AI in Moliri
Where the AI runs
A model can run in four places. The place decides two things: how fast it answers and whether your data leaves the building. For each function, the AI overview in the profile gives the place, with stays in the house or goes to and the provider.
The places
| Place | What runs there | Where the data goes | Who sets it up |
|---|---|---|---|
| In the browser | Speech input, text recognition, field search | Nowhere. The model comes from the internet the first time and stays in the browser afterwards | You, in the profile |
| This PC | Speech input, text recognition and field search, with the desktop app Transcribe by Moliri | Nowhere. Sound, images and search words stay on your computer | You, in the profile and in Transcribe |
| The instance’s server | Transcription, text recognition on the server, search by meaning | To your instance’s server, like any file you upload | Moliri, with the instance |
| AI account | Suggest fields with AI, the AI run under Similar values, dimensions from free text, text recognition by the AI service | Wherever the account points: to a server of your own in the house or to a provider | The administration, in the AI account |
There are also two ways in which you pass data on yourself. An AI assistant that you connect with an access token reads what you are allowed to read; the data goes to that assistant. A copied prompt you paste into a chat window yourself; it goes only there.
For speech input, you can also choose ElevenLabs as the service, or a server of your own. The sound then goes straight to that service, not through Moliri’s server.
Who chooses the place
What runs in the browser and on this PC you set for yourself in the profile . It applies to your account and nobody else’s.
| Profile page | Choice |
|---|---|
| Speech input | Transcribe, in the browser, ElevenLabs or a server of your own |
| Text recognition | The default place: in the browser, Transcribe on this computer, AI service or server |
| Field search | Where the field search model runs, in the browser or in Transcribe |
The default place of text recognition is only where things start: in the OCR import and with Recognise text you can choose another place for each run.
What runs on the server and through the AI account applies to the whole instance. The administration sets up the AI account (status 9). Whether images may go to the AI service is decided with the switch Text recognition by the AI service .
What goes to a model
Moliri sends a model only what the function needs.
| Function | Goes to the model |
|---|---|
| Suggest fields with AI | The texts read from the document, the field names and up to three existing values per field |
| Similar values, AI run | The values of the chosen fields |
| Dimensions from free text | The text of the chosen field, such as sheet 34.5 × 47 cm |
| Text recognition by the AI service | The image |
| Speech input | The sound, while you dictate |
Moliri derives whether an account counts as in the house from its address; how is described in the AI overview . If the model runs on a server of your own, nothing leaves the building.
When a place does not answer
| Function | What happens then |
|---|---|
| Text recognition | If Transcribe does not answer, Moliri starts in the browser. If the server is busy, the note offers the other places |
| Field search | If Transcribe does not answer, the model runs in the browser, provided it is stored there |
| Search by meaning | If the project is not prepared or the service does not answer, the switch searches the full text and says so |
Further reading
- AI overview : place, state and provider of each function
- Models and services : which models run in these places
- Connecting Transcribe to Moliri : one address for three functions