Awesome Infra for AI › Workflow Orchestration for AI

AgentEra/Agently

⭐ 1655 Python repository created 2023-06-30

Agently is an AI application runtime framework designed to build robust and reliable Generative AI services. It addresses common challenges in productionizing LLM applications, such as managing model changes, output drift, streaming user experiences, action execution, workflow signals, and service boundaries. The framework emphasizes transforming 'prompt glue' into structured, observable, and recoverable runtime executions. Key features include normalized provider setup for seamless model switching, guaranteed structured output regardless of model capabilities, and instant streaming that exposes structure before the final token for real-time reactivity. Agently offers observable and portable actions, allowing local functions, built-in helpers, and custom executors to share a common Action Runtime. Its Skills Executor enables dynamic discovery, installation, and execution of runtime capabilities, while Execution Environment providers manage reusable resources like MCP processes and sandboxes. The framework facilitates dynamic task management, converting model-generated or app-generated DAG data into validated, observable task executions. TriggerFlow supports signal-driven workflows with events, fan-out, streams, pause/resume, and sub-flows. Agently also provides composable patterns for common model-app scenarios like routing, planning, reflection, and multi-agent collaboration, along with developer tooling for async APIs, FastAPI integration, and observability to support non-trivial projects.

https://github.com/AgentEra/Agently

AI application runtimeLLM agent frameworkGenAI developmentworkflow orchestrationprompt managementAI application servicesobservable executionstructured outputmodel switchingstreaming AIaction runtimeskills executordynamic taskTriggerFlowmulti-agent collaborationLLM observabilityAPI integration