We Built a Way for Non-Technical Teams to Ship Data Pipelines

Every company has the same problem. The sales team needs Stripe data joined with HubSpot. The product team wants Mixpanel events in a dashboard. The CEO wants a weekly revenue report that doesn't require asking three people. The data team has a backlog of 47 requests and two engineers.

The usual solutions don't work. Give everyone access to the warehouse and they write bad queries. Buy an ETL tool and you still need an engineer to configure it. Hand them a BI tool and they build dashboards with wrong metric definitions. The bottleneck isn't tooling. It's that building a data pipeline still requires someone who knows Python, SQL, cron, and infrastructure.

What if the pipeline didn't need an engineer?

That's the question we kept hitting while building Bonnard. Teams using the semantic layer still needed someone to get data into their warehouse in the first place. The data was governed once it arrived. Getting it there was the hard part.

So we built Loony.

Loony is a self-serve data pipeline platform. You describe what you need — "sync my Stripe charges into a database I can query" — and your AI agent writes the extraction scripts and SQL transforms. Loony handles the rest: schema validation, a managed Postgres instance, scheduled runs, and REST + MCP endpoints so agents and dashboards can query the data.

The difference from handing someone Airflow or Airbyte: Loony validates everything before it touches production. The agent writes code. Loony checks it.

How it actually works

A product manager wants to track deal velocity from Salesforce. They open Claude Code and describe what they need. The agent scaffolds a dlt pipeline, writes SQL transforms, and defines the schema. Loony validates the schema contract, provisions a database, and deploys it. The PM gets a REST API and an MCP endpoint they can query from any agent.

No Airflow DAG. No Docker. No infrastructure tickets. No two-week wait.

The pipeline runs on a schedule. Data accumulates. If the schema changes, validation catches it before bad data lands. Every query goes through guardrails that enforce types, keys, and access rules.

Why not just use X?

We hear this a lot. Usually X is Retool, Airbyte, Fivetran, or "just Supabase."

ETL tools (Airbyte, Fivetran) solve extraction. They don't give you transforms, validation, or query endpoints. You still need a warehouse, a transformation layer, and something to serve the data. That's three more tools and an engineer to glue them.

Internal tool builders (Retool, Appsmith) solve the UI. They don't solve the data pipeline underneath. You still need the data somewhere queryable before Retool can display it.

Databases (Supabase, Neon) solve storage. They don't solve extraction, scheduling, or governance. You get a Postgres instance. Getting data into it is your problem.

Loony does the full loop: extract, transform, validate, store, schedule, serve. One tool. One deploy command. No engineering required.

The stack

Under the hood, Loony uses dlt for extraction, Postgres for storage, and SQL for transforms. Scripts are Python. Schemas are YAML. Transforms are SQL files that run in order after each extraction. Everything is inspectable, version-controlled, and auditable.

The skills library gives agents pre-built patterns for common integrations — Stripe, HubSpot, Salesforce, GitHub, Slack, and 50 more. The agent doesn't start from scratch. It starts from a working pattern and adapts.

Sandboxes isolate each user's environment so one person's broken pipeline can't affect production data. The semantic layer ensures agents query governed definitions, not raw tables.

Who it's for

Teams where the data team is the bottleneck and non-technical people need data access. Product managers, ops leads, founders, sales teams. Anyone who's ever filed a data request and waited two weeks.

If you have a dedicated data engineering team running Airflow in production, you don't need this. If you're a startup where the CTO is also the data team, or a growing company where requests outpace capacity, this is what we built it for.

Try it

pip install loony

Describe your data. Deploy. Query. loony.dev has the full docs and a free tier.