Artificial Intelligence Platform (AIP)

Artificial Intelligence Platform, marketed as AIP, is Palantir Technologies’ orchestration layer for large language models, released in April 2023 and pitched as the connective tissue between commercial LLMs and the operational data already held inside Gotham and Foundry . Where Gotham serves intelligence and defence customers and Foundry serves enterprise data work, AIP sits across both, letting an analyst or operator address those systems in natural language and trigger the same actions a trained user would otherwise execute through the interface.

The technical premise is that an LLM on its own is ungrounded — it can summarise a document, but it cannot reliably act on a fielded unit, a supply order, or a target. AIP bridges that gap by binding model outputs to Palantir’s ontology, the typed graph of entities, properties and permitted actions that customers maintain inside Foundry. A prompt is routed to one of several available models — Palantir is model-agnostic and integrates OpenAI, Anthropic, Google and open-weight options — and the response is constrained to operations the ontology permits. Every model call, retrieval, and action is logged for audit, the feature Palantir leans on hardest in defence and regulated-industry sales. The action-graph layer turns model suggestions into reviewable workflows: a recommended target nomination, a recommended logistics reroute, a recommended sensor tasking, each with the underlying data and chain of approvals attached.

Public-sector customers include the US Department of Defense, where AIP is woven into existing Gotham deployments and the Maven Smart System contract. Commercial uptake has been the louder story — Palantir spent 2023 and 2024 running near-continuous “AIP Bootcamps” with Fortune 500 prospects, and much of the company’s US commercial revenue growth in that period traces back to those sessions.

Development since launch has focused on autonomy — letting AIP-driven agents execute longer chains of action with less human prompting between steps — and on hardening the deployment story for classified environments. The platform competes less with standalone model providers than with system integrators trying to wire those same models into government and enterprise data estates, which is the gap Palantir built its business closing.