I will be out of the USA sometime in September-November. During that time I won’t have access to Palantir Foundry. So I am trying to work as much as possible in Foundry now, and polish the website/docx’s during my trip.


This Enterprise section currently focuses on Palantir Foundry as a concrete, hands-on example of a modern enterprise platform for data integration, ontology, analytics, applications, AI assistance, and governance.

My focus (for now) is on Palantir (Foundry) because

  • Palantir is a market leader.
  • Palantir offers a generous free trial account for Foundry.
  • Palantir enterprise systems utilize AI as a helpful assistant (not as a replacement for human intelligence).
  • Palantir’s built-in AI help (FDE) makes it possible to complete the demos (without human assistance).

Palantir Foundry represents the logical culmination of ZiptieAI:

  • (2) neural networks,
  • (2b) models,
  • (3) agents,
  • (3b) workflows, and
  • (3c) enterprise platforms

brought together in a single system. One reason I find Palantir’s approach appealing is that AI is used as a powerful engineering tool within a governed enterprise platform, rather than as a replacement for human judgment.


TOC:

  • 1 Concepts
  • 2 Getting started hands-on
  • 3 Advanced hands-on (future)


1 Concepts

  • 1.1 The gist
  • 1.2 Basic workflow


1.1 The gist

  • Palantir Foundry makes the enterprise visible and governable.
  • Data from many systems is transformed into a common operational model that can be analyzed, queried, monitored, and updated.
  • LLMs play the role of helpful assistants, but their very presence makes parts of the system probabilistic. As a result, AI actions must be sandboxed, verified, approved, and logged.

Note: I often use GPT to “polish” text. And GPT makes a lot of very good suggestions. But GPT is also programmed to strongly suggest the “politically correct” version of reality when it comes to AI. The title for the screenshot below is a perfect example.

  • My original title:
    • The fantasy world – (left) a crystal ball (called a “palantir” (“seeing stone”) in The Lord of the Rings) and (right) AGI (a myth that digital circuits can host intelligence). Palantir Foundry represents the real world – An enterprise system that provides the infrastructure and safeguards so that AI can be a practical “helpful assistant”.
  • GPT’s strongly suggested “professional” title:
    • The trusted wizard (left) using a crystal ball (“palantir” or “seeing stone” in The Lord of the Rings) and (right) modern AI running on digital hardware. Both promise enhanced visibility and insight, although one belongs to fantasy and the other (AGI) belongs to engineering.


drones drones


1.2 Basic workflow

The following diagram shows the basic PAL workflow.

  • 1a Data source. Raw data from enterprise systems. For example: CSV files, Databases, ERP systems, CRM systems, Sensors, External APIs.
  • 1b Pipeline. Transforms raw data into clean, structured, and consistent datasets. For example: cleaning, joining, filtering, aggregating, validation.
  • 3a Ontology read. Applications and users read enterprise data through the ontology. The ontology provides a common operational model of the enterprise.
  • 3b Ontology writeback. Applications, workflows, and users update enterprise systems through the ontology. For example: change status, create ticket, approve request, update record.
  • 4 Analysis. Code, queries, dashboards, rules, AI, and other logic analyze the ontology data. Purpose:
    • discover information
    • detect problems
    • support decisions
  • 5 UI / App. User-facing applications. For example: dashboards, workflows, forms, operational applications, AI assistants.
  • 6 Security / Governance. Controls who can see data, modify data, run workflows, and use AI. For example: permissions, auditing, approvals, version control, lineage, logging, AI.
  • About AI. AI can appear in many layers: pipeline transforms, model integration, ontology functions, analysis, UI assistants, and governance monitoring. AI should be treated as a helpful assistant, not the center of the system and not the final authority. The responsible pattern is: AI suggests, deterministic logic verifies where possible, humans approve important actions, and the platform logs and governs the workflow.

The ontology is the operational heart of Foundry. Most applications, analysis, workflows, and AI interact with enterprise data through the ontology rather than directly through the raw datasets.
drones

The three crystal balls represent three layers of governance. Foundry controls

  • Crystal ball #1 — Data governance. Controls how raw enterprise data becomes trusted operational data.
    • Examples: ingestion, transformation, validation, lineage.
    • This is the boundary between: Data Source, Pipeline, Ontology.
  • Crystal ball #2 — Operational governance. Controls how applications, users, workflows, and AI interact with the enterprise.
    • Examples: ontology permissions, workflow permissions, write-back controls, approval processes.
    • This is the boundary around: Ontology, Analysis, UI, Writeback.
  • Crystal ball #3 — Global governance. Controls the entire platform.
    • Examples: security, auditing, version control, branching, deployment, compliance.
    • This is the layer that sits above everything else.


2 Getting started hands-on

First

  • 2.1 Palantir free tier sign up
  • 2.2 Palantir Foundry demo sources
  • 2.3 My demo plan

Second the actualy hands-on installable_examples and step-by-step demos

  • 2.5 PAL (installable) examples (START CREATING PALANTIR FOUNDRY APPS HERE). These are complicated, but they have an autoinstaller that allows you to start with a (usually) working version that you can modify.
  • 2.6 PAL demos. These are less complicated, but you start from 0, and getting a working demo at the end requires a lot of skill (including debugging).
  • 2.7 DIY demos. Go free style (not “vibe”) by teaming up with FDE (AI assistant) to create your own demos.


2.1 Palantir free tier sign up

If you cant get a trial version of PAL running, then there’s no reason to read further. You will not understand PAL by just reading about it.

  • I did not record how I signed up, and its not a process I can repeat. But I can tell you the following:
    • You must be in a country that PAL supports (thats only a few countries).
    • You can NOT use a VPN.
    • You must have a passport or some picture ID of good enough quality for facial recognition verification. My passport is almost 10 years old, but it still worked.
    • You need a camera on your PC to take a pic of yourself. PAL will compare your ID and the pic.
    • I think you need a mobile phone number (and email).
  • It was a fairly long process, but very professionally done (I was impressed).

    drones


2.2 Palantir Foundry demo sources

When I first started with Foundry I was a bit overwhelmed (this is a very powerful system). I first checked out

  • palantir.com/docs/foundry. Has a lot of great content. But its probably not where you want to initially start. Note: The most interesting thing on this page for me was the diagram below (I added numbering that matches the workflow diagram in 1.2 above).
    drones
  • Several Youtube sites. The sites I viewed were quite good, but not really suited for an absolute beginner. Documents are the best for absolute beginners (and they tend to be more up-to-date and can serve as the template for your own docs).

I wanted to get hands-on experience from the start. So I checked out several official Palantir sources of demos. There were a lot of demos. There seem to be 3 main sources (see also the chat with FDE about demo sources):

  • build.palantir.com has links to examples/demos organized by type: AI Starter Pack, Document Suite, Reference Implementation, etc.
    drones
  • learn.palantir.com has links to examples/demo organized by role: Data Engineer, App Dev, etc.
    drones
  • [my-org].palantirfoundry.com/workspace/marketplace is a Foundry internal app for finding examples/demos organized by 9 “stores”.
    drones


2.3 My demo plan

I spent the first month doing the demos I thought looked interesting from the sources listed in 2.2 above (lab notes (MS.Word docx files) are available for each example/demo on the GDrive). Those demos are organized (by Palantir) by role and project types. But what I really needed was a (straight, start to finish) learning path for a generalist beginner. So I decided I’d just start doing demos and after a while I’d eventually figure out a learning path.

I still don’t have a learning path. But I do have a rough draft learning plan #699_planning_.docx that documents

  • All the sources
  • What I did so far
  • What I plan to do
    drones

The goal of #699:

  • Clear master list of demos (examples, courses, etc).
  • For each demo: A simple Foundry workflow main diagram with for that demo.


2.5 PAL examples (START CREATING PALANTIR FOUNDRY APPS HERE)

PAL “examples” are complete (usually more complicated) and have an installer that allows you to start with a (usually) working version that you can modify. This is the best way to start.

  • Install a PAL example. The examples are installed and (usually) ready to run immediately.
  • Run the example first.
  • Then rebuild the examples (from top to bottom, from downstream to upstream). The examples are excellent, but are often quite complicated. But you don’t have to do everything first step-by-step before you see the the final results. Once you have a working first (reference) example, you can install a second (experimental) example that you modify at your pace. This allows you to (1) appreciate quickly the capabilities of PAL and (2) focus on learning the UI and basic concepts/terminology.
    PAL example (left, center) and my own build (right) (3c.1b)
    drones drones drones
  • Workflow diagrams for examples are usually a bit more complicted than for demos.
    drones


2.6 PAL demos (speedruns)

These are (typically) less complicated than installable PAL examples, but you start from 0, and getting a working demo at the end requires a lot of skill (including debugging). My future versions of these will take a different tactic: Not trying to do everything at once (get a basic result first, then go back and add the missing parts). But for now I’d recommend doing these after have done some installable examples.

  • Current demos consist of just D1-D8, the first PAL demos I did (step by step tutorials).
  • Demos are simpler, but I think more difficult to get running. I sometimes spent a few hours and at the end it did not work, and I did not know how to find the problem. But if you did the examples first, then you have enough of the environment experience to get through the demos..

For the D1 demo this is the main workflow diagram (for details about this workflow see D1 workflow chats).
drones

A diagram of the configuration steps for demo D2 (I spent days on my initial attempts to recreate D2 without this diagram; with this diagram its fairly simple; its even simpler if you do it in 3 separate phases as described in the latest version of the docx)
drones


2.7 DIY (with FDE assistance) demos

Go free style (not “vibe”) by teaming up with FDE (AI assistant) to create your own demos. I recommend this last because these will be my original demos so I need more time to think of what kind of examples to do.

  • The real goal of the Getting Started phase is to be able to quickly assess how you can create apps to solve requirements relying only on AI for assistance. FDE is an amazing tool that can (1) “see” screenshots that you paste into a chat and (2) guide your through the most complex configurations and problem solving. But you need to have an idea of the big picture direction for any project. The (eventual) goal of the DIY demos is to show how to team up with FDE to work effectively in Foundry (and to a limited extent AIP/PAL Pilot).
    drones

  • When you start free styling, things get complicated easily. You do try things that the experts knew to avoid when they created installable-examples and step-by-step demos. It took me a long time to get the Workshop in that DIY demo working. Even FDE made a mistake during the configuration. Afterwards I created this simple diagram to make sure I could miss those “rabbit holes” in the future (with uniquely numbered steps and variables; they don’t have to be perfect, just something you can trace).
    drones


3 Advanced hands-on (FUTURE)

This will focus on partnering with FDE to cover 2 major topics areas

  • 3.1 Demos for Foundry certification exam guide for application developer topics (the exam involves answering questions; this section uses the exam to determine what demos to do).
  • 3.2 Other demos (Managing Foundry environment, 24/7 operation, streams, deployments, packaging, sharing, etc).


3.1 Demos for Foundry certification exam guide for application developer topics

Use the exam as a guide for what demos to do to cover all the core topics that were not covered earlier.
drones
drones

3.2 “Other” demos (not sure exactly what yet)

Demos not covered earlier (and not in the exam; I don’t have a clear idea yet what the topics in this section would be). Up to this point you only use the apps for testing. They are not setup for end user usage, for real deployment (“deployment in the UI in the previous apps was more like “compile/run”). Demos for topics not covered yet. Maybe

  • managing Foundry environment,
  • 24/7 operation,
  • streams,
  • packaging, sharing, deployments, dev social media, etc

Basically whatever is inspired by Online PAL docs.
drones


26.0814 (v1 26.0702)





















OLD

  • A7 (3c.0) Concepts. My own take (see also Concepts CHATS ONLY).
    • What problems PAL solves.

    • This section is a long term WIP (I started using PAL in July 2026, and I need more experience before I can sum up the concepts).
      drones

    • xxxxA6 (3c.1) PAL AI usage strategy.
      • PAL Foundry has 2 helps systems: (1) AIP and (2) FDE.
      • I used both during the first 8 “Initial PAL demos” (3c.2).
      • During demo 8 (of 3c.2) I asked AIP/FDE to show me how to do something that was not in the demo. The advantages of each tool slowly became apparent.
      • During the AIP/FDE demos (3c.3) I uses only AIP/FDE. It became apparent that
        • AIP is better for getting initial workflows.
        • FDE is the lifesaver that can solve problems and (most of all) recognize screen shots (just paste in the window). In one example, FDE led me step by step in setting up a Workshop demo that I just could not get right (for good reasons).
      • I look forward to when AIP/FDE are combined into a single tool that can “see” dialogs and guide a user through various tasks.
        • NOTE: After almost a month of using PAL, I noticed PAL has a beta product called “Pilot”. That appears to be what I was talking about.
      • In general, a big focus (at least initially) is on prompt techniques and workflows to maximize effectiveness of using AI.
        FDE
        drones

    • A8 (3c.4) Certification (Foundry Application Developer).
      • We’ve been (hopefully) using the PDF as a guide for what to study/build in goals (1-4).
      • Now need some certification (for me its Foundry Application Developer Certification). For a system like PAL, most job will be in big projects. Certification matters.
        drones

Notes