AgentFlow

Overview

AgentFlow lets developers define complex LLM workflows using simple JSON configurations. It orchestrates multi-step chains of LLM calls, passing outputs between steps without writing orchestration code. Built in Python, it is a community-maintained tool for structuring prompt sequences.

Best for

Best for
Python developers who want to declaratively chain LLM calls without writing orchestration code

Use cases

• Setting up multi-turn conversations with context passing
• Chaining several LLM calls for data extraction and summarization
• Defining reusable prompt templates as JSON workflows

Notes

AgentFlow lets developers define complex LLM workflows using simple JSON configurations. It orchestrates multi-step chains of LLM calls, passing outputs between steps without writing orchestration code. Built in Python, it is a community-maintained tool for structuring prompt sequences.

323 stars on GitHub. Last updated 2023-08-11. Licensed MIT.

Use cases

• Setting up multi-turn conversations with context passing
• Chaining several LLM calls for data extraction and summarization
• Defining reusable prompt templates as JSON workflows

Pros

• Workflow logic lives in JSON, making it portable and version-controllable
• Minimal boilerplate – no need to write Python for common orchestration patterns
• Active community with 323 stars, indicating peer review and shared patterns

Cons

• JSON-only definitions can become unwieldy for intricate branching or conditional logic
• Tightly coupled to the LLM model called in each step – no built-in fallback to non-LLM actions
• Limited to Python environments; no native support for other programming languages

Indexed from awesome-langchain and enriched against its public facts.