What it is
Programmatic access to Google's neural networks (the Gemini model family) โ your code can "ask" the AI something and get an answer automatically.
How we use it
The AI assistant in the chat on this website runs on it โ it answers visitors' questions and gathers the basics about their task before a call with you.
Where it helps a business
- Voice, receipt photos or complex text need to be parsed automatically, with no separate recognition software.
- The website needs a consultant that answers from your price list instead of making things up.
- Staff need a first line of support that closes simple questions before a ticket is filed.
How we use it
- Audio in one request. In the voice transcription bot the audio and an instruction go to the multimodal Gemini model in a single request, with no separate speech recognition step. It works through the google-genai library, and the main and backup models are configurable.
- Receipts and statements. The finance assistant sends receipt photos, voice and complex text to Gemini; keys and ordered model lists live in
.env. - First line of support. The IT help desk has an optional Gemini assistant, off by default.
- A consultant on the site. The AI consultant in this website's chat also runs on Gemini and gathers initial information about a task.
Common problems
- Quotas. Keys work as a pool: an exhausted key goes on cooldown (60 minutes by default), the request goes through the next one, and the admin gets a notification; the admin panel shows the state of each key.
- The model was renamed. On a 404 the bot switches to the backup model by itself.
- Data goes to Google. When a complex parse is needed, the photo and text are sent to Google. Without keys only the built-in rules work.
When you do not need it
If rules and keywords are enough, you do not need a model. And if the data must not go to an external service, a cloud API does not fit.