Kestra AI Plugin

---

🧠 Overview

Kestra AI Plugin brings AI-native orchestration to Kestra. It enables workflows to leverage large language models (LLMs), vector databases, AI providers, and intelligent agents — declaratively.

From prompt-based completions to full Retrieval-Augmented Generation (RAG) pipelines, this plugin provides all the tools you need to build AI-driven automations within Kestra.

---

✨ Features

🗣️ Chat & Completion

• ChatCompletion — generic chat interface for LLMs (OpenAI, Anthropic, Gemini, etc.)
• Classification — classify text into predefined categories
• ImageGeneration — create images from text prompts
• JSONStructuredExtraction — extract structured data (JSON) from unstructured text

🧩 Providers

Supports multiple AI backends:

• OpenAI, Azure OpenAI, Anthropic, MistralAI, Google VertexAI, Google Gemini
• Amazon Bedrock, Ollama, LocalAI, OpenRouter, DeepSeek, DashScope
• WorkersAI (Cloudflare)

Each provider integrates authentication, model selection, and latency tracking through TimingChatModelListener.

🧱 Embeddings

Store and search semantic vector embeddings using:

• PGVector
• Elasticsearch
• Milvus
• Weaviate
• Chroma
• Pinecone
• Qdrant
• Redis
• KestraKVStore
• MongoDB Atlas

These integrations make it easy to build semantic search, context retrieval, or memory-augmented workflows.

🧠 Memory

Persistent conversational or vector-based memory via:

• Redis
• KestraKVStore

🔍 Retrieval & RAG

Tools for Retrieval-Augmented Generation (RAG):

• IngestDocument — chunk and embed documents
• Search — perform vector similarity searches
• ChatCompletion — combine retrieved context with a chat model

🌐 Tools & Agents

The plugin includes a flexible Agent framework and tools that can interact with external systems:

• AIAgent — orchestrates tool usage dynamically
• CodeExecution — safely run code blocks
• KestraTask & KestraFlow — trigger Kestra workflows and tasks from within an agent
• GoogleCustomWebSearch, TavilyWebSearch — retrieve web content
• MCP Clients (DockerMcpClient, SseMcpClient, StdioMcpClient, StreamableHttpMcpClient) — connect to external Model Context Protocol tools

---

⚙️ Example Use Cases

• Build chatbots that query your data using RAG pipelines.
• Use LLMs to classify, summarize, or extract information from text.
• Automate image generation tasks.
• Integrate AI directly in ETL or data processing pipelines.
• Orchestrate complex multi-agent reasoning workflows using AIAgent and KestraFlow.