dbt (data build tool)
What it is
dbt is the standard SQL-first transformation framework for the “T” in ELT: you write select statements (“models”), dbt turns them into tables and views in the warehouse, and adds tests, documentation, lineage, contracts and (optionally) a governed metrics layer. It does not extract or load data. (Wikipedia, secondary; dbt-core README.)
Maker, ownership and history
- Started in 2016 at RJMetrics; the commercial company (then Fishtown Analytics, now dbt Labs) shipped a commercial product in 2018 (Wikipedia, secondary).
- 2025-01: dbt Labs acquired SDF Labs (SQL compiler/static analysis tech that became the Fusion engine) (Wikipedia, secondary).
- 2025-10-13: merger with Fivetran announced; 2026-06-01: completion announced by Fivetran. Product brands stay separate (“dbt will still be dbt”). See dbt Labs and Fivetran.
- Tristan Handy (co-founder) is President of the combined company and oversees community and open source; George Fraser is CEO.
Editions and deployment
- Open source: the
dbt-labs/dbt-corerepositorymainbranch now holds the Apache-2.0 source of dbt v2.0, a ground-up Rust rewrite. v1 (Python) continues on the1.latestbranch (v1.12.5 released 2026-09-15). - dbt v2 binary (“dbt”): a distribution of the repository with dbt-specific customisations under the dbt product (Fusion) licence agreement; CLI branding distinguishes
dbt(proprietary) anddbt-oss. Free for local development; sign-in unlocks advanced features on the platform (docs.getdbt.com Fusion page). - dbt platform (formerly dbt Cloud): hosted IDE, orchestration, catalog, Semantic Layer, APIs. Plan names in the docs: Developer, Starter, Enterprise, Enterprise+ (and legacy plans). Billing is seat plus usage based (successful model builds, queried metrics, dbt State usage, AI usage). Pricing page: https://www.getdbt.com/pricing
- Fusion engine history: initial Fusion release 2025-05-28, mostly ELv2-licensed and in beta; the 2026-06-01 release announced dbt Core v2.0 (alpha) with the Fusion runtime under Apache 2.0; dbt v2.0.0 released 2026-09-14 (GitHub release) and declared GA at dbt Summit 2026 (press release 2026-09-16), replacing the Core/Fusion dual-engine model.
dbt-labs/dbt-fusionis now an archive pointing todbt-core.
Core architecture
Project = SQL and YAML files; dbt compiles a DAG and runs it inside the customer’s warehouse (Snowflake, Databricks, BigQuery, Redshift, ClickHouse and others via ADBC drivers in v2). v2 is a single binary (no Python runtime), strictly parses the language spec, and emits Parquet artifacts alongside the JSON manifest.json. Linux and macOS (x86-64, ARM) and Windows x86-64 are supported; Windows ARM not yet. v2.0.0 also adds native Databricks metric view materialisations. Latest v2 patch at fetch time: v2.0.5 (2026-09-18).
Role in an enterprise AI rollout
- Produces the tested, documented, contract-enforced tables that RAG, feature and agent workloads read; lineage, freshness and test results are machine-readable metadata for agents (see Data readiness for enterprise AI).
- The Semantic Layer (MetricFlow) gives LLM/text-to-SQL agents governed metric definitions instead of raw-table guessing; the docs state Claude and ChatGPT can query it through the dbt MCP server. Compare options in semantic-layers-compared.
- Quality companions: Great Expectations (also now stewarded by Fivetran), Monte Carlo, Soda.
AI features as of October 2026
| Feature | Status (source date) |
|---|---|
dbt MCP server (dbt-labs/dbt-mcp, Apache 2.0, v2.6.0 on 2026-10-06) | Released. Tools: Semantic Layer, Discovery (models, lineage, health), SQL (incl. text_to_sql), Admin API, codegen, dbt CLI, LSP column lineage, docs search. Remote MCP on all platform plans; local via uvx dbt-mcp (CLI and codegen). Clients named: Claude, Cursor, VS Code, Snowflake Cortex. |
| dbt Copilot | Metered AI actions per plan (docs billing page); text_to_sql consumes the Copilot allotment. |
| dbt Wizard | Project-grounded agent (lineage, tests, contracts, semantic definitions); public preview (2026-09-16). Also dbt-wizard CLI (public beta). Works with any foundation model. |
| Wizard Explore Mode | Public preview: natural-language analytics over dbt projects. |
| Wizard Desktop | Private beta. |
| dbt State | GA (2026-09-16): warehouse-metadata based build/skip/defer decisions. |
| dbt Charts | Public beta: YAML-defined BI alongside models. |
| Agent skills in v2.0.0 | Installs AgentSkills-format SKILL.md directories from packages, gated by an ai_provider flag. |
| Agents Schema (with Fivetran) | Open-source standard for agent context stored as plain SQL tables; Fivetran Context Layer built on it is private beta. |
Integrations with neighbouring layers
Warehouses/lakehouses (Snowflake, Databricks, BigQuery), Iceberg via Fivetran Managed Data Lake and Lake Compute (DuckDB, private beta), BI tools through the Semantic Layer APIs, catalogs (catalogs compared), and the OSI/Ossie semantic standard (open-semantic-interchange; the Ossie repo ships a dbt converter). Alternatives: SQLMesh, Coalesce.
Strengths and weaknesses (opinion)
- Strengths: de facto standard, huge package and community ecosystem, now an open Rust core, native agent hooks (MCP, skills).
- Weaknesses: licence and branding history (ELv2, then Apache 2.0 plus proprietary distribution) is confusing; AI and Semantic Layer features are plan-gated; the Fivetran merger concentrates ingestion and transformation in one vendor, a concern for neutrality (opinion).
Self-learning
- dbt Learn catalogue (official): https://learn.getdbt.com and course https://learn.getdbt.com/courses/dbt-fundamentals (pages load; free vs paid per course not confirmed).
- Training overview: https://www.getdbt.com/dbt-learn
- Docs and quickstart guides: https://docs.getdbt.com/guides
- Sample project: https://github.com/dbt-labs/jaffle-shop
- Community Slack: https://community.getdbt.com ; forum: https://discourse.getdbt.com
- Certifications (paid per attempt, online proctored, valid two years): dbt Analytics Engineering Certification Exam and dbt Architect Certification Exam - https://www.getdbt.com/dbt-certification
- See dbt-learning-path.
Sources (fetched 2026-10-07)
- https://github.com/dbt-labs/dbt-core (README, releases v2.0.0, v2.0.5), https://github.com/dbt-labs/dbt-fusion
- https://www.fivetran.com/press/fivetran-dbt-labs-complete-merger-to-create-the-data-infrastructure-for-trusted-ai-agents
- https://www.fivetran.com/press/fivetran-dbt-labs-announces-new-capabilities-to-make-enterprise-data-agent-ready-at-dbt-summit-2026
- https://www.getdbt.com/blog/dbt-labs-and-fivetran-merge-announcement
- https://docs.getdbt.com/docs/fusion/about-fusion, /docs/dbt-ai/about-mcp, /docs/use-dbt-semantic-layer/dbt-sl, /docs/cloud/billing
- https://github.com/dbt-labs/dbt-mcp, https://github.com/dbt-labs/metricflow
- https://www.fivetran.com/blog/dbt-wizard-an-ai-agent-that-actually-understands-your-dbt-project
- Wikipedia “Data build tool” (secondary)
Open items
- Exact current wording of the v2 product licence (proprietary distribution terms) not read.
- dbt Copilot per-plan allotments and whether Copilot is being folded into Wizard: not verified.
- dbt Summit 2026 dates and venue not verified (site says it has concluded; next is 2027-09-13..16, Las Vegas). The “Coalesce” name is now used by the separate Coalesce.io company; whether dbt retired the Coalesce conference brand is not confirmed.
- Alias
dbtalone is omitted because it already exists on lightdash.