Firebase AI Logic vs. GenKit

firebase_ai_logic_vs_genkit.md

	Firebase AI Logic	Firebase Genkit
Integration	Client-side SDK (Flutter, Web, Android, iOS, Unity)	Backend framework (Node.js, Go, Python); client apps call backend
Dart/Flutter Support	✅ Full SDK for Flutter	❌ No native SDK; use via HTTP/cloud backend
Deployment	No backend needed (uses Firebase proxy)	Requires backend deployment (Cloud Functions/Run/custom server)
Model Providers	Google only (Gemini, Imagen)	Google, OpenAI, Anthropic, Ollama, more
Input Types	Text, images, video, audio, PDFs (via Gemini multimodal)	Text, images; others depend on chosen model/plugin
Output Types	Text, JSON, images, streaming text/audio	Text, images, structured JSON; streaming via supported models
Multimodal Support	✅ Gemini supports images, media	✅ If supported by chosen model (e.g. Gemini)
Advanced Features	⚠️ Limited: supports tool/function calling, but no RAG, external plugins, or custom embeddings	✅ Full: supports RAG, plugins, embeddings, memory, multi-step logic
Orchestration	Prompt-based orchestration (multi-turn chat, function calls handled manually)	Agent-style orchestration (flows, memory, branching, tool chains)
Function Calling	✅ Supported (Gemini function calling)	✅ Supported with custom tools and plugins
Use Cases	Quick client-side AI features (chat, generate image/text)	Full-featured backend AI systems with complex logic and multi-model access
Streaming Support	✅ Token streaming, audio streaming (via Gemini)	✅ Token streaming supported (model-dependent)
Security/Quotas	Firebase App Check, 100 RPM/user default (configurable)	Fully customizable via backend infrastructure
Firebase Integration	Native (part of Firebase SDKs)	Can be integrated via Cloud Functions or callable APIs
Ease of Adoption	✅ Very easy—console + few lines of code; no backend setup	⚠️ Medium learning curve—requires backend setup & tool familiarity