memfree

Request a pilot

A pilot runs memfree on your own hardware with your own workloads, so the numbers you see are yours. Tell us about your fleet below and we reply to every request by email.

How a pilot works

  1. You tell us about your fleet and what you want to see, with the form on this page.
  2. We agree with your engineers what to measure, and on which node.
  3. We install memfree on that node. Your customers' data stays on your machines.
  4. We run your long-context work with memfree and without it, side by side.
  5. You get the numbers: memory per conversation, conversations per card, right answers and speed.

What we'll need from you

  • One GPU node running vLLM, on NVIDIA H100 or H200.
  • A model you serve today.
  • Long texts or conversations to test with, yours or our test sets.
  • An engineer on your side to work with.

Prefer to write?

Email omar@memfree.io with the same details.

Tell us about your fleet

Fields marked are needed to set up a pilot. When you're done, your email app opens with your request written out, ready to send.

You

Your fleet
GPUs you run

For example: Llama 3.3 70B, Qwen, DeepSeek.

The pilot

For example: how many 1M-token conversations fit on one of our H200 nodes.

Nothing is sent until you press send in your email app. See our privacy note.

Your request is in your email app

Send it from there and it reaches us. If your email app didn't open, copy the request and email it to omar@memfree.io.