An agent that drafts your ontology, and a graph to hold it

What it does

Foundry connects a pile of business data sources — Postgres, MySQL, S3, Stripe, PostHog, GitHub, Linear, and friends — plus uploaded documents and table files, and turns them into a single graph in HelixDB. An agent drafts the ontology: types, links, source mappings, and rules. You review, edit, and make a version current, then a load writes it into the database. From there you can explore the graph, trace any value back to its originating row or quote, and ask questions with cited answers.

The interesting bit

The ontology drafting is delegated to an agent but kept under human review — the machine proposes, you approve. And the AI backend is refreshingly pragmatic: it uses Claude or OpenAI if a key is in your shell, falls back to Ollama, then LM Studio, and tells you in Settings which it picked and why. Local-first without being dogmatic about it.

Key highlights

• Connects PostgreSQL, MySQL, S3, Stripe, PostHog, WorkOS, GitHub, Linear, Granola, and Cal.com, plus Markdown, PDF, CSV, TSV, JSON, and Parquet uploads
• Every value in the graph traces back to its source row or quote
• Answers to questions come with citations
• Runs fully self-hosted via Docker Compose, or hosted at foundry.helix-db.com
• Apache 2.0 across the site, backend, and all three images; HelixDB itself is Apache 2.0 too

Caveats

• The self-hosted install is single-owner with no sign-in and every connector on — fine for one person, not for a shared deployment
• First-run secrets live in a Docker volume; docker compose down -v takes your data with it

Verdict

Worth a look if your company’s data is scattered across SaaS tools and you want one queryable graph over it without hand-writing an ontology. If you need multi-user access control or already have a knowledge-graph pipeline, the hosted version or your existing tooling may fit better.