A chain links a prompt, a model call, and a parser into a single reusable pipeline.
DefinitionWhat it means
A chain composes a fixed sequence of steps, commonly format a prompt, call a model, parse the output, into one callable unit. Chains can be linked further, the output of one feeding the input of the next, to build multi-step pipelines like summarize-then-translate, but the control flow itself stays linear and predetermined.
Why it mattersWhy you should care
Chains are the simplest and most maintainable orchestration primitive, ideal when a task's steps are known in advance and do not need to branch or loop. Most production LLM features start as a chain, and teams only reach for a graph once they need conditionals, retries, or multi-agent branching that a linear sequence cannot express.
At a glanceSee it
When the parser rejects the output, the chain loops a repair prompt back through the model and retries — exhausting the budget raises a ParseError.
Chains compose end to end, each one’s structured output becoming the next one’s input — so a single weak link cascades errors through the whole pipeline.
Where you see itIn the wild
- LangChain-style chain objects composing a prompt, model, and parser.
- Simple retrieval-then-generate pipelines with a fixed step order.
- Multi-step content pipelines like draft, critique, and revise.