Palantir expands workflow tools, model access and governance
Palantir’s November 2026 updates make faster pipelines the default in Pipeline Builder, adding support for Use LLM and Text to Embeddings nodes with Palantir-provided or registered models. Powered by DataFusion, these pipelines can reduce execution time and compute usage compared with standard Spark pipelines. Bulk migrations let teams propose pipeline conversions and model replacements across projects or enrollments, with changes requiring review and merging.
AIP Insight, available in beta on AIP-enabled enrollments, turns natural-language questions about Ontology data into workbooks with visible, editable analysis steps. Table, chart and map layouts still require manual configuration, and some Insight features remain unsupported. Ontology SQL and SQL functions are generally available, while Dataset Preview gains primary-key comparisons across Foundry datasets and Iceberg tables.
Experimental model additions include OpenAI’s GPT Realtime 2.1 and GPT Realtime 2.1 Mini, Google’s Gemini 4 Hydrogen and Carbon, and DeepSeek V4.1 Flash (1017). Availability varies by provider and enrollment. Supported OpenAI models also gain public web search with cited answers through Direct OpenAI and Amazon Bedrock.
Foundry usage and limits monitoring is generally available, offering organization- and user-level breakdowns without controls to change limits. Audit Explorer enters beta on selected enrollments for searching audit logs, with every search recorded for accountability. Palantir also says streaming ingestion on Foundry worker can reduce compute costs by around 97.5% compared with previous costs.