Harness LLM Connector
Use the Harness LLM Connector to run Worker Agents on Claude and GPT models that Harness hosts and manages, with no provider credentials to configure.
The Harness LLM Connector gives your Worker Agents access to models that Harness hosts and manages. You do not need an Anthropic or OpenAI account, API key, or cloud provider connector. Select the connector, choose a model, and run your agent.
The Harness LLM Connector replaces the separate Harness Anthropic and Harness OpenAI connectors with a single connector that serves models from multiple providers.
What you will learn from this topic
By the end of this topic, you will be able to:
Before you begin
Harness AI enabled: Harness AI must be active for your account. Go to Overview for more information.
Managed connector enabled: The Enable Harness Managed LLM Connectors setting under Account Settings > General > Default Settings > Harness AI must be on. Go to Configure Model Connectors for more information.
Worker Agent permissions: If you define an explicit agent permission block, include
ai_llm_gateway: accessso the agent can reach the Harness LLM Gateway.
How the connector works
Name in the UI
Harness LLM
Identifier
harnessLlm (referenced as account.harnessLlm)
Scope
Account
Used by
Worker Agents
Models
Models from Anthropic and OpenAI that Harness hosts and manages
Credentials
None required from you
Editable
No. The connector is view-only.
Harness provisions the connector at the account level. Because it is shared across the account, every project in the account can select it. You choose the model for each Worker Agent, so one connector serves agents that use different models.
Harness adds models to the connector over time. The Model name list in the Worker Agent editor shows the models available to your account today.
Use the connector in a Worker Agent
Go to AI Agents > Worker agents, then open or create a Worker Agent.
For Model provider, select Select model provider. On the Existing tab, select Harness LLM.
For Model name, select a model from the list. The list includes Claude models, such as claude-opus-4-5, and GPT models, such as gpt-5.5.
Save the agent.
If you leave Model name empty, Harness uses a default model that can change over time. Set the model explicitly when you need consistent results between runs.
Use the connector in YAML
Set connector to account.harnessLlm, and set the model with env.MODEL:
To use a GPT model instead, change only the model:
Move from the deprecated connectors
Agents that reference Harness Anthropic (harnessAnthropic) or Harness OpenAI (harnessOpenAI) keep running. To move an agent to the Harness LLM Connector:
Open the Worker Agent and set Model provider to Harness LLM.
Set Model name to the model the agent used before.
Save the agent, then run it again to confirm the result.
Your own Anthropic Model Connector and OpenAI Model Connector are not deprecated. Keep using them when you want agents to run with your own provider credentials, cloud provider connectors, or LiteLLM endpoints.
Troubleshooting
Next steps
Worker Agents: Create and configure agents that use this connector.
Anthropic Model Connector: Use your own Anthropic or Bedrock credentials.
OpenAI Model Connector: Use your own OpenAI or LiteLLM credentials.
Last updated
Was this helpful?