Models that stay in the institution
An administrator can run the small assistant models on a machine the institution controls. The product sends a request only to an address the administrator has configured.
| Where the model runs | How it behaves |
|---|---|
| A machine the institution controls | Receives requests only at the address the administrator configured |
| A cloud CPU service | Serves one model. It checks the model file before it answers, does not start until the administrator has set a key, and listens only on the port the platform assigns |
| A developer machine | The local model service accepts connections only from that machine |
People still see suggestions, not automatic changes. A manager or an agent accepts anything that would change a record.
Night notes
After the organisation's night run, a project page can show a short note about that project, and My insights can show a short note about your timesheet day.
- No note is stored while the organisation has the assistant switched off, or while an emergency stop is on.
- Each note is a suggestion. It does not change the project, the timesheet or anyone's record.
- A support session can read a project note and cannot change it.
Note
If your organisation switches the assistant off or turns on an emergency stop, no new night notes are stored.
Describing a flow
On the automations page, and inside a flow, you can describe what the flow should do in your own words.
Describe the flow
Open the automations page, or open a flow, and write what the flow should do.
Review the suggestion
You see a suggestion and what it would change.
Save a draft
Saving keeps a draft. The flow does not start, and nothing is published.
Publish later yourself
The flow starts only when you publish it yourself later.
A support session can see the box and cannot ask for a suggestion.
FAQ
Only to the address the administrator has configured.
No. Each note is a suggestion. It does not change the project, the timesheet or anyone's record.
No note is stored while the organisation has the assistant switched off, or while an emergency stop is on.
No. Saving keeps a draft. The flow does not start, and nothing is published, until you do that yourself later.
Related pages
Run your own AI model
Install a model server on your own machine or server, start an OpenAI-compatible endpoint and connect it to Akollo.
Running the model in the institution
The model runs on the institution's own machines and questions never reach an outside model. Profiles, sizing, memory and graphics cards.