Lume AutonomaService · Private AI models
Private AI models for US and Canadian businesses
A private AI model is a language model that runs on your own servers, so your data is never sent to a third-party AI. Lume Autonoma fine-tunes open-source models on your documents and conversations with LoRA or QLoRA, then quantizes them to 8-bit or 4-bit so they are fast and cheap to serve.
- 7B model, 16-bit
- ≈ 14 GB
- Same model, 4-bit
- ≈ 4 GB
- Your data
- Stays on your servers
What we build
- Fine-tuning with LoRA and QLoRA
- Quantization to 8-bit and 4-bit (GGUF, AWQ, GPTQ)
- Search and answers over your own documents (RAG)
- On-premise or private-cloud deployment
Who it is for
Businesses whose documents or customer data must stay under their control, such as law, accounting and consulting firms, and any team that wants an assistant that answers the way the business does.
For businesses in the US and Canada
Your model runs on your own servers or in a private cloud, in the region you choose, including the United States or Canada, so your data stays where your policies and your clients need it.
How we work
- ScanA discovery call. We map where your team's time and money go.
- BlueprintA written scope, timeline and fixed quote before any work starts.
- BuildYou see working progress every week and test it as we go.
- DeployWe launch, train your team and hand over the keys.
- EvolveWe monitor, fix and improve, so the system keeps getting better.
What it costs
Every project has a fixed price, agreed in writing after a discovery call, so there are no hourly surprises. You pay on one of our two payment plans, Elite or Normal, set out in our terms. See the payment plans.
After launch, the monthly plan is US$250 a month and covers security updates, monitoring and fixes. You can cancel it at any time.
Anything that runs in accounts in your name, such as phone lines, AI model usage or a hosting provider, is billed by that provider at its own rates.
Questions
How is a private model different from a public AI service?
A public AI service runs on someone else’s servers and receives whatever you send it. A private model runs on your own servers or private cloud, is trained on your documents, and your data stays with you.
How big a server does it need?
Less than you might think. Quantization shrinks a model: a 7-billion-parameter model takes about 14 GB at 16-bit and about 4 GB at 4-bit.
Can it answer questions about our own documents?
Yes. We set up search and answers over your own documents (RAG), so the answers come from your files, on your servers.
What is fine-tuning?
Fine-tuning teaches an open-source model your terminology, your documents and the way your business answers. We use LoRA and QLoRA, which train a small set of extra weights instead of the whole model.
How much does a private AI model cost?
Each model project has a fixed price, quoted in writing after a discovery call. After launch, the monthly plan is US$250 a month and covers security updates, monitoring and fixes.
Who owns the trained model?
Once the project is paid in full, you own the weights of the model we trained on your data. The open-source base model stays under its own licence.
Talk to us
Send us a message and a person from the team will reply, usually the same day.