dlt is an excellent open-source library. The difference is what you take home. From dlt you take home Python you own and operate; from dltHub, a managed runtime for pipelines you still wrote. From Ingest you take home typed tables in your own warehouse, and none of the code behind them.
The short version: dlt gives your engineers a library, and dltHub gives them somewhere to run it. Ingest gives you the finished pipeline and answers for it.
| Ingest | dlt | |
|---|---|---|
| What you take home | An operated pipeline. Name a source and a destination you own; Ingest authors the connector, runs it on your schedule, and answers for it when it breaks. | A Python library you write pipelines with and run yourself. dltHub adds a managed runtime and an observability dashboard on top of code you still own. |
| Who writes the connector | Agents author it from the vendor's API docs as a declarative spec, and a human reviews and merges. No connector code arrives on your side. | You do, in Python. REST sources can be declared as a config dict, and dltHub's agents scaffold a pipeline from a prompt, but the result lands in your repository. |
| Who operates it | Ingest. Every run is scheduled, monitored and retried, and a failure names the failing resource and the reason in the UI. When an API drifts, repairing it is our job. | You. Self-hosted dlt runs wherever you put it. dltHub runs the job and shows you the dashboard, but the pipeline is still yours to fix. |
| How you know it is right | A connector is not live until it has been certified against the real vendor API with exact row and column counts asserted. A run that finished is not the bar. | Your call, and your code. dlt loads what your pipeline tells it to load; deciding whether the numbers are correct is a check you write and run. |
| Source coverage | Seven kinds in one engine: REST, GraphQL, SOAP, gRPC, WebSocket, files and SQL databases. Agents build the long tail, so coverage does not end at a catalog. | REST, SQL and filesystem sources, plus anything you are willing to write yourself. dltHub’s pricing page calls the open-source set "limited verified OSS connectors". |
| Pricing shape | Per GB ingested. Estimate it from the size of your data before you connect anything. | dlt is free and Apache 2.0. dltHub starts at $1,190 per month for 500 credits, where one credit is one hour of runtime; overage steps from $1.00 down to $0.80. |
| A slow, rate-limited source | Costs what its data weighs. An API that trickles 200 MB out over six hours bills for 200 MB. | Costs six credits, the same as six hours moving a hundred times the data. Runtime billing prices the clock, not the payload. |
| Where data lands | Plain typed tables in your warehouse or database, or plain files in your own bucket, with Iceberg on your own catalog if you want it. Ingest keeps no copy. | A wide destination set you configure. dltHub lists Iceberg support, CDC and in-pipeline transformations as source-available capabilities licensed with a paid subscription. |
The goal is any source to any destination, with agents doing the work in between: building the connector, running it, and repairing it. None of it lands on your team.
Point Ingest at the API docs. Agents handle auth, pagination, rate limits, and schema. No SDK, no YAML, no connector code to own.
New connectors go from request to production in days, not quarters. The catalog is a head start, never a limit.
Every run is scheduled, monitored, and retried, with volume stats you can see. Silent failures are the old world.
APIs drift. When something breaks, agents repair the connector, with the Ingest team behind them. Breakage is our problem, not yours.
No. Ingest runs its own engine with no dlt dependency, and no Airbyte or Singer either. dlt's design was studied closely while building it, because the primitives are good ones: sources and resources, incremental cursors, load packages, schema contracts, write dispositions, and pipeline state kept in the destination. Ingest implements that shape itself, so the whole path from an API response to a merged table is one codebase it can verify, gate and repair.
The library is free. Writing the pipeline is not, and neither is being on call for it. Every source you add is Python somebody wrote, reviewed, deployed and now maintains through auth changes, pagination quirks and breaking API versions. Ingest is the opposite trade: no code arrives on your side, agents build and repair the connector with the Ingest team behind them, and you pay per GB ingested.
dltHub starts at $1,190 per month and includes 500 credits, where one credit is one dollar and covers one hour of runtime. Beyond that, credits step from $1.00 down to $0.80 as volume rises. It is a clean model when your jobs are fast and dense. It is an expensive one when your sources are slow: a rate-limited API that takes six hours to trickle out a few hundred megabytes bills six credits, the same as six hours moving a hundred times as much. Ingest bills per GB ingested, which makes the clock Ingest's problem rather than yours.
Yes, and plenty of teams should. They are not exclusive. dlt lands data through your code, Ingest lands data through a service, and both write into destinations you already own. A common split is dlt for the internal databases your team models well anyway, and Ingest for the vendor APIs nobody wants to own.
Anything either can reach, given enough Python. The real question is who writes it and who keeps it working. Ingest ships SOAP, gRPC and WebSocket as first-class source kinds beside REST, GraphQL, files and SQL, all sharing one engine for auth, rate limiting, retries, schema evolution and merge semantics. For a source with no machine-readable spec, agents work from the vendor's documentation and a certification run against the live API decides whether it ships at all.
In a destination you already own: your warehouse, your database, or your own object storage as plain Parquet, JSONL or CSV, with Iceberg on your own catalog if you want it. Ingest holds no copy. The destination keeps the rows, the schema history and the pipeline's saved position, so losing every byte on Ingest's side would lose none of your data.
Tell us the first connector you need. If it is not in the catalog, agents will build it.
Early Accessdlt and dltHub are trademarks of dltHub Inc. Ingest is not affiliated with or endorsed by dltHub. Pricing and capability claims reflect dlthub.com/pricing as read on 7 September 2026, including the $1,190 per month floor stated in its comparison table and FAQ. If something here is out of date, tell us and we will fix it.