Skip to main content

Platform Overview

This section provides a comprehensive understanding of SemanticFed — what it is, what powers it, and what it gives you out of the box.

What is SemanticFed?​

SemanticFed is a data federation platform that unifies your databases, warehouses, lakes, and SaaS APIs behind one governed semantic layer. You write a single query and get a single answer — with access policy, lineage, and audit built in — without copying the data anywhere.

A cost-aware federated query engine pushes filters, projections, and aggregations down to each source and returns live results, so you move bytes instead of terabytes. On top of federation sits a typed semantic layer where a metric like "active revenue" is defined once and resolves identically in every dashboard, notebook, API, and AI agent.

Platform components​

1. Federated query engine​

A cost-aware planner that pushes predicates, projections, and aggregations down to each source, resolves cross-source joins through typed relationships, and returns one result set. Each run records a plan summary, sources touched, duration, row count, and cache status. (The live engine ships on the roadmap; the model, query, run, and plan records exist today.)

2. Governed semantic layer​

A versioned, lifecycle-managed model of your business (DRAFT → PUBLISHED → DEPRECATED → ARCHIVED) made of entities, dimensions, metrics, and relationships — documented, typed, and consumed identically everywhere.

3. Governance at the query edge​

Row filters, column masking, and entity-level ALLOW/DENY policies defined once and enforced on every federated query, plus an attributable audit trail and end-to-end lineage.

Ecosystem​

SemanticFed ships a full ecosystem of clients and tools:

  • Web app — app.semanticfed.com — the full product in your browser
  • Node SDK — @semanticfed/sdk — docs
  • Python SDK — semanticfed-sdk — docs
  • CLI — @semanticfed/cli — docs
  • Browser Extension — Chrome MV3 side panel — docs
  • VS Code Extension — burdenoff.semanticfed-code-extension — docs
  • MCP Server — @semanticfed/mcp — docs
  • Doctor E2E — backend test suite — docs

All clients default to production endpoints. Switch to local or alpha via BURDENOFF_ENV. See Environment Selection.

Key platform characteristics​

Query in place, don't copy​

  • Zero-copy by default — results reflect live operational systems; no nightly pipeline to babysit
  • Push-down for speed and low cost — work executes at the source; you move bytes, not terabytes
  • Virtual views instead of pipelines — model the shape you need as a view over federated sources
  • Data-residency friendly — regulated rows can stay inside their jurisdiction while still being part of a unified model

One model, governed once​

  • Model once, consume everywhere — SQL, BI tools, notebooks, the GraphQL API, and AI agents resolve the same definitions
  • Typed and relational — every dimension has a type, every metric an aggregation, every relationship a kind and explicit fields
  • Versioned as a governed event — publishing a model bumps its version, so a definition change is reviewable

Trustworthy context for AI​

  • Documented metrics, not mystery columns — agents query named, typed metrics rather than guessing at raw tables
  • The same policy as a human — row/column/entity policy applies to a generated query exactly as to a hand-written one
  • Grounded and reproducible — every answer resolves to an inspectable plan

Supported sources​

SemanticFed connects relational databases (PostgreSQL, MySQL, MongoDB), cloud warehouses (Snowflake, BigQuery, Redshift, Databricks, ClickHouse), object-storage lakes (S3 / Parquet), and SaaS REST/GraphQL APIs. Each connector advertises its push-down capabilities so the planner uses every source to its strengths. (Live connector introspection expands through the roadmap.)

Getting started​