AI in Moliri
Models and services
Only a handful of models stand behind the AI functions . Some come with Moliri, some are chosen by the administration in the AI account, and some you load yourself into your browser or into Transcribe. The names in the first column are those of the AI overview in the profile; it also shows which model your instance is using at the moment.
Models of the instance
These models apply to everyone who works with the instance.
| In the AI overview | Model | What for |
|---|---|---|
| Chat model | The language model of the AI account : at a provider (OpenAI, Mistral, Anthropic, Google Gemini) or on a server of your own (Ollama, vLLM) | Suggest fields with AI, AI run under Similar values, dimensions from free text |
| Embedding model | A model of the AI account that compares values by their meaning | AI run under Similar values |
| Image model for text recognition | A model of the AI account that understands images | Text recognition by the AI service, Latin handwriting included |
| Text recognition on the server | Tesseract, with Fraktur | Text recognition on the Text tab, in the OCR import and by rules |
| Transcription | The instance’s transcription service | Transcripts of audio and video with speakers |
| Search by meaning | A language model on your own server that reads in the texts of the objects once | Finding objects by meaning |
The chat model, the embedding model and the image model come from the AI account. Without an entry there, Moliri uses the model the server brings, if it has one. The other models run on the instance’s server; texts and images do not leave it.
Models in your browser and on your PC
You set these models up for yourself. They apply in the browser or on the computer where you loaded them.
| In the AI overview | Model | What for |
|---|---|---|
| Speech input | Whisper in the browser, Transcribe’s speech recognition in Transcribe, Scribe v2 Realtime at ElevenLabs, or a Whisper model on a server of your own | Speech input |
| Text recognition | tesseract.js in the browser, Transcribe’s text recognition in Transcribe | Recognise text and the OCR import, when you choose this place |
| Field search | EmbeddingGemma, in the browser or in Transcribe | Field search and the suggestion by column header in the Excel import |
Moliri downloads a model for the browser once; the AI overview gives its size. After that it stays in the browser and does not have to be loaded again.
Transcribe by Moliri
Transcribe by Moliri is a free desktop app for Windows. It transcribes recordings on your computer, on its own and without an account. Connected to Moliri, it takes on speech input, text recognition and field search. It loads its models once for all browsers on the computer, and what it reads does not leave the computer. You set this up under Connecting Transcribe to Moliri .
Assistants and prompts
| In the AI overview | What it is | What for |
|---|---|---|
| MCP access | An AI assistant of your choice, such as Claude, Cursor or VS Code, connected via the Model Context Protocol | Reads with your rights and submits AI suggestions . Set up under Access tokens |
| Prompts to copy | Ready-made requests that you paste into a language model of your choice | Formulas, Word templates, new fields, value lists, event rules, mapping to standards, see What the AI can do |
Further reading
- Where the AI runs : what runs in each place and where data goes
- AI account : setting up the chat model, embedding model and image model