Deployment
Your company already owns the compute.
An engineer from our team installs and configures PeerLLM across the machines you already have — workstations, servers, on-premise or in your own cloud. No new hardware, no data leaving your network, no per-token bill.
Before you read further
This is engineer-led work, and we take on a limited number of them.
Every deployment is done by our own engineers, in your environment, alongside your team. That is deliberate — it is why they work — but it means we can only run two of them per quarter.
So it is worth knowing early whether this is a fit. It usually is when:
- You have machines with real compute sitting idle — workstations with GPUs, servers with headroom.
- Sending company data to an external model vendor is difficult, slow to approve, or simply off the table.
- Someone in the organization is accountable for AI actually being adopted, not just procured.
- You would rather own the stack than rent it, and you understand that owning it means operating it.
It is usually not a fit if you are looking for a hosted API with no infrastructure of your own. That is a real need, andthe public PeerLLM networkserves it directly, without us.
The first question every IT lead asks
What actually gets installed on a machine?
One of two things, depending on the machine: the PeerLLM Electron application, for workstations where someone wants a visible app they can see and control, or the PeerLLM CLI host software, for servers and for managed rollouts where a background service is more appropriate.
Both register the machine as a node on your own private network — one you run, not one we run. The orchestrator that routes work to those nodes is deployed inside your infrastructure too, on-premise or in your own cloud account.
Nothing routes outside your network unless you configure it to.If you choose to allow overflow when your own capacity is exhausted, you choose the target as well — your own cloud models, an AI provider you already have a contract with, or the public PeerLLM network. Never by default.
Your security team will have more questions than that, and they should. We would rather walk through them on a call than pre-empt them with reassurance here.
What the engagement looks like
Shaped around your environment rather than a fixed script, but this is the usual sequence.
- 01
Environment review
What hardware exists, what it is doing today, how machines are managed, and what your security review will require. We size the deployment from what is actually there rather than from a questionnaire.
- 02
Orchestrator install
The PeerLLM orchestrator goes into your infrastructure — on-premise, your VPC, or air-gapped — and is wired into your identity provider and monitoring.
- 03
Node enrolment
Machines join the network, starting with a small pilot group so you can see real behaviour on real hardware before it reaches anyone's desk.
- 04
Models and policy
Which open-weight models you serve, who can reach them, what gets logged, and what happens when demand exceeds the capacity you have.
- 05
Validation and handover
Load testing against your real usage, documentation written for your team, and a working session so the people who will operate it have actually operated it.
What you own afterwards: the orchestrator, the nodes, the data, and the ability to run all of it without us. That is the point of self-hosting, and it would be a strange thing to sell you and then quietly undermine.
Cost
Two numbers, and they are separate.
| What | How it is charged | Paid to |
|---|---|---|
| PeerLLM licence | annual, by node tier — 5, 15 and up | the platform |
| This deployment | hourly at $300, estimated in the proposal | Best Bytes |
The licence is recurring and the deployment is one-time — the same shape as an annual software subscription plus a professional-services engagement to implement it. A node is a unit of compute: a PC, a workstation, anything with a GPU, CPU and memory. A machine with two GPUs is one node.
Licence tiers, terms and the security-patch policy are published onpeerllm.com, since they belong to the platform rather than to us.