Isolated
On-premise
Models and data stay on your own servers. No external calls — the highest level of control.
- Local GPUs or your own cluster
- Runs without internet access
- On-site assessment and audit
Security and data
By default, models run in your infrastructure.
Third-party cloud models only with your security team's sign-off. Here is how we access your systems and what goes into the contract — plus a data-flow diagram and answers for your questionnaire.
In short
- We access your systems only with your permission, logged on your side.
- We train models on your data only with your consent.
- On-premise means no external calls: models and data stay on your own servers.
- What you keep — code, weights, documentation or a licence — goes into the contract.
Deployment
Where the models run and how we get access.
By default, models run in your infrastructure — on your own servers, in your own cloud tenant or a mix of both. You choose the setup; your security team signs off on the requirements.
Odissey LAB
Access
only with your permission, logged on your side
Your infrastructure
On-premise, your own cloud tenant or hybrid
- data
- models
- access log
Your security team's sign-off
what data goes out is agreed in advance
Third-party cloud model
only for a pilot or if you have no GPUs
Private cloud
Your cloud provider
We deploy in your own account with your cloud provider. You own the infrastructure and the access.
- Your tenant, your keys
- Scales with the load
- Redundancy and SLA
Hybrid
Mixed setup
Sensitive data is processed locally; heavy training runs on dedicated capacity.
- Separation by data class
- Training separate from inference
- Single access control
Data lifecycle
What happens to your data, step by step.
From access to your systems to what you keep after the project — and who decides what.
01
Access
The target sheet, an annex to the contract, lists what access you provide and by when. The 30-day pilot starts once we have access to the data.
02
Logging
We access your systems only with your permission, logged on your side — it's in the contract.
03
Processing
Models run in your infrastructure. A third-party cloud model only for a pilot or if you have no GPUs, and only with your security team's sign-off. What data goes out is agreed in advance.
04
Training
We train models on your data — our own models included — only with your consent.
05
Measurement
We compute the impact with the formula on the target sheet, using exports from your systems, so you can check the calculation without us.
06
Handover
What you keep — code, fine-tuned model weights, documentation, or work under a licence — depends on the task and goes into the contract before kickoff.
Models and people
A person at every decision point.
Which models our projects use, and which decisions stay with people.
Models
- Open models
- Third-party cloud models
- Our own ML models and ensembles
Third-party cloud models only for a pilot or if you have no GPUs, and only with your security team's sign-off.
Where people decide
- On your side, the process owner and the finance director sign the target sheet; IT and security sign too when the project concerns them.
- After the pilot, you decide whether to continue or stop.
- If a result is disputed, we measure again; if that doesn't settle it, an independent expert decides.
Contact
Questions about security and data
Write to the lab's general inbox:
hello@odisseylab.ruSecurity questionnaire
Answers for your security team.
Copy the answers into your own questionnaire: it's all on this page, with no form and no sign-up.
- 01
Where do the models run?
By default, in your infrastructure: on your own servers, in your own cloud tenant or a mix of both. You choose the setup; your security team signs off on the requirements.
- 02
Will you send our data to a third-party cloud model?
Only if your security team signs off — for a pilot or if you have no GPUs. What data goes out is agreed in advance.
- 03
How do you access our systems?
Only with your permission, logged on your side — it's in the contract. The target sheet lists what access you provide and by when.
- 04
Do you train models on our data?
Only with your consent — and that includes our own models.
- 05
Can the solution run without internet access?
Yes, on-premise: models and data stay on your own servers, with no external calls.
- 06
What do we keep after the project?
It depends on the task and the product: code, fine-tuned model weights, documentation — or work under a licence. It goes into the contract before kickoff.
- 07
How much of our people's time does it take?
On a typical pilot, 10–20 hours a week: data exports, access, reviewing results and security approvals.
- 08
Does our security team sign the target sheet?
Yes, if the project concerns it — alongside the process owner and the finance director. The target sheet is an annex to the contract.