What Database Does Palantir Use?

Palantir does not run on one database. Its platforms store data in layers, and each layer uses a different technology. Foundry, the data platform, keeps datasets as Parquet files in cloud object storage such as Amazon S3 or Azure Blob Storage. The Ontology is the layer that turns those datasets into objects such as customers, aircraft, or orders. It is served from Palantir's own object store, with search indexes built on Lucene and Elasticsearch. Gotham, the government platform, was built on AtlasDB, Palantir's open source transactional layer on top of Cassandra. Palantir also connects to whatever databases its customers already run, and it does not sell a database of its own.

Palantir Storage Layer by Layer

Datasets Are Files, Not Rows in a Database

A Foundry dataset is a folder of files plus a schema, not a table in a database server. Structured data is written in Parquet, a columnar file format in which each column is stored together. A query that reads three columns of fifty therefore touches only those three. The files sit in the customer's cloud object storage. Every change to a dataset creates a new transaction, so a pipeline can be rebuilt from any earlier version.

This design is why Foundry scales cheaply. Object storage costs a small fraction of a database server per terabyte, and compute is added only while a pipeline runs. Foundry now also supports Apache Iceberg, an open table format that lets other engines such as Spark, Trino, Databricks, and Snowflake read the same files without copying them.

The Ontology and Its Indexes

The Ontology is the part of Palantir that people usually mean when they ask about "the Palantir database". It maps datasets to object types, such as a Customer object with properties and links to Order objects. Gotham's graph views, Foundry's applications, and AIP all read from this layer.

Objects are loaded from datasets into Palantir's object storage service, which indexes them for filtering, search, and aggregation. Palantir's architecture documentation names Lucene and Elasticsearch in this layer. Writes made by users, such as an analyst editing a property, are recorded as actions and applied back to the datasets. The graph that analysts see in Gotham is a view over these objects and links. Palantir does not run a standalone graph database product.

AtlasDB and Cassandra in Gotham

AtlasDB is the one database technology Palantir has published in full. It is an open source transactional layer on top of key-value stores such as Cassandra. It adds serializable transactions, meaning that concurrent updates behave as if they ran one at a time. Palantir built it for Gotham, where analysts edit shared entities at the same time and must never see a half-applied change. It remains the clearest public statement of how Palantir stores operational data.

Databases Palantir Connects To

Most data in a Palantir deployment starts in the customer's own systems. Foundry ships connectors for PostgreSQL, Oracle, SQL Server, MySQL, Cassandra, MongoDB, Snowflake, BigQuery, Databricks, SAP, and many more. Some data is copied into datasets on a schedule. Some is queried in place through virtual tables, so the customer's warehouse stays the system of record.

That also answers the "how does Palantir collect data" question. Palantir does not collect data itself. Customers bring their data, and Palantir's software integrates it inside the customer's own cloud account or on the customer's own servers. AIP, the product that connects large language models to that data, is not a database either. It reads from the Ontology.

What to Take Into a System Design Interview

The Palantir pattern is a useful example for a system design interview. Use cheap object storage for raw data, a columnar file format for analytics, a search index for interactive filtering, and a transactional key-value store for shared edits. Each store does one job. That layered answer, with a reason for each layer, is what an interviewer wants when the question is to store and query a large, mixed dataset.

Palantir's own design round is described in what to expect in the Palantir system design interview, and the wider loop in the Palantir interview process, round by round. For what the company does with these systems, read what Palantir is famous for.