

I didn’t want an LLM executing MongoDB queries directly, so I built a validation layer
Giving an LLM direct database access felt like the wrong abstraction.
so built ANDI around a different workflow:
intent → query plan → local validation → developer inspection → execution
The model proposes a query, but deterministic code decides whether it can run.
It’s open source and still early. I’d like feedback on whether this API would be useful in real Python/MongoDB projects.
built an open-source safety layer for querying MongoDB with natural language
we are building ANDI, a python library that converts natural-language requests into inspectable, read-only MongoDB query plans.
It includes typed runtime variables, local policy validation, query limits, and a compile → inspect → execute workflow. Database records are never sent to the LLM; only schema metadata.
I’m looking for a few Python/MongoDB developers to test it.