Skip to content

Object Backend Overview

Source: https://www.palantir.com/docs/foundry/object-backend/overview/ Captured during the ontology-parity effort. Concrete feature taxonomy only.

Core storage & indexing

Object databases

  • Object Storage V1 (Phonograph) — legacy canonical DB; consolidated indexing + querying; tracks user edits; sunset 2026-06-30.
  • Object Storage V2 — current; decoupled indexing/querying; horizontal scaling; up to 2,000 properties per object type; incremental indexing on by default; tens of billions of objects per type.

Object Data Funnel

Orchestration service that ingests from datasources (datasets, restricted views, streaming), processes Action-driven user edits, maintains sync with upstream changes, and supports low-latency streaming.

Query & access layer

Object Set Service (OSS)

Read-serving component for filtering/search, aggregation, object loading/retrieval.

Object sets (resource type): - Static — fixed lists of primary keys. - Dynamic — filter-based, auto-updating. - Temporary — 24-hour expiring inter-service transfers. - Permanent — persistent saved resources.

Search & aggregation backend

  • Spark-based query execution (OSv2); Search Around operations (100,000-object default limit); accuracy via distributed computation.

Write & modification

Actions service

  • Structured edits; complex permission enforcement; historical action logging; up to 10,000 objects per action (configurable higher).

Functions on objects

  • Operational code execution for custom aggregations, decision-support computations, rapid in-application execution.

Schema & metadata

Ontology Metadata Service (OMS)

Definitional layer for object-type schemas, link types, action-type specs, entity metadata.

Multi-datasource capability

Multi-Datasource Objects (MDOs) — property-level permission granularity, column-level access controls, composite objects from multiple sources.

Key shift (V1→V2): decoupling indexing from querying enables independent scaling and improved throughput across edit + aggregation workloads.