> This is a page from the ElevenLabs documentation. For a complete page index, fetch https://elevenlabs.io/docs/llms.txt. For the full documentation in a single file, fetch https://elevenlabs.io/docs/llms-full.txt.

# Customizing ElevenAgents Architect

## Overview

ElevenAgents Architect's own instructions are fixed, but you control the context it brings to every conversation. These documents shape how it works:

| Document                                          | Scope                 | Who it applies to                |
| ------------------------------------------------- | --------------------- | -------------------------------- |
| [Agent context](#agent-context) (`agents.md`)     | One agent             | Everyone who builds that agent   |
| [Your preferences](#your-preferences) (`user.md`) | Every agent you build | Only you                         |
| [Prompt library](#prompt-library)                 | Saved prompts         | You, or teammates you share with |

To open them, go to the agent's **Settings** > **Architect** tab, or to **Workspace Settings** > **Architect**. You can also select **Customization** on the Architect tab home page. The workspace page also lists the agent context for each of your agents.

![Customization on the Architect tab](/docs/_fern-img/b374ed85570a53cde1f0617a3e537af3f7a4c92c83c02d63eb59f528d2668c5b.webp)

> **Note**
>
> These documents only guide how Architect helps you build. They don't change the agent's runtime
> behavior or the system prompt it uses with callers.

![Architect settings for an agent](/docs/_fern-img/d12588d007d18de1de31d047da07e5da57cce7f9fca6203becc7449c3ab1e809.webp)

## Defining the standards ElevenAgents Architect follows

There is no separate rules engine. Today, you define the standards Architect follows in three places:

* **Agent context** for standards specific to one agent, such as "Every procedure change needs a simulation test" or "Never change the voice without asking."
* **Your preferences** for how you personally like to work, such as "Always use Plan mode for prompt rewrites" or "Explain changes as a bulleted diff."
* **The agent itself.** Architect reads the agent's system prompt, procedures, and guardrails before changing them, and follows the conventions it finds there.

There is no workspace-wide rules document yet. To apply the same standard to several agents, add it to each agent's context, or save a shared prompt that states it.

## Agent context

Every agent has an `agents.md` document that Architect receives at the start of every conversation about that agent. It's shared with everyone who builds the agent. Use it for the agent's purpose, tone, domain, and the conventions your team follows, so nobody has to repeat that background.

You can write the document yourself, start from a template with **Purpose**, **Tone and voice**, **Conventions**, and **Testing** sections, or select **Set up with Architect**. You can also type `/start` in a conversation inside the agent. Architect asks a few questions and writes the document for you. Architect can update it later when you ask, for example "Add to agents.md that refunds over \$500 always go to a human."

When a conversation uses agent context, a **Shared agents.md context** label appears above the composer.

The document can be up to 100,000 characters. Above 10,000 characters the editor shows a warning, because a shorter, focused document tends to work better.

## Your preferences

Your preferences document (`user.md`) is personal to you and applies across every agent you build. Use it for standing instructions about how you like to work, such as how much detail you want in explanations, or when Architect should ask before proceeding. It has the same 100,000-character limit as agent context.

## Prompt library

Save prompts you use often and reuse them instead of retyping them. Prompts are private by default. Select **Share** to share a prompt with workspace members, so your team starts common tasks such as reviewing triage tickets or auditing guardrails from the same playbook. The library shows **Your prompts** and **Shared with you** separately.

To use a saved prompt:

* In the composer, select **+** > **Use a saved prompt**.
* In the prompt library, select **Run** to start a new conversation with that prompt.

Each prompt can be up to 20,000 characters.

> ElevenLabs provides APIs and SDKs for text to speech, voice cloning, speech to text, sound effects, voice isolator, voice changer, and conversational AI agents. Build voice-enabled applications with lifelike audio generation.