Chains - cognis

Chain examples cover composition: piping primitives into typed pipelines, extracting structured data from prose, and recovering when an LLM emits dirty JSON. Sources under examples/chains/.

Name	Scenario	Source
`chains_pipe_operator`	Triage incoming support tickets through an embed → categorize → tag pipe.	src
`chains_structured_extraction`	Parse a meeting note into a typed `Vec<ActionItem { who, what, due }>`.	src
`chains_structured_parsing`	Repair an LLM reply that’s almost-valid JSON before parsing.	src
`chains_typed_runnable`	Custom `Runnable<UserQuery, RankedResult>` impl — end-to-end type safety with `.pipe()`.	src

How to run

COGNIS_PROVIDER=ollama COGNIS_OLLAMA_MODEL=llama3.1 \
  cargo run -p cognis-examples --example chains_structured_extraction

See also

Runnables

The contract these chains compose.

Structured output

Parsers, fixers, retry strategies.

Parsers examples

OutputFixingParser and RetryParser in action.