feat: migrate @google/generative-ai → @google/genai SDK + Deep Research config

[codercatdev]

Summary

Phase 2 Sprint 1: Migrate from deprecated @google/generative-ai SDK to @google/genai + add Deep Research config fields to pipelineConfig singleton.

Spec: Replace NotebookLM with Gemini Deep Research API
Tasks: Sprint 1 Tasks 1 + 2

Task 1 — SDK Migration (4 files)

lib/gemini.ts

• GoogleGenerativeAI → GoogleGenAI from @google/genai
• Module-level client replaced with lazy-init pattern (getAI()) — prevents crash at import time if GEMINI_API_KEY is missing
• genAI.getGenerativeModel() + model.generateContent() → ai.models.generateContent({ model, contents, config })
• systemInstruction moved into config object
• result.response.text() → response.text (property in new SDK, not method)

lib/sponsor/gemini-intent.ts

• Same SDK migration pattern
• new GoogleGenerativeAI(apiKey) → new GoogleGenAI({ apiKey })
• model.generateContent(prompt) → ai.models.generateContent({ model, contents })

lib/sponsor/gemini-outreach.ts

• Same SDK migration pattern

package.json

• Removed: "@google/generative-ai": "^0.24.1"
• Added: "@google/genai": "^1.0.0"

Task 2 — Deep Research Config (2 files)

sanity/schemas/singletons/pipelineConfig.ts

4 new fields added after enableNotebookLmResearch:

• enableDeepResearch — boolean (default false)
• deepResearchAgent — string (default "deep-research-pro-preview-12-2025")
• deepResearchPromptTemplate — text with {topic} placeholder
• infographicModel — string (default "imagen-4-fast")

lib/types/config.ts

Matching fields added to PipelineConfig interface.

Key SDK Changes for Downstream Agents

// Old SDK
import { GoogleGenerativeAI } from "@google/generative-ai";
const genAI = new GoogleGenerativeAI(apiKey);
const model = genAI.getGenerativeModel({ model: "gemini-2.0-flash" });
const result = await model.generateContent(prompt);
const text = result.response.text(); // method

// New SDK
import { GoogleGenAI } from "@google/genai";
const ai = new GoogleGenAI({ apiKey });
const response = await ai.models.generateContent({ model, contents, config });
const text = response.text; // PROPERTY — no parentheses

This unblocks Sprint 2: @research (Task 3: Gemini Deep Research client) and @videopipe (Task 4: infographic generation).

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
codingcat-dev	Ignored		Mar 5, 2026 7:08pm