An AI API that costs pennies.

For chatbots, Discord bots, side projects and scripts. It speaks the OpenAI API, so you change two lines and keep your code. Prepaid, no subscription, and it never charges past zero.

5pper 1M tokens in
50pper 1M tokens out
£1≈ 20M input tokens

Use the same email to ask for your key and to pay. Top-ups are matched to your key by email, so if they differ the credit won't be added automatically.

your_bot.py — the only change
from openai import OpenAI

client = OpenAI(
    base_url="https://api.bytedev.co.uk/v1",
    api_key="byte_…",
)
r = client.chat.completions.create(
    model="byte-chat",
    messages=[{"role": "user", "content": "hello"}],
)

What it costs

Your balance only drops when you make requests. No monthly fee, no expiry, no overspend.

WhatModel namePrice
Chatbyte-chat5p / 50p per million tokens in / out
Imagesbyte-image0.5p per image, on request
Video—Not offered right now

A chatbot resends the conversation every turn, so input tokens are most of what it uses. That's why input is the cheap side: a bot that costs hundreds a month on a frontier model often runs on a fraction of that here.

Getting set up

If your tool speaks the OpenAI API — most do — it already works: the OpenAI SDKs, Continue, Aider, OpenWebUI, LangChain.

  1. 01
    Ask for a key. Tell me in a line what you're building and I'll set you up.
  2. 02
    Add credit. Pay with the same email you asked with — that's how credit finds your key.
  3. 03
    Change the base URL and key. Chat completions and streaming, the usual shape. A chat key can't touch image gen: it only has what you paid for.
  4. 04
    Top up when it runs low. At £0 requests stop with a clear message. It never overspends.

Ask for a key

Use the same email when you pay, so the credit lands on your key.

Prefer email? [email protected]. By asking for a key you agree to the terms and privacy policy.

Check your balance

Only used to look up the balance. Nothing is stored.
Enter a key above.

Straight answers

Is it as reliable as OpenAI?

No, and I won't pretend otherwise. It runs on my own hardware as a small side project, best-effort, with no uptime guarantee. It's cheap precisely because it isn't an enterprise datacenter. Fine for hobby projects, bots and side-apps; not for anything mission-critical.

Which model is it?

A fast open Mixture-of-Experts model — think "better than GPT-4o-mini" for everyday chat, not a frontier coding agent. Great for chatbots, bots, assistants and everyday app features.

What happens when my credit runs out?

Requests stop with a clear "out of credit" message — no overspend, no auto-charge — and you top up when you want to keep going.

What about my data?

Requests pass through to the model to generate a reply. I keep only minimal operational logs and never sell or share anything. Don't send secrets.