firebase/genkit

Features

• AI Application Frameworks - An open-source framework for building AI-powered applications with composable workflows, model integration, and tool calling.
• Generative AI Models - Sends a text prompt to any supported generative AI model through a single unified API, returning the model's text response.
• Tool Composition Flows - Invokes external functions or data sources during a flow to ground AI responses in live, real-world information.
• Multi-Turn Agent Conversations - Passes message history with roles to maintain context across multiple interactions with the model.
• Multi-turn Interaction Managers - Keeps a running history of messages in a session for multi-turn chat conversations.
• Model Provider Integrations - Uses a unified interface to connect with hundreds of models from multiple providers for text and image generation.
• Agentic Workflow Orchestration - Creates multi-step AI processes that call tools and retrieve external data for context-aware responses.
• AI Observability Tracing - Provides detailed trace inspection of AI feature steps, inputs, and outputs for debugging.
• AI Performance Monitoring - Collects and displays real-time latency, error, and token usage metrics for deployed AI features.
• AI Provider Orchestrators - Provides a unified SDK that connects to hundreds of AI models from multiple providers for text, image, audio, and video generation.
• AI Workflow Orchestration - Provides built-in primitives for structuring multi-step AI interactions like chat, RAG, and tool use.
• AI Workflow Orchestrators - Structures multi-step AI processes that chain model calls, tool invocations, data retrieval, and human-in-the-loop approval steps.
• Chat Model Text Generators - Generates text responses by sending prompts to chat models with optional model-specific configuration.
• Executable Prompt Files - Loads prompt files and executes them as callable functions with streaming support.
• External Tool Integrations - Invokes custom functions like weather lookups or dice rolls during a conversation and renders their results.
• Generative AI Provider Integrations - Connects to hundreds of generative AI models from multiple providers through a single unified API.
• Tool Calling - Invokes external functions during an AI generation to ground responses in live or fetched data.
• AI Chat Interfaces - Supports chat-style interactions with AI models for conversational applications through a unified interface.
• Multi-Role Prompt Construction - Constructs prompts with multiple roles like system and user for structured conversations.
• Model Parameter Configurations - Adjusts generation parameters like service tier, safety filters, and thinking level to control output quality and cost.
• Model Provider Plugins - Connects to hundreds of AI models from multiple providers through a unified plugin interface.
• Multi-Modal Prompt Integration - Embeds media like images into prompts for models that support multimodal input.
• Structured Output Generators - Produces type-safe, structured data from AI models, enabling reliable extraction of formatted information.
• Gemini Model Text Generators - Generates text, reasoning, and multimodal responses using language models with support for tool calls and safety settings.
• Tool-Attached Prompts - Associates registered tools with prompts so models can call them during generation.
• Versioned Prompt Variants - Maintains multiple prompt versions in separate files for side-by-side performance comparison.
• Prompt Iteration Workflows - Runs prompts interactively in a graphical UI for experimenting with wording and parameters.
• Prompt Templates - Defines prompts as files with Handlebars templating, schemas, and model configuration for versioning.
• Structured Output Proxies - Specifies a schema for the model's response, ensuring output is parsed and validated into a typed object.
• AI Flow API Endpoints - Makes AI flows callable by any client through standard HTTP endpoints on a backend server.
• Flow-Integrated Tool Calls - Invokes external functions during a flow to fetch live data and ground AI responses in real-world information.
• Tool Definitions - Defines callable functions with name, description, and typed input schema for LLM invocation during generation.
• Learning and Prompt Engineering - Creates, templates, versions, and tests prompts with structured schemas, media support, and a developer UI.
• AI Application Debugging - Inspects execution traces, monitors performance metrics, and debugs AI workflows locally and in production.
• AI Flow HTTP Endpoints - Serves AI workflows as HTTP endpoints callable by any client, enabling backend deployment of AI features.
• AI Feature Debugging - Provides a CLI and developer UI for running flows, comparing model outputs, and inspecting traces.
• Prompt Input-Output Schemas - Specifies structured schemas for prompt inputs and outputs for type-safe model interactions.
• Semantic Document Retrieval - Fetches relevant documents from an index using vector embeddings for semantic similarity.
• Incremental Result Streams - Sends AI-generated content incrementally to the UI, showing partial results before the full response completes.
• AI Development Toolkits - Provides a CLI, local UI, and trace viewer for testing, debugging, and monitoring AI features.
• AI Flow HTTP Endpoints - Runs AI logic in a standalone backend and makes it callable over standard HTTP requests.
• AI-Powered Full-Stack Apps - Enables building complete AI-powered applications from scratch by integrating streaming output and tool calling.
• AI-Powered Full-Stack Builders - Builds complete AI-powered applications with streaming output, tool calling, and frontend-backend communication.
• AI Application Deployment Platforms - Deploys full-stack AI applications to Firebase, Cloud Run, and other hosting platforms.
• AI Flow Backends - Creates standalone backend services that expose AI flows and model calls over HTTP.
• AI Flow Web Endpoints - Starts a server that serves defined flows as HTTP endpoints, making them accessible over the network.
• Express.js AI Flow Deployments - Serves AI flows as Express.js apps deployable to any cloud or self-hosted platform.
• Composable Workflow Executions - Provides a flow-based execution model for composable, traceable AI workflows with streaming and state management.
• AI Response Streams - Streams AI-generated content incrementally to clients as it is produced, showing results before the full response finishes.
• AI Flow Trace Viewers - Provides a visual trace viewer for inspecting each step of AI flow execution.
• Prompt and Code Decoupling - Defines prompts in separate files for rapid iteration independent of application code.
• Automated Trace Evaluation - Ships automated trace evaluation by sending production traces to an assessment system.
• Model Performance Monitoring - Tracks request volumes, latency, and error rates in a production dashboard for AI features.
• Observability Tracing - Records every step of a flow's execution as structured traces for debugging and monitoring.
• Backend-to-Frontend Bridges - Links a client-side application to a standalone server to send requests and receive AI responses.
• Model Tool Calls - Includes defined tools in a prompt so the LLM can request them to gather information or perform actions.
• Deep Research Execution - Executes extensive multi-turn research tasks using specialized models with options for collaborative planning and task cancellation.
• Interruptible Loops - Manages a configurable tool-calling loop that can pause for human approval during execution.
• Streaming Chat Responses - Sends chatbot replies incrementally as they are generated, so users see text appear word by word.
• Multimodal Media Generation - Produces images and other media alongside text from multimodal AI models in a single call.
• Document Rerankers - Reorders documents by relevance to a query using a scoring model to improve result quality.
• Generation Parameter Configurations - Adjusts settings like output length, temperature, and stop sequences to control model generation behavior.
• Built-in and Custom Tool Combiners - Mixes built-in tools with user-defined tools in a single generation request.
• Text-to-Image Generators - Creates images from text descriptions using models with configurable aspect ratio and quality.
• Human Approval - Interrupts AI workflows to ask clarifying questions and wait for human response before continuing.
• Human-in-the-Loop Tools - Halts the tool-calling loop conditionally to ask the user a question, confirm an action, or request out-of-band approval.
• Image Content Analyzers - Reasons about images provided as inline data or URLs to describe or answer questions about them.
• Video Content Analyzers - Processes video files to describe scenes, answer questions, and reference specific timestamps.
• Image Generation Models - Creates images from text prompts using image generation models with configurable settings.
• Prompt Management Systems - Defines, versions, and iterates prompts with templates, schemas, and variants.
• Handlebars - Uses Handlebars expressions and helpers to dynamically insert user input into prompt text.
• Local Model Generation - Generates text responses using locally hosted large language models configured by model name and type.
• Real-Time Tool Call Streams - Receives tool request and response parts as stream chunks for dynamic message sequence construction.
• Model Response Streaming - Delivers generated content in chunks as it is produced, improving perceived responsiveness for users.
• AI Flow Response Streams - Receives real-time chunks from streaming AI workflows as they arrive, then retrieves the final complete output.
• Dynamic Prompt Definitions - Creates prompts entirely in code using functions for dynamic generation use cases.
• File-Based Variants - Ships a file-based variant system for organizing prompt versions with shared partials and A/B testing.
• Multimodal Input Tuples - Includes images, video, or audio alongside text in prompts for models that support non-text media.
• Nested Prompt Fragments - Defines reusable template fragments that can be included across multiple prompts.
• Reasoning Depth Configurations - Adjusts the thinking depth or token budget of models to improve performance on complex reasoning tasks.
• Text-to-Speech Integrations - Converts text into natural-sounding speech using text-to-speech models with multiple voices.
• System Prompts - Provides the model with initial instructions about persona, tone, or response format to guide its behavior.
• Text Embeddings - Converts text into numerical vector representations for similarity search and NLP tasks using embedding models.
• Text-to-Speech - Produces spoken audio output from text input using text-to-speech models.
• Explicit Tool Call Handling - Takes full control of the tool-calling loop by returning tool requests and fulfilling them with custom logic.
• Tool Call Iteration Limits - Sets a maximum number of back-and-forth tool-calling cycles to control cost and prevent runaway execution.
• Text and Image Embedding Generators - Creates vector embeddings from text or multimodal input using embedding models for semantic search and similarity tasks.
• Text-to-Video Generators - Provides text-to-video generation using Veo models with configurable duration and resolution.
• Web Search Integrations - Allows supported models to retrieve and incorporate current web search results into their generated output.
• AI Workflow Testing Interfaces - Starts a local development interface to test and debug AI workflows before deploying them.
• Web Search Grounding - Augments model answers with real-time information and citations from web search.
• Location-Aware Grounding - Provides location-aware answers by integrating maps data into the model's context.
• Document Chunking and Embedding Pipelines - Provides a built-in pipeline for chunking documents, generating embeddings, and storing them in vector databases.
• AI Component Testing Interfaces - Provides a web interface for interactively testing models, flows, and prompts during development.
• Session State Persistence - Maintains conversation history and arbitrary state across invocations using pluggable storage backends.
• Arbitrary Object Persistence - Stores and updates custom JavaScript objects alongside message history for tracking user preferences.
• Local AI Workflow Debuggers - Provides a local developer interface to inspect and step through AI workflows for debugging.
• AI Frontend-Backend Bridges - Pairs web or mobile UIs with AI backends so frontends can invoke AI flows over HTTP.
• Callable AI Flow Functions - Wraps AI flows in callable functions so they can be invoked from client apps.
• Consistent Multi-Language SDKs - Provides consistent APIs across JavaScript, Go, Python, and Dart by sharing core abstractions and protocol definitions.
• Semantic Video Understanding Tools - Processes video files to describe content, answer questions, and reference timestamps within the footage.
• Cross-Language Host SDKs - Provides consistent APIs across JavaScript, Go, Python, and Dart for flexible project integration.
• Structured AI Output Streams - Streams structured output from AI models to the UI as it is generated, showing partial results before completion.
• Template Helper Registrations - Registers global helper functions for custom formatting and logic within prompt templates.
• Model Safety Filters - Blocks model responses that exceed configurable thresholds for hate speech, harassment, and other harm categories.
• AI Flow Client Invocations - Invokes deployed AI flows from client-side code using a dedicated library for non-streaming or streaming requests.
• AI Task Output Streams - Sends partial results from AI workflows to the client as they are generated for improved responsiveness.
• AI Frontend-Backend Connectors - Pairs web or mobile frontend frameworks with AI backends by calling flows over HTTP from the client.
• MCP Tool Attachments - Connects to external servers to access pre-built tools, databases, APIs, and file systems within the same generation request.
• Client Applications - Framework for building agents and performing data transformations.
• Model Serving & Deployment - Builds AI-powered apps using code-centric patterns.
• Client Applications - Framework for building AI-powered apps with MCP support.

Star-Verlauf