You run your own AI
You host the model yourself, on your own or rented GPUs, and the GPU bill is yours.
- Support bots and assistants
- Document and knowledge chat
- Coding assistants and in-house copilots
AI chat costs · Bree Developers
Your GPUs keepre-reading chats.
When someone comes back to a long AI chat, most set-ups read the whole conversation again from the start. That is GPU time you pay for, and a wait your users feel. We cut it.
In our test17 s → 0.12 sto pick a 22,000-word chat back up
Your numbersWhat could re-reading be costing you each month?
AI chat costs
The hidden cost
A returning chat is read again. In full, every time.
A person pauses a long AI chat, then comes back. In most set-ups what the model had worked out is gone by then, so the GPU reads the whole conversation again before it can answer. You pay for that time, and the person waits for it.
- Long chats cost the mostIn our test, a chat 3.7 times as long took 6.5 times as long to read.
- Busy GPUs lose capacityEvery second spent re-reading is a second not spent answering someone else.
- The first answer is the slowestAfter a break, the person waits longest at the moment they are most likely to give up.
In our testA person comes back to a 22,000-word chat
Today · the whole chat is read again17.1 s
With Bree · the chat is picked up where it stopped0.12 s
about145 timesquicker to be ready to answer
Measured on a 70-billion-parameter AI model running on two NVIDIA H100 GPUs. The bars play at a quarter of real speed. In the same test, 11 of 11 questions about picked-up chats were answered correctly. Your own set-up will differ, which is what the audit is for.
Your numbers
What could this be on your bill?
Put in roughly how you run today. This is an estimate built from our test. The audit replaces it with your real figures.
How it works
Measure first. Change only what pays.
- 1
Share your usage logs
When each chat was active and how long it was. We do not need the conversations themselves.
- 2
We measure the waste
You get one clear number: how much GPU time goes on re-reading, and what that costs each month.
- 3
Pilot on part of your traffic
We run it on a slice of your chats and compare the bill and the waiting time side by side.
- 4
Pay from what is saved
If it saves you money, the fee is agreed from the saving we have measured with you. If it does not, you have lost nothing.
Who it is for
If you run the GPUs, you pay for the waste.
Your chats are long
The longer the chat, the more it costs to read again, and the more there is to save.
- Documents pasted into the chat
- Long-running agent sessions
- Customers who return after a break
You pay per message
If a provider charges you for each message, they carry the GPU bill, so there is nothing for us to cut. We would rather say so now.
- Not sure which one you are? The audit will tell you.
Get the audit
Find out what it costs you.
Tell us roughly how you run your AI today: what it is, what it runs on, and how long the chats get. We will come back with a plan for measuring the waste.
Ask for Brian Mokonyane · Johannesburg, Gauteng