Palantir Foundry: What It Is, How It Works, and Who Uses It
Palantir Foundry is an operating platform for enterprise data. It connects the systems a business already runs, turns their output into governed, versioned datasets, models the business as an ontology of objects, links, and actions, and serves the applications and AI agents that operators use to make and record decisions. Palantir's 2025 annual report describes Foundry as a central operating system for an organization's data, logic, and actions, and says all of Palantir's commercial customers use it. The write path separates Foundry from a warehouse or lakehouse: Foundry is built to run operations, where most data platforms stop at analyzing them.
Product names in this guide are current as of October 2026, and Palantir does rename tools: AIP Agent Studio is now AIP Chatbot Studio. For where Foundry sits among Palantir's other products, see what Palantir does.
Palantir Foundry at a glance
Palantir's Foundry system is one platform with distinct layers, and each layer has named tools. The table summarizes them from Palantir's platform summary and application reference.
| Layer | What it does | Foundry tools, as of October 2026 |
|---|---|---|
| Data connection | Syncs data from databases, cloud storage, APIs, files, and ERP systems such as SAP | Data Connection and its connectors |
| Pipelines | Cleans, joins, and transforms raw data into versioned datasets, in batch or streaming | Pipeline Builder, Code Repositories, Code Workspaces |
| Data quality and lineage | Monitors datasets and builds, and traces every output back to its inputs | Data Health, Data Lineage, Workflow Lineage |
| Ontology | Maps datasets to object types, properties, and link types, and defines the actions and functions that change them | Ontology Manager, Object Explorer, Ontology SDK |
| Applications and analytics | Gives operators and analysts tools that read and write ontology data | Workshop, Slate, Quiver, Contour, Notepad, Map |
| Automation | Runs actions, functions, or AI logic on a schedule or when data changes | Automate, Foundry Rules |
| AI | Connects large language models to the ontology for functions, chatbots, and evaluations | AIP Logic, AIP Chatbot Studio, AIP Evals, AIP Assist |
| Delivery | Deploys and upgrades the platform across environments | Apollo |
| Governance | Controls who sees what and records what they did | Projects, Organizations, roles, markings, Checkpoints, Sensitive Data Scanner |
The Palantir Foundry architecture, layer by layer
Foundry's architecture runs from source systems at the bottom to applications and AI at the top, with the ontology in the middle. Palantir's architecture documentation says AIP and Foundry together consist of more than 300 microservices running in an autoscaling compute mesh on zero-trust security infrastructure, organized as data, logic, and workflow services around the ontology.
Data connection and integration
Data Connection is where Foundry meets the rest of the enterprise. It syncs data through pre-built connectors for databases, cloud storage, APIs, and SAP. Pipeline Builder gives data engineers a point-and-click environment for batch and streaming pipelines, and Code Repositories gives them a web IDE for production transforms in Python, Java, and SQL, with Git version control and code review. Data Health raises an alert when a dataset or build fails its checks, before the problem reaches an operator's decision, and Data Lineage traces every output to its inputs, which is the evidence an auditor or an executive needs when asking where a number came from.
The ontology: objects, links, actions, and functions
The ontology is the layer that makes Foundry behave differently from a lakehouse. Palantir's ontology documentation defines an object type as "the schema definition of a real-world entity or event," a link type as "the schema definition of a relationship between two object types," and an action type as "the schema definition of a set of changes or edits to objects, property values, and links." Functions are code-based logic that takes input parameters and returns an output. Teams define all of it in Ontology Manager, explore it in Object Explorer, and reach it from outside Foundry through the Ontology SDK, which generates TypeScript, Python, and Java clients.
The payoff is reuse: every new application, analysis, and AI agent works from the same semantic layer and the same permissions. The cost is that ontology decisions compound. A stable primary key and a deliberately scoped first model make each later use case faster, and a rushed model turns every workaround into permanent structure. Our Palantir ontology guide covers primary keys, backing datasets, and action design.
Applications: Workshop, Slate, Quiver, Contour, and Notepad
The application layer is where operators work. Workshop is the low-code builder for operational applications, with widgets bound to ontology data and actions. Slate is the framework for custom HTML, CSS, and JavaScript when an application needs its own branding or a public-facing interface. Quiver handles analysis of ontology objects and time series, Contour handles top-down analysis of large datasets outside the ontology, and Notepad is an ontology-aware editor for collaborative reports.
The design distinction that matters is read versus write. A Contour analysis answers a question. A Workshop application backed by action types lets a planner reroute a shipment, validates the change against available capacity, writes the update back to the ERP, and records who made the call. Programs that ship only the first kind end up with expensive dashboards.
AIP: the AI layer
AIP, Palantir's Artificial Intelligence Platform, runs on the same ontology and permissions. AIP Logic is a no-code environment for building, testing, and deploying LLM-powered functions that query objects and propose ontology edits. AIP Chatbot Studio builds assistants grounded in ontology data, documents, and custom tools; Palantir's documentation notes that it was previously called AIP Agent Studio. AIP Evals tests functions and prompts against defined cases. Because an agent acts through the same action types as a human operator, its changes face the same validation and land in the same audit log, and that guarantee is only as good as the ontology underneath. Our AIP guide covers how teams move from a first function to production agents.
Apollo: how Foundry is delivered
Apollo is Palantir's delivery platform, and its solution overview describes delivery across clouds, on-premises data centers, edge hardware, and disconnected networks, so the same Foundry can run in each. Apollo's documentation describes hubs that issue upgrade plans to spokes, constraints that hold a plan until it is safe to run, and automatic recall of a bad release. For government buyers, the relevant facts are the authorizations: Palantir Federal Cloud Service holds FedRAMP High and Defense Department Impact Level 5 and 6 authorizations, and in February 2026 Palantir extended the IL5 and IL6 authorizations to on-premises and edge deployments of Apollo, Gotham, Foundry, and AIP.
How does Foundry handle governance and security?
Foundry enforces security in the data layer, so every application and agent inherits the same rules. The controls below come from Palantir's platform documentation, and the right-hand column states what each one gives an auditor.
| Control | What it does | What it gives an auditor |
|---|---|---|
| Projects | The primary security boundary for organizing work and resources | Access scoped by use case, which maps to least-privilege reviews |
| Organizations | Mandatory controls that separate groups of users into silos | Separation of subsidiaries, regions, or partners on one platform |
| Roles | Discretionary permissions named Owner, Editor, Viewer, and Discoverer | A reviewable record of who can change what |
| Markings | Mandatory controls for sensitive data such as PII and PHI | Regulated data restricted to users who hold the marking |
| Data Lineage | Tracks how every dataset and output was produced | Evidence for where a number came from |
| Checkpoints | Prompts users for a justification during sensitive interactions | A recorded reason for each sensitive access |
| Sensitive Data Scanner | Finds defined sensitive-data patterns and automates actions on them | Detection of personal data that arrives in an unclassified feed |
| Audit logging, encryption, row-level and column-level security | Records user activity and protects data at rest and in transit | Evidence for logging and access controls under SOC 2, HIPAA, and similar frameworks |
Configure markings, purpose-based access, and audit routing in the first build, before the platform holds production data, because retrofitting classifications onto a populated ontology costs more than designing them in.
What is Palantir Foundry used for?
Foundry fits operational decisions that depend on many systems at once and need someone to act on the answer. The recurring patterns are supply chain control towers that see across plants, carriers, and inventory; production and yield work in manufacturing; risk, compliance, and audit operations that need lineage as much as data; customer operations that unify billing, support, and usage; and the data foundation under enterprise AI programs. Public deployments show the range.
| Industry | Organization | What is public |
|---|---|---|
| Aviation | Airbus | Launched Skywise, an aviation data platform, "in collaboration with Palantir Technologies" in June 2017 |
| Reinsurance | Swiss Re | Chose Palantir in 2018 and built its data platform on Foundry; nearly half of its workforce uses the platform each quarter |
| Energy | bp | Has used Palantir software in oil and gas production since 2014 and in September 2024 signed a five-year agreement covering a digital twin fed by more than 2 million sensors |
| Public healthcare | NHS England | Runs the Federated Data Platform, "built on Palantir's Foundry software," under a contract valued at £330 million over seven years |
| Automotive manufacturing | Lear | Expanded in September 2025 to Foundry, AIP, and Warp Speed, with more than 11,000 employees using the software |
Where one report from one warehouse answers the question, Foundry is more platform than the problem needs, and we say so.
A five-question Foundry fit test
BD Emerson's view is that five questions decide most Foundry fits. Score each from 0 to 2 and add the scores.
| Question | 0 points | 1 point | 2 points |
|---|---|---|---|
| How many source systems does the decision draw on? | One | Two to four | Five or more |
| How often do operators make the decision? | Monthly or less | Weekly | Daily or continuously |
| Does acting on it require a write to a system of record? | Never | Sometimes | Every time |
| How strict are access and audit requirements? | Internal reporting only | Regulated data such as PII or PHI | Regulated data plus external auditors or restricted hosting |
| Will AI agents act on the same data? | No plans | Pilot planned | In scope for production |
A score of 0 to 3 suggests a warehouse or lakehouse with BI tools. A score of 4 to 6 means Foundry can fit, and one use case should prove it before an enterprise commitment. A score of 7 to 10 describes the workload Foundry was built for.
Palantir Foundry vs Databricks and Snowflake
Foundry competes with Databricks and Snowflake for data platform budgets, and the three start from different jobs. Databricks and Snowflake store and process data for engineers, analysts, and data scientists, while Foundry adds the ontology and operational applications on top of its own pipelines.
| Dimension | Palantir Foundry | Databricks | Snowflake |
|---|---|---|---|
| Core job | Run operational decisions on an ontology of business objects | Lakehouse for data engineering, analytics, and machine learning | Cloud data platform for SQL analytics and data sharing |
| Main users | Operators and analysts, with data engineers and application builders underneath | Data engineers and data scientists | Analysts and data engineers |
| Business semantics | Ontology of objects, links, actions, and functions | Modeled by the customer on top of the catalog | Modeled by the customer on top of the catalog |
| Write-back to operations | Built in through action types | Built by the customer | Built by the customer |
| Pricing model | Negotiated per contract, with no public list price | Consumption-based, metered in Databricks Units | Consumption-based, metered in credits |
BD Emerson's rule of thumb: if the work ends at a dashboard or a model score, Databricks or Snowflake is usually the better-priced home, and if operators need to act on the answer inside a governed workflow, Foundry earns its cost. Our Palantir vs Databricks comparison goes deeper on architecture and cost.
How much does Palantir Foundry cost?
Palantir does not publish list prices for Foundry and negotiates each contract, so the number depends on platform scope, compute consumption, the operator population, the deployment environment, how much Palantir engineering is attached, and the contract term. The ranges BD Emerson sees in commercial Foundry and AIP deals run from roughly $250,000 a year for a narrow single-use-case deployment to several million for an enterprise program. Implementation and internal staffing commonly match or exceed the subscription in year one, which means a $1 million platform decision is usually a $2 million to $2.5 million program decision. The full breakdown is in our guide to what Palantir costs.
What does a Foundry implementation involve, and how long does it take?
A Foundry implementation is a sequence of use cases, and the first one sets the pattern for everything after it. A realistic first arc runs six to twelve weeks and delivers one operational use case, the data connections it needs, a deliberately scoped ontology, and an application real operators adopt, with security configuration, markings, and audit setup inside that arc.
| Phase | What happens | Typical duration |
|---|---|---|
| AIP bootcamp (optional) | Palantir engineers and your team build working use cases on your own data | 1 to 5 days, per Palantir |
| Use case selection | Pick one decision with a measurable baseline, a named owner, and known source systems; agree object definitions with operators | 1 to 3 weeks |
| First production use case | Connect sources, build pipelines and health checks, define object, link, and action types, build the application, configure markings and audit | 6 to 12 weeks |
| Second and third use cases | Reuse and extend the ontology, add actions, onboard new operator groups | 4 to 8 weeks each |
| Enterprise expansion | New business units, an ontology ownership model, and a handover to the internal team | 6 to 18 months |
Durations after the bootcamp row are BD Emerson planning ranges. Four factors push a phase toward the long end: the number and condition of source systems, unresolved identity across systems, write-back into systems of record that run their own change control, and security requirements such as markings for regulated data or a restricted hosting environment. Operator adoption moves the number too, because an application nobody uses has to be redesigned. Forward deployed engineers who build in your environment against your data compress discovery and transfer capability as they go.
What skills and roles does a Foundry program need?
A Foundry program needs six roles. Forward deployed engineers, Palantir's term for engineers who work directly with customers, build early solutions, and Palantir's S-1 credits them with helping customers identify new use cases. Data engineers build pipelines in Pipeline Builder and Code Repositories. Ontology engineers design object, link, and action types and maintain the definitions as the business changes. Application builders work in Workshop, Slate, and the Ontology SDK. A product owner with authority to settle definitional disputes keeps the model coherent, and a platform security lead owns organizations, markings, and audit routing.
Programs most often underfund the ontology engineers and the product owner, because their output is definitions and decisions that no dashboard shows. When Palantir launched its certification program in January 2022, it offered exams in four tracks, Data Engineer, Application Developer, Data Science, and Data Analyst, which gives hiring managers one external benchmark for the engineering roles.
Where BD Emerson fits
BD Emerson is an independent implementation consultancy for Palantir Foundry and AIP. We are not Palantir, we do not resell licenses, we take no margin on the software, and we claim no Palantir partner tier or badge. We deliver Foundry implementations with security and governance configured from day one, and we model one operational workflow end to end, including action types and write-back, before adding a second. If Foundry is on your evaluation list, start with the use case and let it size the license.
Frequently asked questions
What is Palantir Foundry used for? Companies and civil agencies use Palantir Foundry to run operational decisions that depend on many systems, such as supply chain planning, production scheduling, risk and compliance operations, and customer operations. Public examples include Airbus, Swiss Re, bp, NHS England's Federated Data Platform, and Lear.
Is Palantir Foundry a database or a data warehouse? Foundry stores and processes data, and its purpose is operations. It includes data integration, pipelines, storage, and analytics, then adds an ontology that models the business as objects, links, and actions, plus applications that write decisions back to source systems. Foundry can sit alongside a warehouse or lakehouse such as Snowflake or Databricks, with each doing the job it does best.
What is the difference between Palantir Foundry and AIP? Foundry is the data and ontology platform, and AIP is the AI layer that runs on it. Foundry connects source systems, builds pipelines, and models the business as an ontology. AIP connects large language models to that ontology through tools such as AIP Logic, AIP Chatbot Studio, and AIP Evals, so agents act through the same permissions and action types as people.
How long does a Palantir Foundry implementation take? A first production use case typically takes six to twelve weeks in BD Emerson's planning ranges, after an optional Palantir bootcamp of one to five days. That first arc covers data connections, pipelines, a scoped ontology, an operator application, and security configuration. Later use cases reuse the ontology and move faster.
How much does Palantir Foundry cost? Palantir negotiates Foundry pricing per contract and publishes no list price. BD Emerson sees commercial Foundry and AIP subscriptions from roughly $250,000 a year for a single use case to several million for an enterprise program. Implementation and internal staffing often match or exceed the subscription in the first year, so a $1 million platform decision is usually a $2 million to $2.5 million program decision.
Can Palantir Foundry run on-premises or in classified environments? Yes. Palantir's Apollo platform delivers Foundry to Palantir's cloud, customer clouds, on-premises data centers, and disconnected networks. For US government work, Palantir Federal Cloud Service holds FedRAMP High and Defense Department Impact Level 5 and 6 authorizations, and in February 2026 Palantir extended the IL5 and IL6 authorizations to on-premises and edge deployments.
If you are scoping a first Foundry use case, our Foundry implementation team can help you choose the decision, size the ontology, and set the security model before the build starts.
About the author
