Palantir Puts Ontology into Code: Enterprise AI Enters the “Ontology‑as‑Code” Era

01 AI Can Write Code, but It Still Doesn’t Know How Enterprises Operate

Coding agents can generate functions, modify files and even build full applications, yet the real challenge in enterprises is ensuring that generated code acts correctly on real‑world business objects.

02 SuperRepo Is More Than a Code Repository – It Unifies the Business World

On 2026‑08‑04 Palantir announced SuperRepo (Beta), a monorepo that stores Ontology definitions, Functions and React apps together. In traditional projects these three layers live in separate tools and repositories, leading to semantic drift when business objects change.

For example, adding a “high‑risk supplier” concept requires updating object attributes, risk‑calculation logic, front‑end views and exposing new actions to agents. If these changes are released on different schedules, the front‑end may see a new field while the back‑end still uses the old definition, causing agents to act on stale data.

03 Ontology‑as‑Code: Code Becomes the Source of Business Definitions

Developers declare object types, relationships, interfaces and actions in TypeScript. These definitions are compiled into real Ontology entities in the Foundry environment, making the code the factual source of business semantics. Because the definitions live in the code repository, they can be version‑controlled, reviewed, automatically tested and rolled back like any other software artifact.

SuperRepo automatically generates OSDK bindings from Ontology‑as‑Code. When an Ontology change is detected, the CLI regenerates types so Functions and front‑end code can consume the updated bindings without a separate release cycle.

04 Integrating with MCP – Exposing Controlled Capabilities to Agents

SuperRepo works alongside Palantir MCP and Ontology MCP (GA June 2026). MCP provides a platform for building and consuming Ontology‑based tools, while Ontology MCP exposes objects, actions and Functions to external agents under OAuth 2.0 and the developer‑defined permission model. Agents therefore receive a limited, audited set of capabilities rather than unrestricted access.

05 Why “Ontology‑as‑Code” Beats a Simple Semantic Layer

Traditional semantic layers answer “what does this metric mean?”; Ontology‑as‑Code also answers “what is the current state of this object?”, “how is it related to other objects?” and “what actions are allowed?”. By bringing these definitions into the codebase, business models gain the same versioning, testing and CI/CD discipline as application code, reducing the risk of mismatched models across data, services and agents.

06 Beta Limitations and Roadmap

SuperRepo is still in Beta. It currently supports TypeScript v2 Functions; Python Functions, external data‑source calls, Agent Engine, Agent SDK, PySpark transforms and Compute Modules are not yet available. The product may not be enabled in all Foundry environments, and some capabilities listed in the official table are still under development.

Capability overview (summarised):

Business model : TypeScript Ontology‑as‑Code, import existing Ontology.

Functions : TypeScript v2 supported; Python and external calls pending.

Applications : React/OSDK apps with local preview.

Agent access : Via existing Ontology MCP/Palantir MCP; full Agent Engine integration pending.

Data & compute : Objects can declare data sources; PySpark pipelines not yet supported.

Delivery : Reproducible marketplace products, CI/CD deployment; Automate and full data pipelines forthcoming.

07 Conclusion – Business Semantics Must Become Code

AI coding speeds up code generation, but enterprise AI succeeds only when the underlying business world is expressed as code that can be authorized, constrained, verified and tracked. SuperRepo’s Ontology‑as‑Code moves business objects, relationships, rules and actions into the software‑engineering lifecycle, laying the groundwork for trustworthy, versioned enterprise AI runtimes.