Developer & Vibe-Coding Documentation
Everything developers need to know about the Gemini 3 engine, prompt engineering for GenTabs, on-device AI Edge models, and building for the Disco ecosystem.
Gemini 3: Technical Specifications
Gemini 3 is not just a text prediction model — it is optimized for "System 2 thinking," meaning it can plan, reason through multi-step problems, and generate functional code in a single pass.
| Specification | Gemini 3 Pro | GPT-5.2 (OpenAI) |
|---|---|---|
| Architecture | Mixture of Experts (MoE) | Dense transformer |
| Efficiency | 390× improvement over Gemini 2 | ~50× improvement over GPT-4 |
| Code Generation | HTML, CSS, JS (vibe coding) | General-purpose code |
| Reasoning | System 2 "thinking" mode | Chain-of-thought prompting |
| Local Support | AI Edge for on-device inference | Not available locally |
| Multimodal | Text, image, video, code | Text, image, code |
Vibe Coding: A New Paradigm
"Vibe coding" is the practice of describing what you want in natural language and letting the AI generate functional code. In Disco, this means you never write HTML, CSS, or JavaScript manually — you describe the tool you need, and Gemini 3 builds it from your tab data.
Traditional Coding
const sorted = hotels.sort(...);
const table = buildTable(sorted);
document.body.append(table);
Vibe Coding in Disco
and rating. Sort lowest price first.
Add a bar chart of ratings."
Prompt Engineering for GenTabs
Writing effective prompts for GenTabs requires understanding what Gemini 3 can extract from your tabs. Follow these best practices:
1. Be Specific About Output
Say "show a table and a bar chart" instead of just "compare."
2. Name Your Data Sources
Reference tabs: "from the Booking.com tab and the Google Maps tab."
3. Specify Interactions
"Let me filter by price range and sort by rating."
4. Define Layout
"Put the map on the left and the list on the right."
5. Request Calculations
"Calculate total cost including taxes and show savings."
Google AI Edge: Local Models
For users with privacy concerns, Google AI Edge enables on-device inference — meaning your tab data never leaves your machine. Disco supports a lightweight local model for basic GenTab generation, with the full Gemini 3 cloud model available for more complex tasks.
| Feature | Cloud (Gemini 3) | Local (AI Edge) |
|---|---|---|
| Performance | Full capability | Limited to basic GenTabs |
| Privacy | Data sent to Google servers | Data stays on device |
| Latency | Depends on connection | Near-instant |
| Requirements | Internet connection | 4GB+ RAM, modern CPU |
Start Building Today
Get access to the Disco browser and start creating GenTabs with natural language prompts.