Skip to content

Glossary

Factflow reuses a few words in narrow, specific ways. When the docs say "pipeline" they rarely mean what a newcomer assumes. This page pins each term to one meaning.

A YAML file under backend/config/pipelines/ describing a set of routes wired together by queues. The authoring artifact. The UI and CLI call these "Configs" (not "Pipelines") because that is what you edit. See Author a config.

One named processing stage inside a config: an adapter plus the queue it consumes from and the queue(s) it publishes to. Routes are the atoms a config composes. See Adapters and routes.

The unit of work a route runs — a Python class with a process() method, registered by type. web_scraper, html_to_markdown, embedding_generator are adapters. See the Adapter catalog for every type.

The abstract concept of "a config's routes, running." Avoid the bare word in nav and labels — it is overloaded across the config, the running routes, and the broker topology. Prefer "config" (the file) or "execution" (a run).

One run of one config, identified by an execution ID. Carries its own config snapshot and its own isolated queues (ExecutionScopedQueue). Many executions share one broker without cross-talk. See Executions and multi-execution.

Re-emitting stored messages back into a pipeline — either the same config from storage, or a detached cross-pipeline replay from a chosen stage. The canonical home is Replay and recovery.

The record of every message as it moves hop-to-hop: one row per adapter handoff, with correlation and parent-child links. Tracked independently of the main flow. See Lineage and correlation.

A chunk of converted content (post-markdown) that becomes the unit for embedding and storage. Produced by the markdown segmentation workflow.