Local AI and LLM servers · Private · On your premises
On-premise LLMs: private AI on your own servers.
In-office AI that keeps your data in the building. For security- and privacy-focused work, Code Squirrel builds, optimises and tunes an AI model that runs on your own hardware, inside your own network. You get an industry-standard, high-quality AI solution, and your prompts, documents and data never leave your premises.
That opens up AI for workflows handling sensitive data that must meet data sovereignty requirements and the Australian Privacy Principles.
Prefer managed? See Squirrel AI, our Australian-hosted AI.
Models running on your own servers
Data sovereignty by design
Designed around the Australian Privacy Principles
Tuned for your workload
Why run an on-premise LLM instead of cloud AI?
Bring the AI to your data
Code Squirrel makes AI viable where it previously wasn't. An on-premise LLM runs on hardware you own, inside your network, so prompts and documents never leave the building. You decide who can use it, what it can see and where every log is kept.
Prefer managed?
If your data can leave your office but must stay in Australia, Squirrel AI, our Australian-hosted AI, gives you frontier models without buying any hardware.
What a private, self-hosted LLM can do for your business
Models & hardware
Open-weight models, sized to your workload.
We don't sell a box off the shelf. Every in-office AI deployment is chosen, sized and tuned for the work it has to do.
Models
Open-weight LLMs
We work with leading open-weight model families, such as Llama, Qwen, Mistral and gpt-oss, and pick the one that fits your task, language and hardware. The best choice changes quickly, so we recommend it at scoping time, not from a brochure.
Hardware
Sized to your users and workload
From a single GPU workstation for a small team to a rack server for heavier, multi-user and agentic work. We size the hardware to how many people use it and what they ask of it, and tell you honestly if what you already own will do.
Tuning
Tuned to your documents
We set up search over your own documents and knowledge base, configure prompts and guardrails for your use cases, and tune performance so answers come back quickly on your hardware.
Integration
Connected to the systems you use
Your on-premise AI can plug into your document management, CRM, inbox and line-of-business systems, the same way our custom AI agents do.
Access
Your users, your controls
Access controls and logs stay on your servers, so you decide who uses it, what it can reach and how long anything is kept.
Support
Looked after by our Brisbane team
Updates, monitoring and model upgrades as better open-weight models are released, from the engineers who installed it.
How it works
How an on-premise AI deployment works.
Six steps, and you sign off on every one.
Step 1
Scope the workload
A free consultation, then an AI readiness assessment ($999 ex. GST) to pin down the use cases, the data involved and the compliance rules it has to meet.
Step 2
Size the hardware
We recommend the model and the hardware to run it, including whether your existing servers can do the job.
Step 3
Install and secure
We install the model and its software inside your network, lock down access and, where needed, run it fully air-gapped.
Step 4
Tune and integrate
We connect it to your documents and systems, tune prompts and guardrails, and test it on real examples from your business.
Step 5
Train your team
Short, practical sessions so your people know what it is good at, what it isn't, and how to get the best out of it.
Step 6
Support and upgrade
Our Brisbane team monitors and maintains it, and upgrades the model as better open-weight options arrive.
Privacy & compliance
Private AI for sensitive data, designed around the Australian Privacy Principles.
When the model runs on your own hardware, the hardest privacy questions answer themselves.
Data never leaves your network
Prompts, documents and answers are processed on your own servers.
Air-gapped options
Once installed, the model can run with no internet connection at all.
No overseas disclosure
Nothing is sent to a provider offshore, so there is no APP 8 cross-border disclosure to manage.
Your logs, your rules
You set what is logged, where it is stored and how long it is kept.
Access you control
Only the people and systems you approve can use it.
Mutual NDA from day one
Every engagement is covered by a mutual NDA before you share anything sensitive.
Supports Privacy Act & APP obligations
No APP 8 cross-border disclosure
Air-gapped deployments available
Mutual NDA on every engagement
Hosted or on-premise?
On-premise LLM or Squirrel AI: which fits your business?
On-premise suits data that can't leave your building. If it only has to stay in Australia, our hosted service is quicker and needs no hardware. For background, read sovereign AI in Australia.
What matters
On-premise LLM
Squirrel AI (hosted)
Where it runs
Your own server, in your office or data room
Our servers in Brisbane (NEXTDC B2, Tier IV), with disaster recovery in Victoria
Does data leave your network?
Never
Yes, to Australian infrastructure only
Upfront hardware
Yes, sized to your users and workload
None
Models
Open-weight models chosen and tuned for your workload
Our blended, tuned frontier system
Pricing
Quoted, based on your hardware and workload
From $149/month ex. GST, or pay-as-you-go
Best for
Air-gapped, highly regulated or high-volume workloads
Teams that want frontier AI with onshore data, without the hardware
FAQ
On-premise LLM questions, answered straight.
What is an on-premise LLM?
A large language model that runs on hardware you own, inside your own network, rather than in an AI provider's cloud. It works much like the AI tools you already know, but your data never leaves your premises.
Why choose an on-premise LLM over cloud AI?
Control. Some data, such as patient records, legal files or government information, can't be sent to an outside provider. An on-premise LLM lets you use AI on that data while keeping it under your own roof and your own rules.
Which AI models can you run on our servers?
We work with leading open-weight model families, such as Llama, Qwen, Mistral and gpt-oss. We recommend the best fit for your task and hardware at scoping time, because the strongest option changes quickly.
What hardware do we need?
It depends on how many people will use it and what they ask of it. A small team can often start on a single GPU workstation; heavier, multi-user or agentic work needs a server. We size it for you, and check whether your existing hardware will do.
Can it run without an internet connection?
Yes. Once installed, the model can run fully air-gapped, with updates and model upgrades applied by our team on a schedule you approve.
How is this different from Squirrel AI?
Squirrel AI is our hosted service: frontier AI on our own servers in NEXTDC's Tier IV B2 Brisbane data centre (with disaster recovery in Victoria), from $149/month ex. GST, with no hardware to buy. An on-premise LLM runs on your own servers, for data that can't leave your building at all.
Will an on-premise LLM fit our ISO 27001 or security framework?
It usually fits more easily than a cloud service, because the model sits inside your network and under your existing controls: your access management, logging, backups and change process. Code Squirrel isn't ISO 27001 certified itself, but we work with clients who are, have helped clients achieve certification, and document the deployment so it slots into your ISMS and risk register.
How much does an on-premise AI deployment cost?
Every deployment is quoted, because the hardware and workload vary so much. Most start with a free consultation and an AI readiness assessment ($999 ex. GST), which gives you a clear recommendation and estimate before you commit.
Do you sign an NDA?
Yes. Every engagement is covered by a mutual NDA from day one, before you share anything sensitive.
Ready to bring your AI home?
Chat to us today for an obligation free discovery session. You’ll hear from an engineer, not a sales rep, and you’ll leave the conversation with a straight answer.
