drt — data reverse tool

The reverse leg of your data stack.

dlt loads data in, dbt transforms it, and drt activates it back out — reverse ETL from your warehouse to the tools your team works in. Declarative YAML, one drt run.

dlt _load  →  dbt _transform  →  drt _activate

pip install drt-core          # core (DuckDB included)
drt init && drt run

— or try it in your browser, zero local setup.

Why drt?

What's always free? All connectors, CLI, MCP server, and sync engine. See OPEN_CORE.md for the open core boundary.

Quickstart

No cloud accounts needed — DuckDB + httpbin.org, three commands.

pip install drt-core
mkdir my-drt-project && cd my-drt-project
drt init --template duckdb_to_rest

That scaffolds a runnable syncs/duckdb_to_rest.yml. Seed a tiny DuckDB table and run:

python -c "
import duckdb
c = duckdb.connect('warehouse.duckdb')
c.execute('''CREATE TABLE IF NOT EXISTS users AS SELECT * FROM (VALUES
  (1, 'Alice', 'alice@example.com'),
  (2, 'Bob',   'bob@example.com'),
  (3, 'Carol', 'carol@example.com')
) t(id, name, email)''')
c.close()
"
drt run --dry-run   # preview, no data sent
drt run             # POST each row to httpbin.org
drt status          # check results

Other starter templates

drt init --template list             # see all available templates
drt init --template postgres_to_slack
drt init --template duckdb_to_hubspot

Each template prints next-steps for the env vars / source data it needs. See examples/ for the full collection (Discord, Google Sheets, GitHub Actions, MySQL, ClickHouse, BigQuery, …) and docs/connectors/ for per-connector reference.

Customizing your sync

For a guided wizard that walks you through profile + project setup:

drt init   # interactive — picks a source, configures profile, scaffolds project

Both flows produce the same project shape (drt_project.yml, syncs/, .drt/). drt sources --detailed and drt destinations --detailed print every connector's required env vars and a sample YAML stanza — useful when hand-authoring beyond the templates.

CLI Reference

drt init                    # initialize project
drt list                    # list sync definitions
drt sources                 # list available source connectors
drt destinations            # list available destination connectors
drt run                     # run all syncs
drt run --select <name>     # run a specific sync (globs work: 'users_*')
drt run --select tag:<tag>  # run syncs matching a tag (repeat --select to union)
drt run --select destination:<type>  # run syncs by destination type
drt run --exclude <name>    # subtract syncs (same grammar as --select)
drt run --failed            # re-run only syncs that failed last time
drt run --limit 10          # sampled run: send only N rows (watermark frozen)
drt run --fail-fast         # stop scheduling after the first failure
drt run --vars 'lookback_days: 1'   # override project vars: for this run
drt run --threads 4         # parallel sync execution
drt run --dry-run           # dry run
drt run --verbose           # show row-level error details
drt run --output json       # structured JSON output for CI/scripting
drt run --log-format json   # structured JSON logging to stderr
drt run --profile prd       # override profile (or DRT_PROFILE env var)
drt run --cursor-value '…'  # override watermark cursor for backfill
drt test                    # run post-sync validation tests
drt test --select <name>    # test a specific sync
drt test --store-failures   # sample up to N failing rows per failed test (masked)
drt build                   # run each sync and its tests in one pass
drt validate                # validate sync YAML configs
drt status                  # show recent sync status
drt status --output json    # JSON output for status
drt profile list            # list credential profiles in ~/.drt/profiles.yml
drt profile show <name>     # show a profile (secrets masked)
drt profile test <name>     # verify a profile's source connectivity
drt profile add <name>      # interactively add a profile
drt profile remove <name>   # remove a profile
drt serve                   # start HTTP webhook endpoint
drt docs generate --format mermaid  # print project DAG as Mermaid
drt deploy github-actions   # scaffold a scheduled sync workflow (drt-action + secrets wired)
drt mcp run                 # start MCP server (requires drt-core[mcp])
drt --install-completion    # install shell completion (bash/zsh/fish)
drt --show-completion       # show completion script

Visualize your syncs

Generate a Mermaid DAG from your local drt_project.yml and syncs/*.yml files:

drt docs generate --format mermaid > dag.md

graph LR
    subgraph Sources
        src_bigquery_prod["bigquery_prod<br/><i>bigquery</i>"]
    end
    subgraph Syncs
        sync_users_to_hubspot{{"users_to_hubspot<br/><i>upsert</i>"}}
        sync_accounts_to_hubspot{{"accounts_to_hubspot<br/><i>upsert</i>"}}
    end
    subgraph Destinations
        dst_hubspot_contacts["hubspot (contacts)<br/><i>hubspot</i>"]
    end
    src_bigquery_prod -->|extract| sync_users_to_hubspot
    src_bigquery_prod -->|extract| sync_accounts_to_hubspot
    sync_users_to_hubspot -->|load| dst_hubspot_contacts
    sync_accounts_to_hubspot -->|load| dst_hubspot_contacts
    sync_users_to_hubspot -.lookup.-> sync_accounts_to_hubspot

Shell completion

Shell completion is supported for bash, zsh, and fish:

# Recommended: auto-install for your current shell (idempotent)
drt --install-completion

# Or manually add to your shell config (run once from the target shell)
drt --show-completion >> ~/.bashrc   # bash
drt --show-completion >> ~/.zshrc    # zsh
drt --show-completion > ~/.config/fish/completions/drt.fish  # fish

Note: --show-completion outputs the script for your current shell. Run it from the shell you want to configure. The manual >> append is not idempotent — run it once only.

After installation, restart your shell and tab-complete commands and options.

MCP Server

Connect drt to Claude, Cursor, or any MCP-compatible client so you can run syncs, check status, and validate configs without leaving your AI environment.

pip install drt-core[mcp]
drt mcp run

Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "drt": {
      "command": "drt",
      "args": ["mcp", "run"]
    }
  }
}

Available MCP tools:

AI Skills for Claude Code

Install the official Claude Code skills to generate YAML, debug failures, and migrate from other tools — all from the chat interface.

Install via Plugin Marketplace (recommended)

/plugin marketplace add drt-hub/drt
/plugin install drt@drt-hub

Tip: Enable auto-update so you always get the latest skills when drt is updated: /plugin → Marketplaces → drt-hub → Enable auto-update

Manual install (slash commands)

Copy the files from .claude/commands/ into your drt project's .claude/commands/ directory.

Connectors

Per-connector reference: docs/connectors/ · Discoverable from the CLI: drt sources --detailed / drt destinations --detailed

Sources

Destinations

Integrations

Roadmap

Upcoming releases → ROADMAP.md (scope, themes, targets) Issue-level tracking → GitHub Milestones Looking to contribute? → Good First Issues

Shipped: now on v0.8.3 — --dry-run --diff previews mirror DELETEs (all three strategies, read-only) on a keyed lookup that no longer scans the whole destination · alerts.on_degraded thresholds for the failures that succeed · drt test custom SQL query tests, severity: warn, --store-failures, on top of v0.8.2's single-object drt docs --inline (navigable hosted catalog on GCS/S3) and v0.8.1's sync.match_policy (update_only / create_only on Postgres + HubSpot) · manifest schema v2, on top of the v0.8.0 drt docs lineage site · project vars: · drt build · dbt-style selection v2. Grouped milestones below; the full per-release changelog lives in CHANGELOG.md and GitHub Releases.

Next: v0.8.3 Diff Polish → v0.9 Enterprise Foundation → v1.0 Stable Release → v1.x Rust Engine

Orchestration: dagster-drt

Community-maintained Dagster integration. Expose drt syncs as Dagster assets with full observability.

pip install dagster-drt

from dagster import AssetExecutionContext, Definitions
from dagster_drt import drt_assets, DagsterDrtResource

@drt_assets(project_dir="path/to/drt-project")
def my_syncs(context: AssetExecutionContext, drt: DagsterDrtResource):
    yield from drt.run(context=context)

defs = Definitions(
    assets=[my_syncs],
    resources={"drt": DagsterDrtResource(project_dir="path/to/drt-project")},
)

See dagster-drt README for full API docs (Translator, Pipes support, DrtConfig dry-run, MaterializeResult).

CI/CD: GitHub Action

Run drt syncs straight from CI/CD with the official drt-hub/drt-action — no infrastructure, just a few lines of YAML. Trigger on a schedule, on every push, or right after dbt finishes.

- uses: drt-hub/drt-action@v1
  with:
    select: '*'
    extras: postgres
  env:
    PG_PASSWORD: ${{ secrets.PG_PASSWORD }}

Inputs cover sync selection (select), connector extras, profile, dry-run and threads; outputs expose status, succeeded, failed and duration-seconds (plus a step-summary table). Secrets are passed via env: and resolved by drt's *_env keys. See the action README for the secrets pattern and more examples (run-after-dbt, PR preview).

Ecosystem

drt is designed to work alongside, not against, the modern data stack:

Telemetry

drt collects no telemetry by default. Opting in helps us understand which sources / destinations / sync modes are actually used, so we can prioritise.

drt config set telemetry.enabled true     # opt in
drt config show-telemetry                 # preview the exact payload that would be sent
drt config set telemetry.enabled false    # opt out
DO_NOT_TRACK=1 drt run                    # universal kill switch — overrides everything

When opted in, drt sends one sync_completed event per sync. The only properties we collect are these 9 fields: drt_version, python_version, os, source_type, destination_type, sync_mode, rows_synced, duration_seconds, status. The wire envelope additionally carries event, distinct_id (a per-machine random UUID at ~/.drt/.anonymous_id), timestamp, and api_key. Sync names, model SQL, destination URLs, credentials, and project paths are never transmitted — the allow-list is enforced at the function-signature level in drt/telemetry.py. By default events go to PostHog Cloud (EU region); override with DRT_TELEMETRY_ENDPOINT and DRT_TELEMETRY_API_KEY for self-hosted PostHog or a custom collector.

Note: drt itself never transmits your IP, but the receiving PostHog backend records the TCP source IP as $ip. See docs/telemetry.md for details and how to disable / substitute the backend.

For full details see docs/telemetry.md.

Contributing

We welcome contributions of all sizes — from typo fixes to new connectors. drt has a transparent contributor ladder so your work builds toward greater trust and responsibility over time.

• Get started: CONTRIBUTING.md — setup, workflow, and your first connector tutorial
• Pick something to work on: Good First Issues
• Understand how decisions are made: GOVERNANCE.md
• What's free vs. enterprise: OPEN_CORE.md
• Versioning & breaking changes: VERSIONING.md

Contributors ✨

Every contribution — code, docs, bug reports, ideas — makes drt better. Thank you 💜

Full credits by contribution type (emoji key)

K.Masuda 💻 📖 🚧 🤔 🔧 📆 👀 🧑‍🏫	yodakanohoshi 💻 📖 🚧 🤔 👀 📆	Moavia Amir 💻 📖 🚇 🚧 🤔	Khush Domadiya 💻	Pawan Singh Kapkoti 💻	PFCAaron12 💻	Semy Ingle 💻 🚧
きわみざむらい 💻 🐛	armorbreak001 💻	pureqin 💻	Wahaj Ahmed 💻	cian-ps 💻 🐛 🚇	Erik Estrella ⚠️	Ai (藍) 📖
GokulKashyap 💻 ⚠️
Add your contributions

Disclaimer

drt is an independent open-source project and is not affiliated with, endorsed by, or sponsored by dbt Labs, dlt-hub, or any other company.

"dbt" is a registered trademark of dbt Labs, Inc. "dlt" is a project maintained by dlt-hub.

drt is designed to complement these tools as part of the modern data stack, but is a separate project with its own codebase and maintainers.

License

Apache 2.0 — see LICENSE.