Helix Foundry

Not a Palantir product. Helix Foundry is published by HelixDB. Its repository, NOTICE file and documentation never mention Palantir or Palantir Foundry. The “open-source Palantir Foundry” framing comes from promotional posts about the release, not from the project. Its name is similar, and the two products solve a related problem in different ways. See palantir-foundry-and-aip for the commercial product.

Helix Foundry is a local-first data workspace: it connects a company’s operational sources, proposes one connected ontology across them, and lets people explore that ontology and ask questions of it, all on their own computer. The README’s summary is “Your company’s data, connected into one ontology, on your own computer.” It is built on HelixDB (graph and object store for the ontology and metadata) and DuckDB (SQL over Parquet snapshots). No accounts and no hosting; AI runs locally by default.

Facts (repository, read 2026-10-10)

Repository	HelixDB/helix-foundry on GitHub, TypeScript
Licence	Apache-2.0 (LICENSE file); NOTICE: “Copyright 2026 HelixDB”
Created	2026-10-07; last push 2026-10-10
Activity (point-in-time)	About 700 stars, 100 forks and 11 open issues on the day read; no releases tagged
Status	The validation document calls it “an executable single-server MVP” and says its results are “not a production SLA”
Platform	Docker Compose. Developed and validated on macOS with Apple Silicon; CI on Linux; Windows untested
Pinned HelixDB	ghcr.io/helixdb/helixdb:v0.0.6 (from the README dev command)

What it does

• Connectors: Neon, Supabase, PlanetScale, PostgreSQL, MySQL, Stripe, WorkOS, PostHog, REST APIs, S3-compatible storage, file uploads (CSV, JSON, JSONL, Parquet) and an ingestion API.
• Change capture: native PostgreSQL and MySQL change capture where the database allows it; scheduled snapshot refreshes otherwise.
• Suggested ontology: the AI proposes object types and relationships, for example matching the same customer across Stripe and the application database. Proposals show assumptions and unmatched relationships for review, and only validated definitions are published.
• Versioned data: immutable Parquet snapshots with profiles, schema-drift review and history. Queries run in isolated DuckDB child processes.
• Home and Analyst: key metrics, and answers backed by executed SQL with snapshot citations.
• AI providers: local qwen3:4b through Ollama by default; Claude or OpenAI are explicit choices, with no automatic fallback between providers. Hosted sample data is masked for detected sensitive columns unless allowRawData is set through the API.
• Coding-agent setup: the README asks a coding agent to install and verify the app, following an AGENTS.md runbook.

Architecture

App (React UI and loopback-only API), Worker (durable AI runs, scheduled imports, pipelines), Executor (disposable DuckDB processes with no internet access and no connector credentials), HelixDB (workspace metadata, API tokens, jobs, proposals, ontology records and edges) and Ollama. Only the app publishes a port, on 127.0.0.1. Connector and AI credentials are encrypted with a key in .env.

Security posture (as documented)

Single-user by design: no accounts or sign-in, so anyone who can reach the app controls every workspace. Exposing it through a reverse proxy or tunnel is explicitly discouraged. Scripts and the SDK use read-only workspace tokens unless write access is granted.

Validation (vendor-run, September 2026)

The project’s own VALIDATION document reports tests on a development machine, including a Chromium end-to-end run and a reference workspace in which the local model produced an ontology, SQL pipeline and relationships over sample data in about 153 seconds of model time with six calls. These are the project’s results, not independent ones. No third-party review was found.

How it compares (opinion)

• Commercial ontology platforms such as palantir-foundry-and-aip are hosted, multi-user and governed at enterprise scale. Helix Foundry is single-user, local and early.
• Catalogs and semantic layers (data-catalogs-compared, semantic-layers-compared) describe and govern data; Helix Foundry also proposes and materialises an object model over it.
• Graph and ontology stores are covered in knowledge-graph-and-ontology-platforms; this project is an application built on one of them.

Useful if you want to test the ontology-first idea on your own data without a server. Not yet a fit for shared, production or regulated use.

Open items

• Relationship to Palantir: none found in the project. The framing in promotional material is not supported by the repository.
• No tagged release or versioned stability promise was found; the project is three days old at the time of reading.
• Star and fork counts change daily and are recorded only as a snapshot.
• The model-quality results come from one development machine and one local model.

Sources (read 2026-10-10)

• https://github.com/HelixDB/helix-foundry (README, LICENSE, NOTICE, GitHub API metadata)
• docs/GUIDE.md, docs/VALIDATION.md and AGENTS.md in the same repository