Analytics that stays home
We build DataAgent because the fastest way to understand your data shouldn't require handing it to someone else.
Most tools want your data first
You upload a file, a copy lands in someone else's bucket, and from then on it is someone else's compliance problem — even when you only wanted one number. For a growing set of teams that question is answered before the tool is ever chosen: the rows are not allowed to leave the building.
DataAgent takes the other route. It is a Windows desktop app: the warehouse is a DuckDB file on your disk, the model is one you choose (a local Ollama model, if you want nothing to leave the machine), and the whole product is one installer you can read top to bottom. The trade-offs are real (one laptop, one writer, no shared dashboard URL) and we would rather name them than hide them.
Open source, on purpose
“Local-first” is only trustworthy if you can verify it. The whole stack (the FastAPI backend, the DuckDB warehouse, the ETL pipelines, authentication and packaging) is open and inspectable under the GNU AGPL v3 licence. Security claims that cannot be read are not claims; they are marketing.
The changelog is the honest record of what is in the current build, including the parts that are still open. Read the code, or open an issue — both reach the same handful of people.
"Your data already lives on your machine. The tools you use to understand it should too." — The DataAgent principle everything else follows from
What we optimize for
No telemetry from the desktop app and no default egress. If data leaves, you configured the destination yourself.
The query behind an answer travels with the answer and is written to the audit log. An assistant you can't audit is a liability, not a feature.
Single writer, parameterized queries, secrets out of git. Security that fits in one paragraph.
Read the code, fork it, self-host it. Trust is earned in diffs, not in policy PDFs.