Pay as you go
Buy credit, then each request is paid from it by the tokens it uses. Prices are per million tokens. Input is what you send (your prompt and chat history); output is what the model writes.
Standard pool
Apple Silicon Macs running popular open models (Llama, Qwen, Gemma, Mistral and more). Great for chat, coding help, summaries and embeddings.
$10 of credit ≈ 28.6 million tokens (at a typical 3:1 input:output mix)
Credit
Credit never expires. Pay by card, Bitcoin or bank transfer; sales tax is added at checkout.
Monthly plans
Prefer a fixed monthly bill? Plans are the best value: each includes more usage than the same money buys as credit. Anything past a plan's monthly cap comes from your credit, so your apps keep running if you have some.
Standard pool
Starter
$25/month
Includes $35 of usage each month (≈ 100 million tokens)
40% more usage than credit
For a team app or a busy chatbot.
Standard pool
Pro
$100/month
Includes $150 of usage each month (≈ 429 million tokens)
50% more usage than credit
For production apps with steady traffic.
The included usage resets on the 1st; unused usage doesn't carry over. Plus applicable sales tax; cancel any time in your account (it runs to the end of the month).
Need a lot more? Get your own Mac.
A dedicated Apple Silicon Mac in our datacenter runs your models with no per-token charges and no sharing: a fixed monthly price, and your keys point straight at it.
How it works
How do I start?
Sign in to your account, open AI API, create a key and point any OpenAI-compatible app at https://datagram.net/api/v1. That's it: same code as OpenAI, different base URL and key.
What happens when my credit runs out?
Requests stop with a clear "credit has run out" error until you add more. We email you when you're running low, so you can top up first. You're never billed more than you prepaid.
Does credit expire?
No. Credit you buy (and the welcome credit) stays in your account until you use it. Only a monthly plan's included usage resets each month.
I have a plan and credit. Which is used first?
Your plan's included usage is used first, for keys on that plan's pool. After that, or for keys on another pool, usage comes from your credit.
What is a token?
A token is a piece of a word, about four characters of English text. "Hello, how are you?" is about 6 tokens. Your app's request and the model's answer are both counted, and every answer shows its token counts.
Am I charged for errors?
No. Only answered requests use credit. Refused requests (rate limits, unknown models) and failures are free.
Is my data stored?
No. Prompts and answers pass straight through to the Mac and back. We keep only counts (requests, tokens, timings) for billing. Models run on our own hardware in our datacenter, not a third-party cloud.