Getting Started

Customize an agent

Update the agent name, system prompt, model settings, allowed origins, and server access before you embed or share the agent.

8 sections2 locales

Once the agent exists, the next job is to make it behave like the agent you actually want to ship.

In Emcy today, agent customization is split across the Overview, Configuration, Embed Code, and Settings tabs.

1. Update the core configuration#

  1. Open Agents.
  2. Click the agent you want to edit.
  3. Open the Configuration tab.

This tab gives you the main runtime controls:

  • Name
  • System Prompt
  • Max Tokens
  • Temperature
  • Allowed Origins

After changing anything here, click Save Changes.

2. Write the system prompt carefully#

The System Prompt is the highest-leverage setting on the agent page.

Use it for instructions that should apply across every conversation, such as:

  • the agent's role
  • what success looks like
  • how cautious it should be with destructive actions
  • how it should summarize or format results

Keep the prompt product-specific. If the agent is for support, say that. If it is for internal operations, say that too.

3. Tune Max Tokens and Temperature#

Emcy exposes both controls directly in Configuration.

Use them like this:

  • lower Temperature for more deterministic, operations-style assistants
  • higher Temperature only if you want a more flexible or creative tone
  • raise Max Tokens when the agent needs longer summaries or multi-step answers

If you are unsure, keep temperature conservative and increase only when the agent feels too rigid.

4. Set Allowed Origins before embedding#

If you plan to use the agent through the SDK in your app:

  1. Stay on the Configuration tab.
  2. Fill Allowed Origins with the domains that should be allowed to use the agent.
  3. Separate multiple domains with commas.

Example:

TEXT
https://yourdomain.com, https://app.yourdomain.com

If you leave this empty, the agent is more permissive. For production, it is better to lock it to the exact origins you expect.

5. Adjust which servers the agent can use#

The agent server list lives on the Overview tab.

From there you can:

  • add or remove MCP servers
  • review the tool counts coming from each server
  • save the updated agent-to-server mapping

This is the right place to tighten scope if the agent has access to too many tools.

6. Test it in Playground#

From the agent header, click Playground.

The playground lets you test the agent in two display modes:

  • Inline
  • Floating

Use this before embedding the agent anywhere else. It is the fastest way to catch:

  • missing OAuth configuration
  • overly broad tool access
  • a weak system prompt
  • output that is too long or too short

7. Know what the dashboard does not customize yet#

Today, the agent dashboard focuses on runtime behavior, not full visual widget theming.

That means:

  • use the agent page for name, prompt, tokens, temperature, and allowed origins
  • use SDK props like title, welcomeMessage, placeholder, and mode when you want to shape the embedded chat experience itself

After the agent is configured: