Delivery options
Where your AI runs.
You choose where your AI agents run and where your data stays. We explain the trade-offs plainly and confirm your choice in writing.
Your data stays where you choose: on your own servers, in your own cloud account, or with a contracted model provider. We confirm which in writing.
Where each part lives.
Pick an option to see where the model, your data and the log sit.
Confirmed in writing, before launch
Three ways to run it.
The agents, the approval points and the log work the same way in all three. What changes is where the model runs.
Your own servers
- Where the model runs
- Open-source models, on servers inside your organization.
- Where your data stays
- On your servers, inside your network. The log stays there too.
- Suits
- Organizations whose rules or policies keep data inside their own buildings.
- What it takes
- The most setup of the three: servers sized to the work, space, power, network rules and a named owner on your side. Fine-tuning is possible, only with your written approval.
Your own cloud account
- Where the model runs
- A commercial model, inside your own cloud account.
- Where your data stays
- In your cloud account, under your contract and your controls. The log stays there too.
- Suits
- Organizations that already run on a cloud provider and want the model inside their own account.
- What it takes
- Access to a model service in your account, and your IT team's approval of the setup. Which models are available depends on your provider and region, so we check before we propose.
A contracted model provider
- Where the model runs
- A commercial model, run by a model provider under our contract.
- Where your data stays
- Requests go to the provider under contract and are never used to train shared models. Your stored records and the log stay where we agree in writing.
- Suits
- Work where data may leave your organization under contract, and teams that want to start quickly.
- What it takes
- The least setup of the three. We check the provider's terms against your rules before we start.
We check which local rules apply to your project, and confirm in writing how the setup meets them.
How we choose with you.
- 1
Start from your rules
What your policies, regulators and contracts say about where data may go.
- 2
Look at the work
How much data there is, how sensitive it is, and how quickly answers are needed.
- 3
Compare the options
What each would take, and an estimate of its running cost, side by side.
- 4
Confirm in writing
The chosen option goes into the proposal and the contract, with the model named.
Questions about delivery options.
Which models do you use?
Open-source models on your own servers, and commercial models in your cloud account or with a contracted provider. We name the exact model in your proposal and contract.
Does the setup meet our regulator's rules?
We don't make that claim in general. We check which rules apply to your project and confirm in writing how the chosen setup meets them.
Can we start with one option and move later?
Usually, yes. The agents, approvals and log are built the same way in every option, so a move mainly changes where the model runs. We scope it like any other piece of work.
Will you fine-tune a model on our data?
Your data is never used to train shared models. If fine-tuning helps, we do it only with your written approval, inside your environment.
Running AI on your own servers.
When it makes sense.
It takes more setup than the other delivery options, so we suggest it when at least one of these is true:
- Your rules or policies say the data can't leave your buildings.
- Your contracts with your own clients rule out sending their data to an outside provider.
- You want full control over the model, its version and when it changes.
- The work is steady and large enough that owning the servers costs less over time than paying per request.
What we do.
- 1
Size the servers to the work
From the volume and kind of work, we work out what the servers need before anything is bought.
- 2
Your supplier provides them
Your usual hardware supplier provides and installs the servers to our specification, and we check the setup.
- 3
Set up open-source models
We choose the models that fit the work, install them, and test them on examples from your process.
- 4
Connect them to your systems
Read-only access wherever possible, and a person's approval before any write.
- 5
Monitor them in production
We watch the agents and the servers through remote access you approve, visit when needed, and report what we measured.
What we need from you.
- Space and power
- A place for the servers, with the power and cooling they need.
- Network access rules
- Which systems the servers may reach, and whether and how our team may connect to them.
- A named owner
- One person on your side who owns the servers and approves changes.
- Examples of the work
- Real or anonymised samples, so the models are tested on your work before go-live.
What you own.
The servers
They sit in your building, under your control.
The data and the log
Everything the agents read and write stays on your servers.
The setup
The configuration, the runbook and the access list are handed over to your team.
Fine-tuning happens only with your written approval, inside your environment.
Questions about running on your own servers.
Do we need our own servers?
Only if one of the reasons above applies. Otherwise your own cloud account or a contracted provider is usually simpler, and we'll tell you so.
Which hardware do you use?
We size it to your work and don't tie you to one brand. The specification goes into your proposal.
Can it run without an internet connection?
Yes, if your rules require it. We then agree with you how updates and monitoring happen without a connection.
Who maintains it after go-live?
It is set in each contract: what we look after, what your IT team looks after, and how we reach the servers.
Not sure which option fits?
Tell us about your rules and the work, and we'll walk you through the options for your case.
Talk to usYou decide. We answer for it.