Browse docs
On this page

Customize your setup

Once your first message gets a reply, it's time to make the assistant your own. This page collects the changes most teams make next, roughly in the order they need them.

Your organization starts with a working setup:

  • A starter agent with general-purpose instructions.
  • An OpenAI connection with GPT-6 Luna, assigned to that agent.
  • An enabled Monthly spend limit of $10 per calendar month, shared by the whole organization.

Identify your real users

The dashboard guide signs every token with the test IDs test-user and test-org, so everyone shares one conversation history. Its examples refuse requests in production until you change them. Before going live, let's read both IDs from your signed-in session:

  1. Set user.id to the signed-in person's stable ID from your database.
  2. Set tenant.id to the customer account, workspace, or organization they belong to. Users with the same tenant share its directory and usage limits.
  3. Return an authentication error when there is no valid session, then remove the example's production check.

Never read these IDs from the request body, because anyone could then chat as anyone else. See Authentication for complete examples, other languages, and switching workspaces.

Shape your agent

Open Agents and choose the starter agent. Rewrite its system prompt to describe your product, its users, and what the assistant should and shouldn't do. You can also turn file attachments and web access on or off, and create more agents for different parts of your app.

See Agents for every option, and The system prompt for writing tips.

Choose your models

The starter OpenAI connection lets you try chat without a provider account. To use your own account, another provider, or a gateway such as OpenRouter, open Models and either replace the starter key or add a provider. Then assign the model to your agent.

See Models.

Control spend

The starter limit refuses new messages once the organization has spent $10 in a calendar month (UTC). Open Usage Limits to raise it, or add per-tenant, per-user, or per-conversation limits that match your pricing plans.

See Usage limits.

Match your brand

Pass a title and theme options to the widget so it looks like part of your app. You can set the color scheme, override theme tokens, or add custom CSS.

See Theming and Configuration.

Work with your app's data

Let the assistant read and change records through functions in your frontend, using the signed-in user's own permissions, and show results as live components in the chat.

See Tools and widgets, or follow the Todos tutorial to build both end to end.

Add more capabilities

  • Attachments let users share images, PDFs, and documents.
  • MCP servers connect the agent to remote tools such as your issue tracker or CRM.
  • Sandboxes let the agent run code and produce files.
  • Directories show a tenant's users and saved conversations in your app.

Before you go live

  1. Replace the test IDs, as described in Identify your real users.
  2. Keep ASTRALBEAM_API_KEY in your server's secrets, never in browser code or version control.
  3. Review the security model before connecting tools that change data.
  4. Check that your usage limits match what you're willing to spend.