31 July 2026 · 5 min read

Kortexio is now OpenAI-compatible — for real

Kortexio started as a memory proxy with an Ollama-shaped chat contract at /api/chat. That worked. It also created an awkward gap: marketing said “OpenAI-compatible,” but pointing the official OpenAI SDK at Kortexio did not work without adapters.

That gap is closed.

What changed

The public API now exposes a true OpenAI-compatible surface under /v1:

POST /v1/chat/completions — sync JSON and SSE streaming (data: … / [DONE])
GET /v1/models and GET /v1/models/{id} — the BYOK model configured for your App
POST /v1/embeddings — pass-through to your BYOK provider

/api/chat remains available and unchanged for Ollama-compatible clients. Both paths share the same API key, rate limits, session memory, and agentic loop.

Drop-in with the OpenAI SDK

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.KORTEXIO_API_KEY,
  baseURL: "https://api.kortexio.io/v1",
});

const completion = await client.chat.completions.create({
  model: "gpt-4o-mini",
  user: "user-42", // or send X-User-Id
  messages: [{ role: "user", content: "What did we decide last time?" }],
});

Kortexio injects compiled session memory before your BYOK model runs. Your app still speaks chat completions — memory is ambient.

Identity: header or `user`

OpenAI clients do not send X-User-Id. On /v1/chat/completions you can identify the end user with either:

the X-User-Id header, or
the standard OpenAI user field in the request body

One of them is required so memory stays scoped per person. Ollama /api/chat still requires X-User-Id.

Embeddings are a proxy, not our memory engine

/v1/embeddings forwards to the LLM provider you configured for the App. It does not feed Kortexio session memory. That is intentional: conversational memory stays a structured wiki; embeddings stay available for your RAG over documents, catalogs, or search indices.

If your BYOK provider does not expose embeddings (Anthropic is the usual example), the upstream error is returned as-is.

Dashboard format picker

In the user dashboard (App overview and Getting started) you can toggle OpenAI vs Ollama. The preference only changes the URL and curl snippet shown — both APIs stay live. The choice is remembered in localStorage.

What stayed the same

Bring Your Own Key — Kortexio never bills your model usage
Session memory and Global Wiki
Agentic tools / MCP when enabled
EU-hosted infrastructure defaults

If you already integrated via /api/chat, keep shipping. If you want LiteLLM, Cursor-style OpenAI bases, or the official SDK — set baseURL to https://api.kortexio.io/v1 and go.

← All posts Documentation