LlamaIndex Workflows:开源多步 Agent 工作流框架
Workflows
LlamaIndex 推出开源的 Workflows 框架,支持多步编排、循环与并行路径、全异步运行和事件驱动的状态暂停恢复,各步骤可用类型安全程序自定义数据传递。它可与 LlamaParse 配合,把 PDF、PPT 等非结构化数据转为 AI 可用格式后构建多步 Agent,也可组合多个 Agent;无生产环境或商用限制,页面还引用了 Salesforce Agentforce 团队等用户反馈。
-
25M+
package downloads a month
-
1.5k+
contributors
-
20k+
community members
01
Multi-step orchestration
Easily chain together multiple steps, introduce loops and parallel paths for more complex applications.
02
Fully async
Leverage the speed of async workflows for blazingly fast runs and easy integration with python apps like FastAPI.
03
Event-driven
Customizable workflows that can start, pause, and resume statefully and seamlessly—on demand.
04
Maximum control
Customize each step with type safe programs that let you dictate what and how arguments and data are passed across steps.
We’ve helped leading teams go from prototype to production with real-world results.
Testimonials
LlamaIndex’s framework gave us the flexibility we needed to quickly prototype and deploy production-ready RAG applications. The state of the art document parsing capabilities of LlamaParse have been particularly valuable – it handles our complex documents, including tables and hierarchical structures, with remarkable accuracy. The active community support and responsiveness of the LlamaIndex team meant we could quickly troubleshoot and optimize our implementations. What really stands out is how seamlessly we could customize the retrieval pipeline for our specific use cases while maintaining enterprise-grade performance. Salesforce Agentforce team has been leveraging LlamaIndex heavily.
LlamaIndex felt like it was six months to a year ahead of alternatives. Setting up agent workflows was straightforward, and it actually feels like the system thinks — not just runs through steps.
LlamaIndex has been a game-changer for us, accelerating our AI agent development efforts, embedding reliable conversational interfaces directly within applications, and providing a flexible and scalable agentic framework.
Other frameworks felt rigid. LlamaIndex gave us the control and flexibility we needed to meet enterprise constraints. Everyone was building their own bots. Some teams were ahead, some just starting. We wanted to create a shared foundation that made it easier to build and learn together.
01
Is Workflows Open Source?
Yes, Workflows is fully open source, granting developers full control over how they build applications and without any restrictions on leveraging your app in production or for commercial use.
02
What are common ways to use Workflows?
Workflows is used to stitch together multiple steps, including combining human-in-the-loop review, document processing, and more, all in a single end-to-end workflow.
03
How does Workflows work with LlamaParse?
LlamaParse allows builders to turn unstructured data sources like PDF, PPTs, etc. into AI-ready markdown JSON, and other formats. Workflows empowers builders to then use that data to build a multi-step AI agent that can reason, understand, and act.
04
How is Workflows different from LlamaIndex?
LlamaIndex provides developers with the core building blocks for building agents like state, memory, and more. Workflows allows developers to build highly controlled multi-step workflows that can even combine multiple agents.
来源:LlamaIndex:产品、工程与评测 · llamaindex.ai