open baseline]
- Kind: concept · Last content update generation: 2026-07-27 · Review by: 2027-07-27
- Answers: What is a persona? · How is it different from pasting a long prompt every time? · Can I attach my own reference material to one? · Can I pin a model, its settings, and a voice to a persona?

A persona is a saved role: instruction, reference material, a pinned model, a voice, and more. Build one for consistent, predictable work with an entity that knows you, your job, and the task at hand. It carries instructions and advanced chat directives, custom Beam merges included. It carries data, images included, which most AI applications do not support inside a persona. Its memories and data are editable by you and by the persona itself, and travel into every new chat it works on. The instruction defines personality, interaction, behaviour, and constraints. A typical Big-AGI expert keeps two to four personas in frequent use.

## What one holds

| Part | What it gives you |
|---|---|
| **Chat System Instruction** | The behaviour, written once. Its placeholder says it plainly: `Defines the persona's behavior or AI task`. |
| **Documents** | Up to ten reference files, sent with every chat that uses this persona. |
| **Chat Starters** | Clickable openers shown above an empty composer, `Shown at the start of a new chat`. One can be set to ask for a file before it runs. |
| **AI Models** | At most one locked model, optionally with its own settings. |
| **Voice** | The voice it speaks in, and the line it opens a call with. |
| **Custom Merges** | `Extra buttons in Beam's merge selector when launching Beam with this persona.` |
| **Reinforcement Instructions** | Text added before or after each of your turns - the discipline that keeps a long thread from drifting. |

Instructions can hold placeholders, filled in when the message is sent. The editor offers them as two pickers:

| Picker | Placeholders |
|---|---|
| `Advanced output hints:` | `{{PreferTables}}` · `{{RenderMermaid}}` · `{{RenderPlantUML}}` · `{{RenderSVG}}` · `{{RenderHTML}}` · `{{RenderChartJS}}` |
| `Variables:` | `{{Today}}` · `{{LocaleNow}}` |

Those hints are why a conversation produces charts or diagrams at all. The renderers are always on; what varies is whether the model has been *told* it may emit those blocks. Switch to the persona that ships for that job - Chart Maker carries the chart contract - or add the hint to your own. Naming the format in your message works too ([what renders](/docs/feature-attachments#what-renders)).

## Building one

| Choice | What it offers |
|---|---|
| `Create New` | `Configure all details yourself` |
| `Copy Existing` | `Duplicate and modify existing persona` |

1. In the personas app, pick one of those. Personas are best written by hand; the creator turns pasted text into a starting instruction.
2. Give it a `Name` and a single-emoji `Symbol`; both are validated, and a short description shows in the list.
3. Write the instruction. Add the output hints you want from the two pickers rather than typing them.
4. Optionally lock a model, add documents, pick a voice.
5. **Save & Finish**.

Verify it took: the persona appears in the chat persona dropdown (`Ctrl + P`), and starting a new chat with it shows its starters above the empty composer.

## Reference documents

| Documents | What applies |
|---|---|
| sources | files · a web page · the clipboard · drag and drop · a blank note you type into |
| pipeline | the same as chat attachments - drafts, converter chips, `Review & Confirm Documents` |
| cap | ten per persona |
| effect | the persona has those documents in-context during chats and Beams |

Same pipeline, same chips ([attachments](/docs/feature-attachments)).

## Pinning a model, a voice, and merge buttons

A persona pins one model at most.

| Setting | What the panel says |
|---|---|
| unpinned | `Unlocked (default). <Name> will use the model selected in each chat` |
| `Lock Chat Model` | `<Name> will always use this model for chatting.` |
| `Customize Model options...` | `Your persona will use this exact configuration and ignore any prior model settings:` |

Locking overrides the dropdown in every chat that uses the persona. Settings are opt-in on top of the lock ([model settings](/docs/model-settings)), and changing the pinned model clears them, by design.

A voice is stored as an engine plus its parameters, and is what the persona speaks with in a call ([voice](/docs/feature-voice)).

Each entry under **Custom Merges** becomes its own button in Beam's merge selector, ahead of the built-in ones. That is where a merge prompt you wrote lives permanently ([combining answers](/docs/multi-model-fusion)).

## Applying one to a chat

```
🧠 Default · 👨‍💻 Developer · 📈 Chart Maker · 🔬 Scientist · 🚀 Catalyst · 👔 Executive · 🖌️ Designer
```

Seven ship with the product, Designer hidden. Assign a persona to the whole chat from the dropdown at the top (`Ctrl + P`) - that is the primary way to use one. The chat's persona carries into Beam: every card of a run works with it. Type `@` in the composer to address a different persona for a single message.

## What Open gives you, and what the hosted app adds

On your own instance the persona **creator** turns pasted text into one saved system-prompt instruction. Built-in roles ship alongside it, selectable in a chat.

The persona as an editable object is the hosted surface. It covers the dashboard, the editor, its own documents, a locked model with settings, a voice, merge buttons, and `@` addressing. Backing personas up or having them follow you to a second device is Pro **[Pro]**.

> **Pro:** Chats, personas and rambles stay on this device and browser only, unless you subscribe to Pro: cloud backup and multi-device sync (1 GB). On Open there is no sync. Disabling sync never deletes local data.

## Related

- [Model settings](/docs/model-settings)
- [Attachments and outputs](/docs/feature-attachments)
- [Talk to it instead of typing](/docs/feature-voice)
- [Combining answers](/docs/multi-model-fusion)
- [Sync and devices](/docs/feature-sync)