Maintenance: Company-Wide Rollout of Local Models
This use case makes the Cost Control for Running AI Agents and Local AI Operation for Your IT Infrastructure solutions concrete. The scenarios are illustrative and show how DAOS works day to day.
Alongside local models, you can just as easily control access to cloud provider models.
At a glance
The problem
Rolling out new local models today depends on specialist knowledge: is the hardware enough, is there free storage, which old model has to go first? Without that check, rollout stays manual work for a handful of specialists.
The customer
Head of IT
Rolls out new AI models across the entire fleet of machines and owns smooth operation.
The system
DAOS Policy Management
Defines model policies centrally and checks hardware requirements automatically before every rollout.
The goal
Deploy models smoothly on every machine
Without outdated models blocking storage or new models failing on insufficient hardware.
The result
The Policy Engine checks hardware and storage locally before every rollout automatically – models are only installed where they actually run, and outdated versions are removed first.
The path at a glance
Policy defined
Targets selected
Rollout started
Hardware checked
Status at a glance
Illustrative path – the detailed walkthrough is below.
The journey from the customer's side
- 1
Set the rules once
The head of IT sets once which models are allowed, prioritized, or deprecated – including the minimum memory and compute a model needs to run.
- 2
Select target machines
An overview of all machines with names and hardware data shows where the update should land. The head of IT checks a box per machine or user group – whole departments are covered in one step.
- 3
Rollout at the push of a button
One click distributes the new policy to every selected machine. No manual per-device login is needed.
- 4
Hardware checks itself
On every machine, the local Policy Engine checks whether there is enough free storage and whether the hardware is sufficient for the new model before installing it or removing an outdated one.
- 5
Result at a glance
The head of IT sees which machines completed the rollout successfully and where a check failed – including the reason, instead of a silent failure.
Further scenarios
Old models removed automatically
A model marked deprecated is automatically removed from every machine that still has it installed at the next rollout – no separate cleanup needed.
Staged rollout
A new model can be rolled out to a test group first and only extended to the rest of the fleet after it checks out.