Lesson 8 · AI inside the product
Explain which AI outcome your customers need
DeepSeek develops projects inside Studio. When customers need AI inside their website or bot, Studio connects Gemma through Google AI Studio.
- Windows
- Linux AppImage
- Linux .deb
- macOS coming soon
Windows 10 or later or Ubuntu 22.04 or later (64-bit). macOS coming soon. About 1 GB of free disk space.
Windows may show a SmartScreen warning: click “More info”, then “Run anyway”.
Updated: . Checked in Studio 0.11.4.
1. Describe the customer outcome
For example: "Add an assistant that answers from my documents and understands photos, audio and video." Do not request DeepSeek for the customer-facing product.
2. Add permitted sources
Add text, files or links the assistant may use. Emails, leads and external pages remain data, never trusted agent instructions.
3. Test each mode
Send text, an image, short audio and video. For RAG, ask a question answered only by an uploaded document and verify the source.
4. Test failure handling
The project must explain timeouts, limits and provider outages, preserve data and offer a safe retry.
Questions
Which model runs inside the customer project?
Gemma through Google AI Studio. DeepSeek remains Studio’s internal developer.
Does a non-technical customer need a Google key?
The default connection should use Dialektai infrastructure without asking that customer for a key.
Test AI with real permitted data.
Check text, media, RAG, limits and provider outages separately.