Haystack Built Composable NLP Pipelines. The Pipeline Components Have No Agent Schema.
by Nick Clark | Published March 28, 2026
Haystack provides a composable framework for building NLP and RAG pipelines with retrievers, readers, generators, and custom components. The pipeline composition model is flexible. But Haystack pipeline components are functional units with inputs and outputs, not structurally defined agents. There is no canonical schema for what a component or agent is, how governance relates to capability, or how memory and identity persist across pipeline executions. The gap is between composable NLP tooling and canonical agent definition.
Haystack's composable pipeline architecture and rich component ecosystem provide genuine NLP engineering value. The gap described here is about structural agent definition, not about pipeline flexibility.
Components without structural identity
Haystack components are Python classes with defined inputs and outputs. A retriever takes a query and returns documents. A generator takes a prompt and returns text. Components are functional units. They have no structural identity, no governance field, no memory that persists across invocations, and no lineage.
A pipeline assembled from Haystack components is a data flow graph. It is not a structurally defined agent with canonical fields.
Pipeline state without governed memory
Haystack pipelines pass data between components through typed connections. The data flowing through a pipeline is transient. There is no governed memory accumulating across pipeline executions, no lineage tracking how memory evolved, and no policy constraining memory mutations.
What a canonical agent schema provides
A canonical agent schema would allow Haystack pipeline assemblies to be wrapped as structurally defined agents with identity, memory, governance, capabilities, execution state, and lineage as typed fields. The pipeline would provide the processing capability. The schema would provide the structural definition that makes the agent validatable, governable, and interoperable.
