Agent Actions
Declarative framework for multi-step LLM workflows
Start new thread
Muizz Kolapo

Agent Actions - Declarative framework for multi-step LLM workflows

by•
Agent Actions is a declarative framework for building multi-step agentic workflows. Define your pipeline in YAML: what each step does, what model it uses, what data it sees, how to validate the output. The framework handles orchestration, retries, and batch execution. The unit of work is the action. Every action is a self-contained task that can be completed by an LLM, a Python tool call, or a human-in-the-loop dashboard.

Add a comment

Replies

Best
Muizz Kolapo
Maker
📌
For the past three years I have been working on this as a side project. Two lectures shaped the design more than anything else. Stephen Wolfram: "I see LLMs as kind of an important layer of linguistic user interface... they will be just the linguistic transport layer for the actual content of what's going on." As someone who is always tasked to build data pipelines, that made me think of how we build data transformation layers. If LLMs are a transport layer, then we could reduce each call to be seen as a transformation step. Data goes in, structured data comes out, and feeds the next stage. I wanted this to be a tool where the outcome depends on your engineering decisions, not just generic knowledge. The framework gives you the primitives, but how you decompose the problem, how you scope the context, which steps you make deterministic, that is where your imagination matters. Try it out: - `pip install agent-actions` - `agac init example book_catalog_enrichment`. - Docs at docs.runagac.com. Feedback and contributions welcome.