OpenAI Compatibility

Point the official OpenAI SDK at Ablatic and keep your code. Three things change: the base URL, the key, and the model.

What to change

Keep the openai SDK you already use. Then:

1. Set base_url to https://api.ablatic.ai/v1.
2. Set api_key to your Ablatic key, which looks like sk-ablatic-....
3. Set model to talos.

Everything else stays the same. Auth uses the Authorization: Bearer header that the SDK already sends.

Python

from openai import OpenAI

client = OpenAI(
    base_url="https://api.ablatic.ai/v1",
    api_key="sk-ablatic-...",
)

resp = client.chat.completions.create(
    model="talos",
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)

JavaScript

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.ablatic.ai/v1",
  apiKey: "sk-ablatic-...",
});

const resp = await client.chat.completions.create({
  model: "talos",
  messages: [{ role: "user", content: "Hello" }],
});
console.log(resp.choices[0].message.content);

Streaming

Set stream: true and read tokens over Server-Sent Events. The stream terminates with a data: [DONE] line, the same as OpenAI.

stream = client.chat.completions.create(
    model="talos",
    messages=[{"role": "user", "content": "Write a haiku about Linz"}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

Compatibility notes

response_format is supported on this surface. You can request text, json_object, or json_schema. This is the OpenAI-style surface, so structured outputs work here. On the Anthropic-style surface they do not.

Two norms are worth knowing if you carry options over from another OpenAI-compatible provider:

• Expect n to be 1. Request a single completion per call.
• Very high temperature is treated conservatively. The accepted range is 0 to 2. Keep values modest for predictable behavior.

Vercel AI SDK

The AI SDK's OpenAI provider defaults to the Responses API. Ablatic serves it, so the default path works:

import { createOpenAI } from "@ai-sdk/openai";
import { generateText } from "ai";

const ablatic = createOpenAI({
  baseURL: "https://api.ablatic.ai/v1",
  apiKey: "sk-ablatic-...",
});

const { text } = await generateText({
  model: ablatic("talos-preview"), // -> /v1/responses (stateless)
  prompt: "Hello",
});

The Responses surface is stateless and does not surface the reasoning trace (see Responses). If you want the reasoning trace or the Chat Completions shape, use ablatic.chat("talos-preview"), which targets /v1/chat/completions.

Next

• Function calling
• Chat completions reference
• Migrate in one line