Agent Orchestration Framework
A library or runtime for composing multiple LLM calls, tools, and sub-agents into a single reliable workflow — handling state, retries, and hand-offs between steps.
Once an LLM application needs more than a single prompt-response call — chaining tool calls, running steps in parallel, looping until a task is done, or coordinating multiple specialized sub-agents — an orchestration framework provides the scaffolding: a graph or state-machine model for defining the workflow, built-in retry/error handling, and observability hooks so each step can be traced. Frameworks in this category range from lightweight (a thin wrapper around tool-calling loops) to full graph-based runtimes with persistence and human-in-the-loop checkpoints. Most are open source at the core, since lock-in at the orchestration layer is a common buyer concern.
Last verified: