Docs

Glossary

The terms Amfelt uses, in one place.
Agent
The lightweight component that executes data operations inside your infrastructure. One binary is both the agent and the command-line interface.
Bridge layer
The generated, analytics-facing views over a table’s history: <table>_act holds the history table’s open records, those with no amf_valid_to, and <table>_hst all of its records, which under SCD Type 2 is every version. They are the endpoint your BI tools and consumers query.
Contract (ODCS)
An Open Data Contract Standard  document describing a source’s schema, types, and quality expectations. Generated by discovery, versioned in your Git repository, and validated on every sync. See the Data Contract Reference .
Control plane
The managed application where you configure, orchestrate, and monitor. It sends agents configuration, run commands, credentials, and upgrades, and it keeps the schedules itself, starting each scheduled run. Agents send it run events and reports, contracts, profiles for the advisor, and the credentials you store. See Security & Trust .
Data plane
The side of the boundary where your data is moved and transformed: the agent, its plugins, your sources, and your data platform, inside your infrastructure. Some values drawn from the data, such as the advisor’s profiles, still reach the control plane.
Discovery
Introspecting a source’s schemas, tables, columns, and constraints, and profiling its data. It is the input for contract and layer generation.
DTAP
The four environments (development, test, acceptance, production) a project can bind. Each keeps its own sources and data platform credentials.
Historization
Preserving the history of changing records in the generated history layer, so analytics and audits can see data as it was. Governed per object by historyMode.
History mode
The setting that decides what the history layer keeps as a source changes: full SCD Type 2 history (slowly changing dimension Type 2, which keeps every change as a new dated record), overwrite-in-place, snapshot, append, and others. Overridable per object. See Write Modes and History .
Job
A configured pipeline from a source to a target. Its runs set the load strategy and the schedule.
Layering
The generated data platform structure (staging, history, and bridge layers), maintained per table as sources evolve.
Metadata
Data about your data: schemas, contracts, lineage, run metrics. It is not all that crosses to the control plane; Security & Trust lists what does.
Plugin (connector)
The component that speaks to a specific source or target. Each plugin declares its capabilities (discover, extract, load) and by default runs as a local subprocess of the agent; a plugin configured with the grpc registry runs at an address you set.
Schema drift
A change in a source’s schema. Amfelt detects every change and classifies it as a patch, minor, or major bump by its impact on consumers. Your driftPolicy declares how runs should respond; applying it automatically is on the roadmap. See Versioning and Compatibility .
Sync (run)
One execution of a job: extract, load, and layer processing, ending in a run report.
Type resolution
Deciding a column’s data platform type, by pivoting from the source’s own type through the contract’s logical vocabulary to the target’s. See Type Resolution .
Write mode
The setting that describes how records land in staging: append, upsert, or replace. Set per job or per selection, not per object. Distinct from history mode, and not yet applied at run time. See Write Modes and History .
Last updated on October 31, 2026