News & Intel
LiveDaily AI-curated intelligence on Palantir Foundry, Ontology, AIP, Apollo, contracts, and community feedback. Updated automatically via GitHub Actions every day at 7 AM UTC.
What are promoted variables and how to create them on workshop?
I had an embedded module which has an object set filter, this object set filter is populated using an object set which is function backed, so the issue was whenever the user would close the embedded module and reopen it, the filter gets recomputed and loses its selection But this was solved using promoted variables as shown in the image, but I don’t know how it was solved as it was not done by me, could I understand the concept of promoted variables and how to make one on workshop 1 post - 1 participant Read full topic
How to conditionally reset / retain Object Set Filter in embedded module when parent selection changes?
Hi all, I’m trying to solve a state retention issue with an embedded module inside an overlay in Workshop. Setup I have: a parent screen where the user makes a selection a button that opens an overlay inside the overlay: a close button an embedded module The parent screen passes the user’s selection into the embedded module via interface variables . Inside the embedded module, the screen contains: an Object Set Filter an Object Table The Object Set Filter is backed by a function-backed Object Set . Current behavior The filter state is being retained (apparently via promoted variables , although I did not originally implement that part). Example: User selects A in parent screen Opens overlay / embedded module Embedded module loads filter options, and filter value 1 is selected User closes overlay User later selects B in parent screen Opens overlay ag
Academic Use of Foundry for University Paper
Hi I have been directed to post this here by Palantir Support, so here goes: As part of my postgraduate studies in Machine Learning, I am planning to write a paper that builds on my previous work with financial time series data. The focus of the work is on data modeling, pipeline design, and system integration. I am considering using Foundry to implement data pipelines (based on publicly available market data such as OHLCV time series), and integrating the results into a small, self-developed application (already existing). The application itself would remain external and consume processed data and model outputs via a defined interface. Before proceeding, I would like to clarify a few points regarding acceptable academic use: * Is it permitted to use Foundry for a university paper and demonstrator setup? * Are screenshots or architecture diagrams derived from Foundry allowed in such a paper? * Are there any restrictions on describing platform concepts (e.g. ontology, pipeli
Iceberg Virtual tables limitation
We would like to use apache iceberg inside aws using glue data catalog but we have this question , can you confirm or not this is true : If virtual tables are out because they only support append only transactions (for now) and BYOB means AWS storage + Foundry managed catalog will the iceberg data really be useful to other AWS processes like Redshift with Foundry managing the catalog? The decision then just becomes one of cost efficiency for storage via BYOB vs Foundry. I don’t have a definitive answer for this but that’s my understanding of the Foundry docs 2 posts - 2 participants Read full topic
Foundry SDK SQL Queries Errors
Hello! On my stack, I been using the Foundry sdk to run SQL queries against thousands of data sets to sample. I then package it up and run an LLM against it to get a description of those datasets. This is very much a wide governance and “what’s in our foundry” question. I know foundry sdk is in preview for this purpose. It’s actually generated several unique errors that I believe are because of the platform. If it would be helpful for the product to show the query failure errors, I’d be happy to share them to product team. 1 post - 1 participant Read full topic
Palantir Foundry and EPIC Integration - Biller Notes retrieval / submission
Hey Dev Community! I’ve built an application to enable case workers to embed AI into Professional Billing Denials management. The application is complete and working as expected with the Sandbox data provided by EPIC showroom. We are completing the EPIC Showroom Vendor Services application but cannot identify the EPIC endpoint need to bring in the Biller’s Note . Another question is, how to submit those notes back? You can image, that within a platform we have built an application that allows us users to send back the notes that they have written within the application back to EPIC. We have already tried to ingest Document References from the sandbox patients, but there is no evidence in the category field that there are some considered to be Biller notes. Can anyone point me in the right direction? Is it likely to be found via the DocumentReference.Search, Account.Read or ExplanationOfBenefit.Search endpoints? Thanks! References: https://fhir.epic.com/Sandbox?api=1047
Type Mismatch Error for Start_TS in Foundry Builds
Hello Palantir Community, I am encountering a build failure in Foundry related to the Start_TS property. Below is a summary of the issue and the troubleshooting steps I have taken. Summary of Error: The build failed due to a type mismatch in the property Start_TS. The system expected an offsetdatetimeudt type, but the field is being treated as a struct type during transformation/output generation. Because of this mismatch, the ontology build fails. What I Observed: When I import the dataset, Start_TS has values like 2026-03-26T02:45:52.604Z and Foundry identifies the type as datetime. However, after I transform the dataset or generate the output dataset, Foundry changes the datatype of this field to struct. Later, when I try to build the ontology type and ontology objects, Foundry expects this field to be offsetdatetimeudt, which causes the build to fail. Why the Failure May Be Happening: It looks like the source dataset initially has the column in a datetime-compatible form
Markdown support in Media Sets
Given the ubiquity of Markdown these days, I would like to request first class .md support in Media Sets including as an alternative to PDFs as the primary format. Markdown can, of course, be saved an a property on an object or a column on a dataset, but there are plenty of instances these days when neither of those is ideal. Maybe you don’t yet know the object type to make. Maybe you want to give users a better reading experience than viewing the content of a dataset. Maybe you want to let users upload markdown as a manual input to a pipeline… Thanks for considering! 1 post - 1 participant Read full topic
Unofficial starter for deterministic local contract testing in Foundry oriented workflows
I put together a small open source starter for a problem I kept wanting a cleaner answer to: how to test the TypeScript contract layer around a Foundry oriented workflow before deployment or later platform evaluation. Repo: https://github.com/vitron-ai/aip-foundry-themis-starter This is an unofficial example. It is not affiliated with, endorsed by, or maintained by Palantir. The repo is intentionally narrow. It uses one intake-triage workflow to show how to: shape a request payload validate raw agent or function output with runtime schemas normalize that output into a stable app-facing contract preserve telemetry like request ID, latency, retry count, and fallback usage export a local handoff artifact for later evaluation work It also includes deterministic local tests for failure modes like malformed JSON, missing fields, unsupported status values, and empty output, plus a small React demo and an OSDK-like seam showing where a real client co
Dynamically Generating Transform
Hi, I hope you are well. I am trying to dynamically make multiple output datasets based on a list of files paths I have. I managed to do this when I have the list of file paths written out manually. However i want to do this completely dynamic, by fetching the list of files. However can I do this? Because I’m not allowed to use the transform decorator. Is there anyway I can fetch information from a dataset, file, media set without using that decorator? Also I understand that in a production workflow this isn’t common as we normally define our datasets and schemas, however I was doing this for a specific use case and just wanted to know if it’s possible. Thanks Sam 5 posts - 2 participants Read full topic
Feature Request: Solution Designer — Sample Data Rows
Today in Solution Designer we can add custom columns (metadata) to a dataset, but we cannot populate example rows . It would be extremely helpful if Solution Designer allowed users to enter a few sample rows (like Excel) under those columns. This would make it much easier to: Communicate dataset structure Show example values Align stakeholders before building transforms Request: Allow Solution Designer datasets to include editable sample rows for columns (Excel-style preview data) (only need like 10 rows). 2 posts - 2 participants Read full topic
Issues extracting rows from geojson files
Hi, I’ve imported several GeoJSON files (all from data.gov) into Foundry and have been trying to run transforms on them. Following the tutorials, I imported them as “Unstructured” and then used a series of transforms — including the “Prepare Geometry” transform — to properly format the geometry column. It seemed very straightforward, and I was able to get it working perfectly… but only with the exact same data file used in the tutorial. I have several other GeoJSON files I’d really like to use. The issue is that when I run the first transform ( “Extract Rows from JSON” ) and paste my GeoJSON into the text field to auto-generate the schema, it does a decent job overall but does not correctly parse the geometry field (and everything downstream from it). The feature to automatically generate the schema from a GeoJSON file is really nice in theory, but it doesn’t seem to work reliably most of the time. Am I doing something wrong? I’ve occasionally gotten it to work by manua
Issue Configuring Action In Workshop When Other Ontology Objects Exist
Hi All! I have built out an application in workshop and I am now trying to assign an action to a button group widget. I have the button set as On Click Action but when I go to the drop down to select an action there are no results. In a completly empty workshop app if I do the same thing with no other objects configured so just the button group all actions are visible and selected from the drop down. I have the Ontology that the action is created against as a object type in workshop but it seems as though as soon as I add any other Ontology objects to the object types field I am unable to use actions at all. If I configure the action against the button before adding anything else to workshop and continue building out the app the action appears to work completly fine. I strongly believe I am just missing something really simple as currently it would mean I have to start the workshop application from scratch every time I wanted to add a new action to it which just seems really
Technate, Ohio: How Leslie Wexner and Jeffrey Epstein Built The Silicon Heartland
Ohio has become the new destination for Big Tech data centers and AI infrastructure. The state owes much of its rapid transformation into the so-called “Silicon Heartland” to the now vast array of public-private partnerships pioneered by the state’s richest m…
Was the Iran War Caused by AI Psychosis?
AI sycophancy, RLHF bias, and Ender's Foundry simulations shaped Operation Epic Fury. 7 planning assumptions failed in 23 days as the Iran war defied every AI prediction.
Filter on Array Column via Typescript OSDK
Looks like in Typescript V2 repos I can’t filter on Array Columns? Or am I just not using the OSDK correctly. At least I couldn’t find the right API syntax in the docs. https://palantir.mcloud.merckgroup.com/docs/foundry/ontology-sdk/typescript-osdk/ E.g I was trying: orders.where({ manufacturNames: { $eq: “name“ } }) manufacturNames is an Array String Field. I think in the Typescript V1 Repo’s this kind of filters are possible as well as in Workshop. 2 posts - 1 participant Read full topic
Best ways to search for a dataset
I currently have an rid for which I need to track a dataset. But when I search using the default search option it returns missmatched results. How/What is the best approach here? 1 post - 1 participant Read full topic
Are there any benefits of decreasing lightweight transform RAM (or resources in general) from defaults?
As per title. Couldn’t find anything in the documentation pertaining to Code Repositories. My understanding is decreasing CPU might decrease Resource queue utilization, maybe used compute seconds, maybe faster resource allocation. Decreasing RAM is unclear. Defaults mentioned in documentation (old lightweight decorator specifies default memory, new syntax via transform.using().with_resources() - doesn’t): transforms.api.lightweight (cpu_cores=2, memory_mb=None, memory_gb=16, gpu_type=None, container_image=None, container_tag=None, container_shell_command=None) So, is to worth assigning lower resources to lightweight transforms, or is it intended to be used just to increase the required resources? 1 post - 1 participant Read full topic