Awesome AI Agents › AI Workflow Orchestrators

run-llama/workflows-py

⭐ 449 Python repository created 2025-06-04

This repository holds LlamaAgents, the Python framework from the LlamaIndex team for building document-centric agents, whose core is the llama-index-workflows package. A workflow is a class whose steps are async Python methods annotated with a decorator; each step declares the event type it consumes and the event types it returns, and the runtime routes events between steps. Branching, looping, parallel execution, state persistence and recovery after failure are expressed in ordinary Python control flow rather than in a separate configuration language, which matters for document pipelines that combine OCR, model calls, structured extraction, classification, custom validation heuristics and human review, much of which is in-process Python such as embedding models or vision analysis that does not fit neatly into a microservice. The library is designed to be adopted incrementally. At the smallest scale it is a dependency you call from a script or notebook and await, with durability that can be left off, backed by a file, or connected to a database. When an API is needed, llama-agents-server wraps any workflow as a REST service with streaming, persistence and human-in-the-loop support, either standalone or mounted inside an existing Starlette or FastAPI application, and llama-agents-client is the matching async client for calling workflows from other services. Beyond that, a coordination backend adds durability and replication for scaling, without the workflow code being rewritten, so the same definition runs in a prototype notebook, in a product backend, or inside a customer environment where documents cannot leave the network. Dependencies are kept minimal so the library embeds cleanly. The project is tested in continuous integration with coverage reporting, distributed on PyPI, documented on the LlamaIndex developer site, and accompanied by an examples directory. It suits teams running document and agent pipelines that need reliable orchestration in plain Python.

https://github.com/run-llama/workflows-py

agentic-aiagentic-frameworkagentic-ragagentic-workflowagentsaievent-drivenorchestrationllamaindexpythondurable-executionhuman-in-the-loop