Skip to main content
Every managed agent run, including Signal, needs a model behind its harness. You have three options: To see which company processes agent data under each option, see Where your data goes. If you don’t pick a provider, runs use the Arize-managed model. Free and Pro accounts always run Signal on the Arize-managed model.

​
Use the Arize-managed model

Nothing to set up. Pick Arize-managed in the AI Provider picker, or leave the provider empty. Arize pays for the model and applies per-tier usage limits. The Arize-managed model isn’t available in the CA region or on AWS-hosted deployments, so you need your own provider there.

​
Bring your own provider key

Use your own key when you want model traffic covered by your own agreement with the provider, for example a zero-data-retention agreement with Anthropic or OpenAI.
1

Add the AI provider

Under Settings → Account Settings → AI Providers, add a provider the harness supports. See Connect your harness for the provider each harness supports.
2

Allow it in agents

Tick Managed Agents on the provider’s row in the AI Providers table. You can also do this from the AI Provider picker with Enable Key.
3

Select it

Choose the provider on an agent config, during agent setup, under Configure Signal, or as the space default under Space Settings.

​
Connect your own gateway

If your company sends all model traffic through its own AI gateway, connect the gateway as a Custom Model Endpoint and use it for managed agents like any other provider. Requests go to your gateway’s URL, and your gateway decides which model serves them and what is logged. Any endpoint that speaks the OpenAI or Anthropic API format works, for example LiteLLM, an internal proxy, or a self-hosted model server. Arize translates Claude Code’s requests when the endpoint uses the OpenAI format.
1

Add a Custom Model Endpoint

Under Settings → Account Settings → AI Providers, add a Custom Model Endpoint. Enter the base URL, choose an authentication method, add any extra headers, and list the model names your gateway serves. See Custom Model Endpoint for each field.
2

Test it

Use Test integration to confirm that Arize can reach the gateway and get a response.
3

Allow it in agents and select it

Tick Managed Agents on the row, then choose the endpoint on an agent config, under Configure Signal, or as the space default.
Claude Code can run on Anthropic, OpenAI, Azure OpenAI, and compatible custom endpoints. Codex needs an OpenAI provider, OpenCode needs Anthropic, and Cursor needs a Cursor key. The provider you pick must match the harness.

​
Where your data goes

While an agent works, it sends the model everything it reads. That includes project traces, your instructions, issue context, and data from any connected repo, Datadog, or Dynatrace skill. Arize doesn’t redact this content before it reaches the model, so the provider you choose is what protects it. Some details hold for every option:
  • Arize’s gateway relays requests and keeps no copy of them. Requests pass through Arize’s agent gateway, which adds your credential and forwards each one to the provider. Request and response bodies aren’t written to Arize’s logs.
  • Your credentials never enter the sandbox. The sandbox holds a placeholder. Arize adds your key or OAuth token to each request as it leaves.
  • Arize keeps run history in your account. Each run’s messages, final answers, and tool activity stay in your Arize account so you can review the run, subject to managed agent permissions.
If your security policy doesn’t allow agent data to go to a provider you don’t contract with, use your own provider key or gateway. Contact Arize to turn off Arize-managed models for your account, so nobody can start a run on them.