Skip to main content

Get started with Infracost AI

Connect a source of AI spend, check the first import, then use the dashboard to see where the money goes. You can also build what-if scenarios to explore open-weight models, investigate model swaps, routing changes and environment fixes from your own billing and usage data.

Before you begin​

Contact support@infracost.io to enable Infracost AI for your organization. You can then sign in and select your organization.

To add and manage integrations, you must be an organization admin or owner. You also need access to the billing credentials or telemetry settings for the source you want to connect.

For a billing integration, you usually need a provider API key or token that can read cost and usage reports, or read access to your cloud billing data. Some sources also accept a spend CSV or subscription fees from your invoice. For telemetry, you need to set up the coding agent or application that will send usage data. The integration guide lists what each source needs.

1. Connect an integration​

Open Infracost AI and go to Integrations > Add integration.

Choose a source and give the connection a recognizable name, such as Production Bedrock or Engineering Claude Code. Use separate connections when you need to manage different accounts or credentials independently.

  • Billing and subscriptions: Enter the required credentials, optionally use Test credential, then add the integration. An Anthropic spend CSV or subscription fees need no credentials. For AWS Bedrock, you can select an existing AWS connection from Infracost Cloud instead of entering credentials again.
  • AI gateways: Enter your LiteLLM proxy URL and key, or your Kong Konnect token, then add the integration.
  • Telemetry: Select your coding agent or OpenTelemetry. For OpenTelemetry, choose the Format your application sends. Add the integration and copy the connection token shown on the result screen. Open Show setup instructions and apply the generated settings to your client.
note

Connection tokens are shown only once. Copy the token when it is created. If you lose it, create another token in the integration's settings.

2. Verify that data is arriving​

For a billing integration that uses credentials, or a gateway integration, the first import starts automatically when you add it. Check Import status on the result screen or the integration page before you continue. Use Sync now to refresh it later. The provider determines how much past data is available.

For telemetry, restart the configured client and generate some usage. Return to the integration and check that it has received data. Telemetry starts collecting from the configured client; it does not import old sessions.

3. Explore your spend​

Open Explore, select the integration in Source, and choose a date range that includes the imported data.

Explore opens on the cost mode that has spend, and each mode shows its total. Use Billed to see provider charges and Notional to see measured usage from telemetry and gateways. Add Model or Developer as a breakdown to see where the spending comes from. The details you can see depend on what the source reports.

If the chart is empty, check the integration's latest activity, clear any extra filters, and confirm that the selected cost mode matches the data you connected.

4. Add more sources​

Connect your other providers and add telemetry where you need developer or application detail. As you add integrations, review cost overlaps to see which connections track the same activity.

Copy the Explore URL to share a useful view with another member of your organization.