Complex AI workflows, RAG pipelines, LLM application workflows, ML training and deployment pipelines, AI-powered data processing, and custom AI workflows. Anything that combines LLMs or ML models with data and APIs.

Agentic workflows involve AI systems making multi-step decisions, calling tools, and coordinating actions with minimal human intervention. Kestra orchestrates these workflows with built-in guardrails: human-in-the-loop approval steps for high-risk actions, retry logic for transient failures, timeouts to prevent runaway execution, and stateful execution for long-running processes. We support MCP (Model Context Protocol) server integration for tool discovery and execution.

Try Kestra locally with Docker in 2 minutes. Browse AI Blueprints for AI workflows, RAG pipelines, or LLM applications. Start with a template, customize for your use case, and deploy. Or book a demo to see AI orchestration in action.

Yes. We have native integrations for OpenAI, Anthropic Claude, Cohere, Mistral, and Hugging Face. You can also call any LLM API (including open-source models) using HTTP tasks or Python scripts.

Absolutely. Kestra orchestrates the overall workflow while you use LangChain/LlamaIndex for complex logic and LLM interactions. Run them in Python tasks, capture outputs, and coordinate multi-step workflows declaratively.

Yes. Use conditional logic to route workflow paths based on LLM responses, execute parallel LLM tasks, and chain multiple AI operations. For complex patterns, integrate LangChain or LlamaIndex within your workflow tasks.

Kestra has built-in retry strategies with exponential backoff and configurable rate limiting. Set it once in YAML (e.g., max 10 retries, 60 requests/minute) and it applies automatically to all LLM calls.

We support Pinecone, Weaviate, Qdrant, Chroma, and Elasticsearch. Orchestrate document ingestion, embedding generation, vector store updates, and retrieval in one workflow.

Native LLM tasks automatically capture token counts and calculate costs based on provider pricing. View costs per workflow run, per task, or aggregated across all workflows. Export to your analytics tools.

Full execution traces show every LLM call, workflow decision, and data transformation. View logs, inspect intermediate outputs, check token usage, and see exactly where failures occurred. Replay failed workflows after fixes.