Configuration guide.

Quick setup

1. 01

   Create your API key

   Open your console, name a key, and copy it once.

   Get an API key ↗
2. 02

   Set two values

   Use your prixis key and this base URL:

   https://prixis.dev/v1
3. 03

   Choose your model

   The default is GPT-6 Astra. Check the required available balance before your first request.

OPENAI_BASE_URL=https://prixis.dev/v1
OPENAI_API_KEY=your_prixis_api_key

An .env file must be loaded by your application. For a terminal session, use the commands below.

Environment variables

Replace the placeholder with the key you copied from prixis. Set these values in the process that runs your app. Keep the key on your server.

export OPENAI_BASE_URL="https://prixis.dev/v1"
export OPENAI_API_KEY="your_prixis_api_key"

After changing a deployment secret, restart or redeploy the app so its process receives the new value.

Your first request

Use a unique request ID for each new generation. Keep the same ID and request body when retrying that generation.

GPT-6 Astra · Required available balance: $7.48

Install the SDK with pip install openai.

from openai import OpenAI
import os
import uuid

client = OpenAI(
    api_key=os.environ["OPENAI_API_KEY"],
    base_url=os.environ["OPENAI_BASE_URL"],
    max_retries=0,
)

request_id = str(uuid.uuid4())
reply = client.chat.completions.create(
    model="gpt-6-astra",
    messages=[{"role": "user", "content": "Explain this in one line."}],
    extra_headers={"Idempotency-Key": request_id},
)
print(reply.choices[0].message.content)

Check the response, then open your console to see the actual input, cached input, output tokens and final charge.

Models & reasoning

The exact model ID selects the model for a request. Choose low, medium or high with reasoning_effort; leaving it out uses medium.

USD per million tokens. Input usage may include service instructions. Reasoning tokens count toward output usage.

Retries & billing

The Idempotency-Key belongs to one generation, including its retries. Reusing it with changed messages or a changed model returns a conflict. A completed request replays its saved result and does not charge again.

While an execution remains uncertain, its hold remains pending. Check the request in your console before sending a replacement. Unused held funds return to your available balance after settlement or a confirmed failure.

Streaming keeps the connection open while the request runs, then sends the completed answer as SSE. It does not deliver tokens incrementally.

Client compatibility

Current Codex custom providers require the Responses protocol. A base URL change alone cannot make a Chat Completions-only service work with Codex. Check the official provider configuration reference before using Codex-specific providers.

Codex advanced settings

These settings apply to Codex itself. They do not add endpoint support to prixis. Use the official reference for the complete configuration options.

Troubleshooting