Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/firebase/genkit/llms.txt

Use this file to discover all available pages before exploring further.

The Genkit CLI provides powerful commands for developing, testing, and managing your AI applications. It includes tools for running flows, evaluating models, and launching the Developer UI.

​
Installation

Install the Genkit CLI globally using npm: 
npm install -g genkit-cli
The CLI is distributed as the genkit-cli package and provides the genkit command for interacting with the Genkit AI framework.

​
Core Commands

​
start

Run your application in development mode with telemetry and Developer UI: 
genkit start -- <command to run your code>
Options:
  • -n, --noui - Do not start the Developer UI
  • -p, --port <port> - Port for the Developer UI (defaults to 4000-4099 range)
  • -o, --open - Open the browser on UI startup
  • --disable-realtime-telemetry - Disable real-time telemetry streaming
  • --cors-origin <origin> - Specify the allowed origin for CORS requests
Example: 
genkit start -- npm run dev
This command wraps your development server with telemetry collection and launches the interactive Developer UI for testing and debugging.

​
ui:start

Start the Developer UI standalone (connects to runtimes in the same directory): 
genkit ui:start
Options:
  • -p, --port <number> - Port to serve on (defaults to 4000)
  • -o, --open - Open the browser on UI startup

​
ui:stop

Stop the running Developer UI server: 
genkit ui:stop

​
Flow Commands

​
flow:run

Run a flow with provided input data: 
genkit flow:run <flowName> [data]
Arguments:
  • <flowName> - Name of the flow to run
  • [data] - JSON data to use as input (optional)
Options:
  • -w, --wait - Wait for the flow to complete
  • -s, --stream - Stream output
  • -c, --context <JSON> - JSON object passed to context
  • --output <filename> - Name of the output file to store results
Examples: 
# Run a flow with inline JSON data
genkit flow:run simpleGreeting '{"customerName":"Sam"}'

# Run a flow and save output to file
genkit flow:run simpleGreeting '{"customerName":"Sam"}' --output result.json

# Run a flow with streaming enabled
genkit flow:run menuQuestion '{"question":"What drinks do you have?"}' --stream

​
flow:batchRun

Batch run a flow using a set of data from a file: 
genkit flow:batchRun <flowName> <inputFileName>
Arguments:
  • <flowName> - Name of the flow to run
  • <inputFileName> - JSON file containing batch data
Options:
  • -w, --wait - Wait for the flow to complete
  • -c, --context <JSON> - JSON object passed to context
  • --output <filename> - Name of the output file to store results
  • --label [label] - Label flow runs in this batch
Example: 
genkit flow:batchRun simpleGreeting inputs.json --output results.json
The input file should contain an array of inputs: 
[
  {"customerName": "Alice"},
  {"customerName": "Bob"},
  {"customerName": "Charlie"}
]

​
Evaluation Commands

​
eval:run

Evaluate a dataset against configured evaluators: 
genkit eval:run <dataset>
Arguments:
  • <dataset> - Dataset to evaluate (JSON file)
Options:
  • --output <filename> - Output file for evaluation results
  • --output-format <format> - Output file format (csv, json) - defaults to json
  • --evaluators <evaluators> - Comma-separated list of evaluators to use
  • --batchSize <batchSize> - Batch size for parallel evaluations (default: 1)
  • --force - Automatically accept all interactive prompts
Example: 
genkit eval:run dataset.json --evaluators faithfulness,relevance --output eval-results.json

​
eval:flow

Evaluate a flow against configured evaluators: 
genkit eval:flow <flowName> [data]
Arguments:
  • <flowName> - Name of the flow to evaluate
  • [data] - JSON data to use as input (optional)
Options:
  • --input <input> - Input dataset ID or JSON file for evaluation
  • -c, --context <JSON> - JSON object passed to context
  • -o, --output <filename> - Output file for evaluation results
  • --output-format <format> - Output file format (csv, json)
  • -e, --evaluators <evaluators> - Comma-separated list of evaluators
  • --batchSize <batchSize> - Batch size for parallel evaluations
  • -f, --force - Automatically accept all interactive prompts
Example: 
genkit eval:flow menuQuestion --input test-dataset.json --evaluators answer-relevance

​
eval:extractData

Extract evaluation data for a flow from the trace store: 
genkit eval:extractData <flowName>
Options:
  • --output <filename> - Output file for extracted data
  • --maxRows <maxRows> - Maximum number of rows (default: 100)
  • --label [label] - Filter by flow run label
Example: 
genkit eval:extractData simpleGreeting --output extracted-data.json --maxRows 50

​
Configuration Commands

​
config

Set development environment configuration: 
genkit config
This command guides you through configuring your Genkit development environment.

​
init

Initialize a project directory with Genkit: 
genkit init [options]
This command sets up a new Genkit project with the necessary dependencies and configuration.

​
Global Options

  • --no-update-notification - Do not show update notification
  • --non-interactive - Run in non-interactive mode (use default choices)
  • --version - Display the CLI version
  • --help - Display help information

​
Tips

  • Use genkit start during development to automatically collect traces and test flows in the Developer UI
  • The CLI automatically detects your project root and runtime environment
  • Set the GENKIT_ENV environment variable to dev when running your application to enable development features
  • Use --stream flag with flow commands to see output as it’s generated
  • View evaluation results in the Developer UI at /evaluate/<evalRunId>

​
Next Steps

  • Learn about the Developer UI for visual testing and debugging
  • Explore Testing strategies for AI applications
  • Read about Debugging techniques for flows

Build docs developers (and LLMs) love

Get started for freeTalk to us