Skip to main content
Version: v2.0

Choose a harness and model

An agent's runtime is three separate settings: the harness, the model, and the credential that serves the model. This page sets all three. For what each one does, see Harnesses and models.

Open the settings​

Go to Sidebar > Agents > <agent name>, set Playground mode to Build, and open the Model & harness section at the top of the Configuration column. Its header summary shows the current harness and model, for example Pi · gpt-5.6-luna. A Connect key pill means no provider credential is set yet.

If you only want to change the model or add a provider key, do it here in the inline panel. To change the harness, or to see which models and hosting options a harness supports, click Detailed configuration and use the Model & harness dialog.

Choose the harness​

In the Model & harness dialog, open the Harness section. It offers Pi and Claude Code. The one in use carries a Current pill.

Each option lists the providers it can reach under PROVIDERS and the hosting options it accepts under HOSTING. Claude Code reaches Anthropic models and accepts hosting through Bedrock and Vertex AI as well as direct and custom. Pi reaches a wider set of providers, with direct and custom hosting. Read the two lists in the dialog before you pick, since they are the current state for your deployment.

Each option also says whether it works with the model you have selected: ✓ supports your model, or ⚠ model not available.

warning

If the harness you pick shows ⚠ model not available, the agent has no working model until you choose one in the Model section below.

Choose the model​

Open the Model section and use the combobox. The list is filtered to the models the selected harness can reach, and choosing a model also sets its provider.

The picker has a Search box and a provider list with the number of models each one offers. Hover a provider to see its models in a second column. If your model comes from somewhere not in the list, use + Add custom provider at the bottom.

Choose the credential​

The credential decides where the model runs. Open Provider credentials, which has two modes.

API key​

Pick the provider tile, or one of the tiles under USE CUSTOM PROVIDER:

TileUse it for
The provider tile, for example OpenAIThe model vendor directly. Add the key and Agenta lists its models
Azure OpenAIModels served through Azure
OpenAI-compatible endpointA proxy or gateway of your own that speaks the OpenAI API

Paste the key into API key * and click Save. The key is encrypted in transit and at rest.

To manage keys for the whole project instead of one agent, go to Sidebar > Settings > LLMs. Each unconfigured provider has a Configure now button, and + OpenAI-compatible endpoint adds an endpoint of your own.

Subscription​

The Subscription mode runs the agent against a Claude or ChatGPT subscription instead of a metered API key. It requires a self-hosted Agenta deployment, and the card is marked Unavailable in the cloud. See Use your own Claude or ChatGPT subscription.

Save and commit​

Click Save in the dialog footer. The change is a draft on the agent configuration.

To publish it, click Commit in the Configuration header. Deployments and triggers run a pinned revision, so commit before an automation should use the new harness or model.