dltHub AI harness
The dltHub AI harness: guardrails your coding agent can't skip
Your interface to dltHub from local development. The harness equips Claude Code, Codex, or Cursor with 10 toolkits of skills, rules, and MCP tools covering the whole lifecycle: ingest, transform, explore, and operate. Your agent follows workflows dltHub maintains instead of improvising, on infrastructure dltHub runs.
Your coding agent
Claude Code, Codex, or Cursor
the agentic layerdltHub AI harness
Skills, commands, rules, and MCP: the agentic primitives to build, run, and fix pipelines
you are here
dltHub context graph
Lineage, schema, data quality, governance, run state, and API context assets
Sources
REST APIs
SQL databases
Files on S3, GCS, Azure
dltHub
the managed infrastructure layer
Ingestion & transformation
Scheduling & runs
Data quality & lineage
Warehouse
Snowflake
Databricks
BigQuery
DuckDB
Serving
Notebooks & dashboards
Data apps
Agents & MCP clients
A pipeline built and deployed, start to finish
One prompt, and the skills doing the work: a source scaffolded, checked, scheduled and running on the platform. No slides.
Building and deploying a pipeline with the dltHub AI Harness
Ten toolkits your agent can call
Skills, rules and MCP tools covering ingest, transform, explore and operate. Your agent follows workflows dltHub maintains rather than improvising, on infrastructure dltHub runs.
Every skill in the harness
What your agent reaches for, step by step, from first prompt through the pipelines it keeps running.
The guided entry point. Names a use case, checks the workspace, and hands off to the right toolkit in a few prompts.
Opus 5.0 · Quick Start · ~/pipelines
Take me through the full workflow with the GitHub API
The guided entry point. Names a use case, checks the workspace, and hands off to the right toolkit in a few prompts.
Opus 5.0 · Quick Start · ~/pipelines
Take me through the full workflow with the GitHub API
Start with one prompt
For builders
Build your first dlt pipeline
Paste this prompt into Claude, Codex, or Cursor. The agent does the rest.
Run uvx dlthub-start@latest to build my first pipeline and run it on dltHub
Pipelines your agent builds and repairs
Skills, commands and MCP servers your coding agent invokes, grounded in a graph of your own lineage, schemas and run state. An analyst ships a working pipeline in an afternoon, and the agent has the context to repair it when the source changes underneath.
AI harness
Skills, commands, rules and MCP that your coding agent invokes to build, run and fix pipelines. 38 published skills, versioned with the toolkits that ship them.
- dlthub-routerrouted to sql-database
- create-sql-database-pipelinescaffolded oracle_bicc
- setup-secretswrote the dev profile
- deploy-workspacedeploying
Context graph
Lineage, schemas, quality results and run state, connected. Your agent reads it before it writes anything.