What is GenKit Go 1.0?

Key Concepts

• Genkit Go: An open-source framework by Google for building full-stack AI-powered applications using Go.
• Flows: Functions wrapped with observability, streaming, and other AI-specific features.
• AI Primitives: Composable and easy-to-use functions provided by Genkit for AI development (e.g., generateData).
• Generate Data: An AI primitive specifically for generating structured output (JSON) and unmarshalling it into Go structs.
• Genkit CLI: Command-line interface for Genkit, used for development and debugging.
• Genkit Dev UI: A developer UI providing tools for model exploration, flow execution, tracing, and evaluation.
• Open Telemetry: Observability framework used by Genkit for tracing and monitoring.
• Model Providers: Services like Gemini that provide access to large language models (LLMs).
• Evaluation Framework: Tools for building datasets, batch running flows, and attaching evaluators (LLM-powered or custom) to assess output quality.

Getting Started with Genkit Go

• Installation: Genkit Go is installed as a standard Go library using go get. The Genkit CLI is also recommended.
• API Key: An API key is required to access LLMs. The video uses the Gemini API as an example, with instructions provided for obtaining a free key. Instructions for other model providers are available in the documentation.
• Project Setup: The video demonstrates creating a new Go project with a go.mod file and installing the Genkit Go library.

Defining Schemas and Flows

• Schemas: Go structs are used to define the input and output schemas for flows. JSON tags and schema metadata (e.g., descriptions) can be added to the struct fields.
  • Example: An input schema with ingredients and dietaryRestrictions fields, and an output schema with a recipe field.
  • Importance of Descriptions: Descriptions are sent to the LLM and help it generate the correct data.
• Flow Definition: Flows are defined as functions wrapped with Genkit's flow functionality.
  • Example: A recipeGenerator flow that takes ingredients and dietary restrictions as input and returns a recipe.
  • Streaming: Flows can be defined with streaming capabilities, allowing for chunk-by-chunk processing of the output.
• AI Primitives: Genkit provides AI primitives like generateData for interacting with LLMs.
  • generateData: Takes a prompt and a schema as input and returns the generated data unmarshalled into the specified Go struct.
  • Example: Using generateData to generate a recipe in JSON format based on a prompt and the recipe schema.

Running and Debugging with Genkit Tooling

• Running with Genkit Start: Prefixing the go run command with genkit start -- enables Genkit tooling.
  • Command: genkit start -- go run main.go
• Genkit Dev UI: Provides access to various development and debugging tools.
  • Models: Allows exploring and testing available models. Supports multi-modal input.
    • Example: Testing the Gemini 2.5 Flash model with a text prompt and the Nano Banana image preview model with an image generation prompt.
  • Flows: Enables running flows with code completion support and viewing traces.
    • Example: Running the recipeGenerator flow with specific ingredients and dietary restrictions.
  • Traces: Provides detailed information about flow executions, including inputs, outputs, model calls, and token usage. Built on Open Telemetry.
    • Token Usage: Displays input and output tokens for model calls, helping optimize for cost.
  • Prompts, Embedders, Tools: Other features in the Dev UI for working with prompts, embeddings, and tool calling.
  • Evaluation Framework: Allows building datasets, batch running flows, and attaching evaluators to assess output quality.

Building Full-Stack Applications

• Client Libraries: Genkit provides client libraries for various UI frameworks (e.g., Angular, Dart/Flutter) to access Genkit logic running in the backend.
• HTTP Endpoints: Flows can be easily exposed as HTTP endpoints using Genkit's helper functions.
  • Automatic REST API: Flows are automatically converted into REST APIs with full typing.
  • CORS Handling: Genkit provides options for handling CORS and other web app security concerns.
• Example: An Angular app that takes ingredients and dietary restrictions as input and displays the generated recipe.
• Tracing from Client Apps: Flows invoked from client apps are also traced in the Genkit Dev UI, allowing for end-to-end debugging.
• Debugging Workflow:
  1. Run the backend with genkit start.
  2. Invoke the flow from the client app.
  3. Inspect the trace in the Genkit Dev UI.
  4. Open the flow in the flow runner to debug and rerun with different inputs.
  5. Open specific model calls in the model playground to tweak prompts and inspect outputs.

Conclusion

Genkit Go provides a comprehensive framework for building full-stack AI-powered applications. It simplifies the development process with features like automatic schema handling, AI primitives, and a powerful Dev UI. The ability to trace flows from client apps and debug individual model calls makes it easier to build and maintain complex AI systems. The stable release of Genkit Go encourages developers to explore its capabilities and contribute to the open-source community.