# OpenAI Chat Completions

Send a server-side Chat Completions request with explicit Kastra and upstream credentials.

Updated: 2026-09-22

Canonical: https://docs.kastra.ai/proxy/openai

New to this integration? Follow the [model-proxy quickstart](https://docs.kastra.ai/start/model-proxy) for a complete ALLOW, DENY, and HOLD example.

## Prepare a test request

Set `KASTRA_API_KEY` and `OPENAI_API_KEY` in your server environment. Choose a model your provider account can access. Create a `request.json` file:

```json
{
  "model": "REPLACE_WITH_YOUR_OPENAI_MODEL",
  "messages": [{"role": "user", "content": "Reply with a short greeting."}],
  "stream": false
}
```

Replace the model placeholder before sending. This example produces no tool side effect.

```sh
curl -i https://api.kastra.ai/v1/chat/completions \
  -H "Authorization: Bearer $KASTRA_API_KEY" \
  -H "X-Upstream-API-Key: $OPENAI_API_KEY" \
  -H 'Content-Type: application/json' \
  -H 'X-Jurisdiction: US' \
  --data-binary @request.json
```

Choose the jurisdiction your policy expects; it is policy context, not a data-residency selector.

## Read the response

Inspect `X-Policy-Decision` and, when present, `X-Policy-Pre-Decision` and `X-Policy-Post-Decision`. Decode structured errors even on non-2xx responses. A provider success status alone is not sufficient permission to execute a generated tool.

## Add tools and approvals

Start with a harmless test tool and configure a policy that matches the actual output tool name and arguments. Handle blocked or held output before dispatching the tool. [Output and streaming guide](https://docs.kastra.ai/proxy/streaming), [approval contract](https://docs.kastra.ai/proxy/approvals).

## Endpoint boundary

This guide covers Chat Completions. It does not document `/v1/responses`, assistants, image generation, or arbitrary OpenAI endpoints as Kastra proxy routes. Use the [evaluation API](https://docs.kastra.ai/reference/evaluate) for a custom cooperative authorization integration.