> ## Documentation Index
> Fetch the complete documentation index at: https://agent-compass.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# OpenAI Chat Completions

> Connect AgentCompass to OpenAI-compatible Chat Completions endpoints.

Use the `openai-chat` protocol for endpoints compatible with the OpenAI Chat Completions request shape, commonly exposed at `/v1/chat/completions`.

## Configure the Protocol

```bash theme={"system"}
export MODEL_NAME=""
export MODEL_BASE_URL="https://your-endpoint.example/v1"
export MODEL_API_KEY="sk-..."

agentcompass run <benchmark> <harness> "$MODEL_NAME" \
  --env <environment> \
  --model-base-url "$MODEL_BASE_URL" \
  --model-api-key "$MODEL_API_KEY" \
  --model-api-protocol openai-chat \
  --model-params '{"temperature":0}'
```

The model id is the third positional argument. AgentCompass sends messages through the configured harness and forwards supported `--model-params` values to the endpoint.

## Common Parameters

| Parameter     | Purpose                                                               |
| ------------- | --------------------------------------------------------------------- |
| `temperature` | Control sampling randomness when supported by the endpoint.           |
| `max_tokens`  | Limit generated completion tokens.                                    |
| `timeout`     | Set the request timeout used by AgentCompass direct protocol clients. |
| `max_retries` | Set SDK-level retries for AgentCompass direct protocol clients.       |

Harnesses may add, transform, or restrict model parameters. Check the selected [Harness](/en/user_guide/modules/harnesses/overview) page before copying a configuration between agent implementations.

## When to Use It

Choose `openai-chat` when:

* the provider documents Chat Completions compatibility;
* the selected harness supports `openai-chat`;
* the evaluation does not require Responses-specific state or request fields.

For stateful Responses endpoints, use [OpenAI Responses](/en/user_guide/modules/models/openai_responses). For Anthropic-compatible endpoints, use [Anthropic Messages](/en/user_guide/modules/models/anthropic_messages).