Helix Cloud is open. Start free ↗
Helix Foundry
Docs Source Start free

Your company’s data, connected into one ontology

Connect your database, Stripe, WorkOS, PostHog and your files. Foundry suggests one ontology across them that you explore and ask questions of.

Start on Helix Cloud Self-host under Apache 2.0

$ git clone github.com/helixdb/helix-foundry.git
$ cd helix-foundry && ./scripts/setup.sh --compose

Built on HelixDB / DuckDB / Parquet / Ollama

Schema

A suggested ontology that matches the same records across sources.

The types, links and rules drawn from your data, stored as real objects and relationships in HelixDB. A generation stays a draft until you make it current.

The Schema view of an ontology called Supabase Retail Operations, showing seven types including DocOrder, DocCustomer and DocProduct as cards, joined by labelled links such as doc includes product and doc placed by.

Explorer

Everything a load wrote: its entities and facts.

Open a record to see its properties and every relationship, in and out, with the load and the version that wrote it.

The Explorer, showing 59 of 84 nodes and 53 of 74 edges as a graph, with a node panel open on order ORD-0000002 listing its properties and its three outgoing and three incoming relationships.

Fig. 1

01Ontology

The same customer, matched across every source.

  1. 1.1Real objects and relationships in HelixDB
  2. 1.2Immutable Parquet snapshots, with history
  3. 1.3Proposed changes are tested, then wait for you
Build your ontology

Fig. 2

02Isolation

Generated SQL runs where it can reach nothing.

  1. 2.1No internet, no connector credentials
  2. 2.2No Linux capabilities, 512 MB a job
  3. 2.3Credentials encrypted with a key you hold
Read the security model

The point

Nobody has to agree on a schema first.

Connect a source. Foundry suggests the ontology, you keep or change it, and nothing is published until you say so.

Start on Helix Cloud

Fig. 3

Answers backed by executed SQL.

Ask anything about your data. The answer comes back with the query that produced it and a citation to the exact snapshot it ran against, and those snapshots are immutable.

Ask your own data

Which accounts have an open support ticket and renew in the next 30 days?

Nine accounts, holding $14,200 of monthly recurring revenue.

SELECT a.name, count(t.ticket_id), s.current_period_end
FROM accounts a
JOIN subscriptions s ON s.account_id = a.account_id
LEFT JOIN support_tickets t ON t.account_ref = a.name

accounts v14, subscriptions v9, support_tickets v3

Fig. 4

1,000,000 rows to Parquet472 ms
1 GB of events imported10 s
100,000 records published64 s
Everything within two links16 ms
25 requests at once, median1,690 ms

Measured on one laptop.

Every figure below was measured on one M1 Pro laptop with 16 GB of RAM. Run BENCH_HTTP=1 pnpm benchmark to produce them yourself.

Start on Helix Cloud

Connect what your company already runs on.

Automatic sync through native PostgreSQL and MySQL change capture where the database allows it, and scheduled snapshot refreshes otherwise.

PostgreSQL / MySQL / Neo4j / Neon / Supabase / PlanetScale / Stripe / WorkOS / PostHog / REST / S3 / CSV and Parquet / Ingestion API

Now there is nothing to host. Helix Cloud runs the same build we publish, so you connect a source and explore the ontology without standing up HelixDB, DuckDB and Ollama yourself.

Start on Helix Cloud Read the self-hosted setup

Self-hosted Foundry listens on 127.0.0.1 only and has no sign-in: anyone on that machine is already in.